(John G. Webster (Editor) ) 50.quantum Electronics

50 Quantum Electronics
50 Quantum Electronics
Distributed Bragg Reflector Lasers Abstract | Full Text: PDF (228K) Distributed Feedback Lasers Abstract | Full Text: PDF (210K) Dye Lasers Abstract | Full Text: PDF (250K) Electronic Speckle Pattern Interferometry Abstract | Full Text: PDF (310K) Excimer Lasers Abstract | Full Text: PDF (254K) Free Electron Lasers Abstract | Full Text: PDF (210K
file:///N|/000000/0WILEY%20ENCYCLOPEDIA%20OF%20ELE...TRONICS%20ENGINEERING/50.Quantum%20Electronics.htm15.06.2008 20:02:05
file:///N|/000000/0WILEY%20ENCYCLOPEDIA%20OF%20ELECTRICAL...ONICS%20ENGINEERING/50.%20Quantum%20Electronics/W6301.htm
}{{}}
HOME
q
ABOUT US
q
CONTACT US
q
Home / Engineering / Electrical and Electronics Engineering
HELP
q
Wiley Encyclopedia of Electrical and Electronics Engineering

Distributed Bragg Reflector Lasers Standard Article S. D. Roh1, R. B. Swint1, J. J. Coleman1 1University of Illinois Copyright 1999 by John Wiley & Sons, Inc. All rights reserved. DOI: 10.1002/047134608X.W6301 Article Online Posting Date: December 27, 1999 Abstract | Full Text: HTML PDF (228K)
q q
Recommend to Your Librarian Save title to My Profile Email this page Print this page
Browse this title

q
Search this title Enter words or phrases

r
r r
Advanced Product Search Search All Content Acronym Finder
Abstract
The sections in this article are Dbr and Laser Basics The Bragg Period Edge-Emitting Single Frequency Lasers Vcsel Conclusion Keywords: bragg period; coupled mode theory; single frequency semiconductor lasers; reflectivity; DBR fabrications; VCSEL; tunable DBRs About Wiley InterScience | About Wiley | Privacy | Terms & Conditions Copyright 1999-2008John Wiley & Sons, Inc. All Rights Reserved.
file:///N|/000000/0WILEY%20ENCYCLOPEDIA%20OF%20ELE...0ENGINEERING/50.%20Quantum%20Electronics/W6301.htm15.06.2008 20:03:48
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering Copyright c 1999 John Wiley & Sons, Inc.
DISTRIBUTED BRAGG REFLECTOR LASERS

One of the distinguishing characteristics of a laser, as opposed to other sources of light, is that its emission consists very nearly of a single frequency, or color, of light. However, close inspection of the light emitted from a simple semiconductor laser source reveals that its light output consists of several closely spaced frequencies, and thus its spectrum is not innitely narrow. The characteristic of single-frequency emission is critical for some applications, including optical communications and spectroscopy. One method of narrowing the emission spectrum of a simple semiconductor laser source is to incorporate in the laser a structure that is capable of selectively reecting only one frequency of light. A distributed Bragg reector (DBR) is just such a structure. Semiconductor lasers that have DBRs incorporated in them have single-frequency output. Emission spectra from two semiconductor lasers, one with and one without a DBR, are shown in Fig. 1. While other singlefrequency laser sources exist, semiconductor DBR lasers are preferred for many applications because they are relatively simple, compact, and robust and operate over a large temperature and current range, whereas alternative sources may be complex and bulky and require precise alignment. The use of a DBR to make a single-frequency laser source is only one of the several ways that a DBR can be used to improve or otherwise make possible certain operating characteristics of a semiconductor laser. This article describes the basic operating principles of a DBR and specically how it is implemented in two types of semiconductor lasers: edge-emitting lasers and vertical cavity surface-emitting lasers. Design and fabrication issues for both types of laser are presented, as well as some of the advantages afforded by each design.
Dbr and Laser Basics The DBR Concept. When light crosses a boundary between two materials that have different indices of refraction, ni , the light experiences a partial reection given by
A DBR is a structure that has a change in its refractive index that is repeated several times in a set period, termed the Bragg period, . The periodic change in the refractive index causes multiple partial reections that add constructively to create a strong reection. Figure 2 is a simplied picture of how these multiple partial reections can add to form a strong reection. As will be shown later, this type of additive reection is maximized when the incident light has a wave-length that is equal to twice the period of the Bragg reector, . A DBR is useful in two respects: it can create a strong reection, and the reection created is wavelength specic, that is, it reects some frequencies of light while allowing other frequencies of light to pass through unreected. The Semiconductor Laser. Before developing further how a DBR can be used to improve a lasers performance, it is important to have clear understanding of the fundamentals of laser operation. Three factors are required for laser operation: an amplifying or gain medium, a resonant cavity for feedback, and some means
1
Fig. 1. Optical spectra of InGaAsPInP ridge-waveguide lasers. Several longitudinal modes exist in the simple Fabry Perot laser, whereas the DBR laser exhibits single-frequency operation.
Fig. 2. Simplied pictorial of how multiple partial reections can add to create a strong effective reection in a DBR with a period .
of excitation. In a semiconductor diode laser, recombination of electrons and holes in the diode junction results in light emission, and a population inversion of these carriers provides optical gain. The resonant cavity for a semiconductor laser is formed by an optical waveguide with partially transparent mirrors on either end. An electric current owing through the diode is the source of excitation. The onset of lasing action occurs when the excited gain medium begins to create just enough light to offset the loss of light due to internal losses within the cavity and the loss of light through the semitransparent mirrors. This condition is given by the equation
Fig. 3. A schematic diagram of (a) wavelengths at which FabryPerot modes exist, (b) the gain distribution in a semiconductor laser, and (c) the superposition of (a) and (b) showing the possible laser modes in a simple FabryPerot laser. All three graphs are plots as a function of wavelength.
where int is the internal loss, g is the gain, L is the length of the cavity, and R1 and R2 are power reectivities of the two semitransparent mirrors. The frequency of the laser light is governed by the energy distribution the gain medium is capable of providing and by the geometry of the resonant cavity. The cavity constrains the emission to discrete frequencies of light, or FabryPerot modes, whose half-wavelengths will t in the cavity an integral number of times. The smaller the cavity, the greater the spacing between adjacent FabryPerot modes. Because the round-trip cavity loss is fairly constant for different modes, the laser emission from a simple semiconductor diode laser, usually called a FabryPerot laser, will consist of multiple FabryPerot modes that coincide with the highest gain of the material (see Fig. 3). Edge-Emitting Lasers. Simple edge-emitting semiconductor lasers are formed by rst growing a planar optical waveguide and diode junction by a suitable epitaxial growth technique. The laser is completed when the semiconductor is cleaved in two places to produce reective facets that terminate the planar waveguide. Laser light propagates in the plane of the semiconductor wafer and is emitted from the facet formed by the
Fig. 4. A schematic diagram of a simple FabryPerot semiconductor diode laser. Laser light propagates in the plane of the semiconductor wafer and is emitted from the facet formed by the cleave at the edge of the wafer.
cleave at the edge of the wafer. Figure 4 shows a diagram of a simple FabryPerot edge-emitting laser fabricated in this fashion.The facets formed by the cleave reect due to the change in the index of refraction between the semiconductor (n 3.5) and the air (n 1). The simple mirrors formed by the cleave reect almost all frequencies of light equally well, and hence many frequencies can and do experience enough reective feedback to lase. By replacing one or both of the facet reectors with DBRs, the feedback created by the reection can be made wavelength selective. Thus, only the frequency of light that experiences a strong reection from the DBR will have enough feedback to lase. The DBRs are formed by etching a grating in the semiconductor. The area where material is removed from the etched portion of the grating will have an index of refraction different from that of the unetched material, creating a periodicariation in the refractive index. Vertical Cavity Surface-Emitting Lasers. Advances in epitaxial growth techniques have recently made possible an alternative laser design that promise to provide several advantages over the more traditional edge-emitter laser. The geometry of the vertical cavity surface-emitting laser (VCSEL) is rotated 90 from that of the edge-emitter, and light emits from the surface of the wafer, rather than from the edge (see Fig. 5). The feedback mirrors are formed by epitaxially grown DBRs, which consist of alternating layers of two materials with different refractive indices. This method of creating a periodic change in refractive index is clearly different from the etched grating utilized in edge-emitting lasers. The reason DBRs are used in each of the lasers is also different. In edge-emitting DBR lasers, the grating provides a wavelength-selective mirror, which is useful because it constrains the laser to operate at a single frequency. VCSELs, on the other hand, operate at a single frequency because theresonant cavity is very short, usually only one wavelength long. While the short cavity creates a single-mode operation, it also causes the gain path to be very short. Consequently, the mirror reectivies must be very high in order to satisfy the requirements for lasing. In VCSELs, the DBRs are used to obtain a mirrors with extremely high reectivities.
The Bragg Period

A simplied model of wavelength-selective reection can give insight into how a DBR works. Consider a strong electromagnetic wave, which can be represented by the real part of E0 e jkz , incident upon a structure of period at z = 0 (see Fig. 6). Assume only a negligibly small part, of the incident wave is reected back at each
Fig. 5. A schematic diagram of a VCSEL. The geometry of the VCSEL is rotated 90 from that of the edge emitter, and light emits from the surface of the wafer. The feedback mirrors are epitaxially grown DBRs.
Fig. 6. A schematic diagram of light incident on a structure with a periodic refraction index. In order for the small reections to interfere constructively and provide a large effective reection, the periodic structure must satisfy the Bragg condition, = m/2.
interface. The reected wave at z = 0 can be expressed as the sum of the small reections at each of the interfaces
Note that the phase components of the individual reected waves are integer multiples of 2k , where k is the wave number (2/) of the propagating wave. For these small reections to constructively interfere (sum) and provide a large effective reection, 2k must equal an integer multiple of 2 radians. This model is not strictly valid because we have ignored the fact that the magnitude of the incident wave, E0 , decreases and that the backward traveling reections will be partially reected in the forward direction again. Nevertheless, the model leads to an important concept: a periodic structure can provide a wavelength-selective reection when
DISTRIBUTED BRAGG REFLECTOR LASERS is equal to an integer multiple of half wavelengths
the period,
where we have used the fact that k = 2/ and dened = 0 /n0 . The parameters 0 and n0 are, respectively, the free-space wavelength and the effective index of the laser mode. The order of the grating is designated by the integer m.
Edge-Emitting Single Frequency Lasers

Both distributed feedback (DFB) lasers and DBR lasers utilize Bragg reectors to induce single-frequency operation. The distinction between DFB lasers and DBR lasers lies in the placement of the Bragg grating (see Fig. 7). In a DFB laser, the grating that provides distributed feedback is placed along the entire length of the laser, whereas in the DBR laser, the grating does not overlap the active region, and is used only at the end of the cavity as a wavelength-selective mirror. In 1972, Kogelnik and Shank outlined the principles behind the operation of DFB lasers using a coupled wave model (1). Coupled-mode theory explains the operation of both DBR and DFB lasers; however, the lasers realize single-frequency operation by two distinctly different methods. In a DFB laser, only modes that can propagate in the periodic structure will exist. Of the allowed modes, the mode nearest the Bragg wavelength will lase. In a DBR laser, where the Bragg grating is used as a reector, modes that can propagate through the periodic structure will not experience any feedback and therefore cannot resonate and will not lase. The cavity mode that experiences the strongest reection from the DBR will lase, assuming the mode overlaps the gain spectrum of the material. DFB lasers were demonstrated in 1974 (1), and demonstration of DBR lasers followed in 1975 (2).
Coupled-Mode Theory. Coupling Coefcient. To create a high degree of wavelength selectivity, the periodic structure must
satisfy the Bragg condition, and it must effectively interact with the optical mode in the laser structure. Within a laser cavity, there are both forward- and backward-propagating waves. These waves are coupled through the distributed reections within the periodic structure. The coupling coefcient describes the degree of interaction between the forward- and backward-propagating waves. The coupling coefcient can be determined by applying coupled-mode theory. Let us start with a periodically varying index of refraction prole and use the coordinate system dened in Fig. 6
Note that any index variation prole, n(z), can be expressed in a similar expression by the Fourier expansion of the function that describes the index variation. By making the approximation that the variations of the optical eld in the x and y directions are negligible, the wave equation for the optical eld along the laser cavity ( z direction) can be written as
Fig. 7. Schematic diagrams of a (a) DFB laser and (b) DBR laser, with a diagram of the optical eld intensity inside the laser superimposed. The feedback mechanism (grating) is distributed along the entire length of the cavity in a DFB laser, but is separated from the gain section in for DBR laser.
When n(z ) is squared, the term including (
n)2 can be neglected. Therefore
where = (2/)n0 and = (/) n. The complex propagation constant is assumed to be very close to the Bragg propagation constant 0 .
where is the detuning parameter that represents the separation of from the Bragg propagation constant, and 0 represents the gain or loss in the medium.
Next, we assume that the solution for the eld, E(z), can be written as the superposition of a forward- and backward-propagating components, with propagation constant 0 .
The assumptions of Eq. (10) will result in A(z) and B(z) being slowly varying functions of z. Equations (9) and (11) can then be substituted into Eq. (7). The rst term of Eq. (7) becomes
The second derivative terms, d2 A/dz2 and d2 B/dz2 , can be neglected because, as mentioned before, A and B are slowly varying functions
The second term of Eq. (7) becomes
where the Euler identity has been used and third-harmonic terms have been neglected. Collecting terms with common phase components and utilizing the assumption that coupled-mode equations
, result in the
These equations show that the forward-propagating term A(z) is coupled to the backward-propagating term, B(z), by , and vice versa. Thus far, the periodic index variation has been assumed to be innite in the x and y directions. However, in a laser structure,the gratings (the periodic structure) have a nite dimension. Therefore, only a portion of the optical mode overlaps the periodic index variation. The expression for the coupling coefcient now must account for the spatial extent ( x direction) of the index variation
Fig. 8. Schematic diagram of the coupling coefcient , as a function of grating duty cycle for three different ordered gratings. The coupling coefcient is larger for lower order gratings. For odd-order gratings, there is maximum coupling at 50% duty cycle. For even-order gratings, a null exists at 50%.
Here, d(x) is the Fourier expansion of the periodic index variation. Therefore, to obtain a large coupling coefcient , the optical mode and the periodic structure should overlap strongly [large E(x)], or the index variation d(x) should be sufciently large. Depending on the order of the grating, different duty cycles are required to achieve maximum coupling. A duty cycle is the length of the high-index region within the period devided by the period. Figure 8 shows a schematic diagram of how the coupling varies with the duty cycle. For odd-order gratings, there is maximum coupling at 50% duty cycle. For even-order gratings, a null exists at 50% duty cycle. The coupling coefcient is larger for lower-order gratings. In summary, several factors must be considered to achieve the desired coupling coefcient: the placement of the gratings within the laser structure, the order of the gratings, and the duty cycle are all factors that affect the coupling coefcient. Reectivity. The reectivity of a periodic structure is determined from the coupled-mode equations. Note that the coupled-mode equations are linear, rst-order differential equations. Therefore, the general solutions to the coupled-mode equations have the form of
Substituting Eqs. (18) and (19) into Eqs. (15) and (16), the coupled-mode equations, yields
10
Fig. 9. Reectivity of lossless ( 0 = 0 ) DBR gratings for L = 100 m with = 100 cm 1 and = 400 cm 1 . Note that larger yields a higher reectivity and a wider stop band.
In order for these equations to result in a nontrivial solution, the determinants of both matrices must equal to zero. Making this assignment yields
Now, consider a periodic structure of length L with a forward traveling wave incident on it. For simplicity, assume that the strength of the incident wave, A(z = 0), is known, the structure is lossless ( 0 = 0 ), and the backward traveling wave is zero at L, B(z = L) . These boundary conditions reduce Eqs. (18) and (19) to a system of two equations and two unknowns. Using Eq. (22), it is possible to derive expressions for A(z) and B(z) in terms of A(z = 0) . The reectivity at z = 0 is then expressed as a ratio between the backward- and forward-propagating components:
A plot of |r(0)|2 versus , the detuning parameter, for two different values of is shown in Fig. 9. The region of high reection near = 0 is called the stop band. The plot clearly shows that the reectivity and the width of the stop band increases with increasing . The reectivity also increases with the length of the grating, L.
11
Fig. 10. A schematic diagram of a holographic interference lithography setup. An interference pattern is formed when the two beams are brought together on the surface of the sample. The interference pattern is photolithographically transferred to the sample, which is spin-coated with photoresist.
Grating Fabrication. In 1975, Reinhart et al. used holographic interference and ion milling to create a third-order DBR grating. Since then, the fabricationand epitaxial technologies have improved steadily, and these advances allow the design and fabrication of more sophisticated and complex devices. The small dimensions of the gratings, on the order of a few hundred nanometers, preclude the use of conventional photolithography for their fabrication. Although recent advance have greatly reduced the minimum feature size attainable by photolithography, two other methods are more common for gratings fabrication today: holographic interference and direct-write electron-beam lithography. A typical setup for holographic interference is shown in Fig. 10. The output of a laser is split into two beams that are expanded and collimated. An interference pattern is formed when the two beams are brought together on the surface of the sample. The interference pattern is photolithographically transferred to the sample, which is spin-coated with photoresist. By choosing a laser with the proper laser wavelength () and controlling the angles of the sample and beam incidence ( and ), it is possible to fabricate gratings with different periods (pitches)
This simple technique has been used since the rst fabrication of DBR lasers. Because of its high throughput, the holographic interference method is the most common fabrication technique used to produce DBR and DFB lasers commercially. More complicated exposure schemes using multiple resist and phase shifting, among others can be used to create gratings with more complex characteristics. Electron-beam direct-write lithography is an alternative method for fabricating gratings. An electron beam is used to write gratings on a sample spin-coated with a resist, typically polymethylmethacrylate (PMMA). The accuracy of the period and duty cycle of the grating generated by electron beam direct-write lithography depends on several factors, including the electron-beam current, the electron-beam size, and the scanning system. Although electron-beam direct-write lithography has been used togenerate gratings for high-performance lasers, slow writing speed and high system cost limit its application in the commercial sector. Once gratings have been lithographically transferred into a resist, the next step is to etch the grating pattern into the underlying semiconductor.Wet etching is a simple and easy process that causes little damage to the semiconductor crystal. Precise etch depths can be achieved by utilizing selective wet etches. Lateral dimensions can be much more precisely controlled by using dryetching techniques rather than using a wet etch. However, dry etching can cause crystal damage that may need to be repaired before further processing.
12
The location of the gratings in the laser structure and the material system of the epitaxial layers determine the subsequent processing. Typically,gratings are placed near the active region of the laser structure, where the optical mode is strongest. Locating the grating near the peak of the optical mode creates strong coupling, but it requires that the epitaxial growth of the laserbe done in two steps, with the grating etch performed in between the two growths. Because the optical mode interacts strongly with the grating, large variations in the refractive index are not necessary [see Eq. (17)] and the depth of the gratings can remain small, which is desirable when regrowth is necessary. Special care must be taken to preserve the shape of the gratings during subsequent epitaxial growth because they can be deformed by mass transport. In the InPInGaAsP material system, regrowth is fairly trouble-free when the sample is properly prepared. However, in the GaAs AlGaAs material system, regrowth over gratings is can be problematic because of the highly reactive nature of Al-containing compounds. To circumvent this problem, aluminum-containing laser structures areusually grown in a single step, and gratings are etched on the surface of the laser. However, the placement of the gratings on the surface reduces the amountof interaction between the optical mode and the grating because the optical mode is tightly conned within the cladding layers, and only the tail of the optical mode can interact with the grating. As seen in Eq. (17), to achieve an appreciable value for , a large index change in the grating structure is necessary to compensate for the small overlap of the optical mode with the grating. To create this large change in refractive index, the gratings must be etched fairly deeply ( 0.8 m 1.0 m ) into the epitaxial layers. Because the grating dimensions are only a few hundred nanometers and the etch depth may be a 1 m or more, the aspect ratio of the grating is very large. The task of etching these large-aspect-ratio features into the semiconductor while preserving the period, duty cycle, and shape is difcult to perform even with dry etching processes such as reactive ion etching (RIE). More sophisticated dry etching techniques such as chemically assisted ion beam etching (CAIBE) are often necessary to achieve the highly anisotropic etch demanded by the high-aspect-ratio grating. DBR lasers with a thinner upper cladding, often called asymmetric cladding lasers, can be used to circumvent this difculty. Because the upper cladding is thinner, typically 0.3 m to 0.4 m, the eld is stronger at the surface and adequate coupling can be achieved even with a shallow grating etch ( < 0.25 m ). Ridge waveguides, buried ridge waveguides, and buried heterostructures are some of the device congurations to provide lateral connement, both optical and electrical, necessary for improved performance of the DBR laser. Wavelength-Tunable DBR Lasers. For various applications such as wavelength-division multiplexing and light detection and ranging LIDAR spectroscopy, wavelength-tunable DBR lasers are highly desirable. It is possible to tune DBR lasers efciently by injecting a current into the DBR section of the laser (3). Current injection causes the refractive index of the semiconductor to change, which in turn changes the Bragg wavelength of the grating, as seen in Eq. (5). Figure 11 shows a schematic diagram of a wavelength-tunable DBR laser. The contact pads of the gain section and the tuning section of the laser are isolated so that each section can be biased independently. There are two mechanisms by which current injection changes the refractive index of a semiconductor. Injected free carriers and band-lling effects cause a decrease in the index of refraction. This phenomenon is referred to as the plasma effect and is described by
where n0 is the index of refraction, ne is the electron concentration, and me is the effective electron mass. Current injection also causes heating of the laser structure, which creates in an increase in the index of refraction. Therefore, injecting carriers to the DBR section can tune the output of a DBR laser to a longer or shorter wavelength, depending on which mechanism dominates. The plasma effect increases as a function
13
Fig. 11. Schematic diagram of a tunable DBR laser showing the isolated contact pads of the gain section and the tuning section of the laser so that each section can be biased independently. It is possible to tune DBR lasers efciently by injecting a current into the DBR section of the laser.
of 2 . Thus, in the long-wavelength InP material system, current injection into the tuning section decreases the Bragg wavelength. The heating mechanism dominates in the shorter-wavelength GaAs material system, resulting in an increase of the Bragg wavelength. Tuning is not continuous with tuning current but is interrupted by mode hops. As the peak Bragg reection shifts, the lasing wavelength also shifts. When the peak shifts over far enough, an adjacent cavity mode will experience a higher reectivity. This selection of successive cavity modes causes mode hopping in the wavelength tuning characteristic. The addition of a third phase controlling section can extend the continuous tuning range (4). The tuning range is typically limited by the amount of index shift that can be achieved in the mirror ( / = n/n ). A maximum tuning range of 1% of the wavelength is common (5). Increased tuning range can be achieved when more complex structures and tuning schemes such as superstructure gratings and sampled gratings are employed (6,7). Integrated DBR lasers. Because the gratings eliminate the need for cleaved facets, and the reectivity can be controlled by the length, depth, and order of the grating, it is possible to design and fabricate DBR lasers with monolithically integrated waveguides, modulators, and ampliers. The separation of the active and grating sections of the laser enables the laser to be interconnected with other optical components using a continuous waveguide that permits high optical coupling between the source and these components. The laser structure can be optimized for improved performance with these devices by smoothly integrating regions with different bandgaps. One technique for achieving in-plane bandgap tuning within a single epitaxial growth is selective area epitaxy (SAE). For example, Lammert et al. have demonstrated an electroabsorption modulator integrated with a DBR laser utilizing SAE (8). The bandgap of the modulator section is designed to be larger than the bandgap of the laser. Therefore, the modulator is transparent until a modest reverse bias is applied to shift the absorption peak to extinguish the output signal from the DBR laser.
Vcsel
The concept of the VCSEL, a surface-emitting diode laser formed by sandwiching a pn junction between two epitaxially grown DBR mirrors, was initially conceived by Soda et al. at the Tokyo Institute of Technology in 1979 (9). The VCSEL, although requiring more complex and exacting crystal-growth processes, has several advantages over the edge-emitting laser. Aside from the more complicated epitaxial growth, the VCSEL is a more attractive candidate for manufacturing. Because the VCSEL does not require etched or cleaved facets, these devices can be fully tested at the wafer level, before committing to further processing steps. It is also much smaller than an edge-emitting laser, so more devices can be produced from each wafer. The all planar processing of the VCSEL also facilitates integration of the VCSEL with other electronic devices.
14
The attribute of surface emission, in addition to eliminating the need for cleaved facets, makes possible the fabrication of two-dimensional arrays of lasers, which lends itself to applications in parallel communications. Or, by allowing the growth thickness to change across the wafer, each VCSEL in the array can be made to lase at a slightly different frequency, which allows for wavelength-division multiplexing (WDM). One serious drawback of the edge-emitting laser is its highly elliptical and astigmatic beam pattern, which arises from the aspect ratio of the lasers aperture (thin and wide). Such a beam pattern requires the use of additional optics in many applications, such as coupling to an optical ber. There is much more control over the design of the aperture and beam pattern of the VCSEL, and consequently the emission beam pattern has much better characteristics. The smaller size of the VCSEL improves operating performance because of smaller drive currents, capacitances, and power requirements. Moreover, because the cavity of the VCSEL is very short, only one longitudinal mode can exist, so the emission is inherently single longitudinal mode. A drawback of the small size is that the VCSEL output power is small as well. VCSEL Design. Because the gain path in a VCSEL is very short (twice the length of the cavity, typically 2 ), the reectivity of the cavity mirrors must be extremely high ( >99% ) to satisfy the requirement for lasing
where u and l are the total eld reectivities of the upper and lower mirrors, L is the length of the cavity, and 0 is the gain coefcient. The equation for the net reectivity of a VCSEL DBR mirror is given by the formula for a plane wave experiencing multiple reections and is found in many textbooks
Here i is the net eld reectivity at layer i, i is the propagation constant in layer i, li eff is the effective thickness of layer i, and ri is the local reectivity between the layers i and i+1. In order to make large, two conditions are required: (1) a large number of quarter-wavelength-thick layers in the Bragg reector, and (2) two materials with contrasting indices of refraction out of which to make the pairs. Because the VCSEL has only one longitudinal mode, which is determined by the length of the cavity, the cavity length must be grown with great precision to attain the desired wavelength. The length must be such that the resulting mode overlaps the gain spectrum and the stop band of the DBR. Mirror Fabrication. The VCSEL is perhaps the most challenging optoelectronic structure to be created by crystal-growth techniques. While the concept for the VCSEL has been around for a long time, its performance had been limited largely due to the complexity of mirror fabrication. InGaAs and AlGaAs lasers, which operate at shorter wavelengths (800 nm to 1100 nm), utilize AlAs GaAs DBRs. AlAs and GaAs have large differences in their indices of refraction and are almost perfectly lattice matched, which makes them ideal for use in a DBR. AlAsGaAs mirrors, as semiconductor materials, can be doped to allow current to ow directly through them to the active region of the laser. Even though the AlAs and GaAs have strongly contrasting indices of refraction, more than 20 pairs are needed in the DBR to obtain the necessary reectivity. The multiple abrupt heterointerfaces in the mirror can create a signicant electrical resistance, which leads to heating and higher power requirements. This effect can be reduced by compromising
15
to some degree on optical design. By grading the composition and heavily doping the heterointerfaces this series resistance can be reduced and overall performance of the VCSEL is improved (10). Lasers based on InP-based material systems, which provide the longer wavelengths suitable for use in optical communications, lack semiconductor materials that are both lattice matched and have sufcient contrast in their indices of refraction to be good candidates for materials in a DBR. Without a substantial difference in refraction indices, the number of DBR pairs required poses problems both in the precision of the extended growth and with incurred diffraction losses. One solution has been to use dielectric materials having disparate refraction indices, such as ZnSeCaF2 or TiO2 ZnO2 , in the Bragg reector in place of semiconductors. Because of the insulating nature of these dielectrics, this technique leads to problems in making electric contacts and with heat dissipation. Another solution has been to use the same AlAsGaAs DBRs that are used to make VCSELs at the shorter wavelengths. While AlAsGaAs mirrors cannot be grown directly on a material to which it is not lattice matched, such as InP, it can be grown separately and then later fusion bonded onto another material (10). Fusion bonding is a process using heat and pressure to adhere two semiconductors together. Lateral Connement. Another design aspect that has received a great deal of attention is the denition of the lateral dimensions of the VCSEL. Lateral denition of the cavity is needed both to efciently funnel carriers to the active region and to provide connement for the optical mode. Several techniques have been used for this process. One method is simply to etch away all the surrounding material from the VCSEL, leaving a so-called air post VCSEL. This provides strong electrical and optical connement. A disadvantage of this technique is that the resulting morphology is nonplanar, making placement of electric contacts difcult. In a second technique, rather than etching away the surrounding material, it is rendered electrically insulating by ion bombardment. This technique has proved to be highly reliable and is currently used to produce VCSELs commercially. For VCSELs using AlAsGaAs mirrors, selective oxidation is a recently developed technique that has proved most successful in creating small apertures, leading to smaller devices with lower threshold currents and higher efciencies (11). In selective oxidation, aluminum-containing layers of the DBR mirror are oxidized, providing both electrical and optical connement. Holes are etched in the perimeter of the VCSEL, and the VCSEL is placed in a steam environment at elevated temperatures. Oxidation initiates from the etched holes, creating a ring of oxide, the center of which forms the aperture of the VCSEL. The aperture size can be closely controlled by how much material is allowed to oxidize.
Conclusion
Both edge-emitting DBR lasers and VCSELs are important laser sources for a variety of applications ranging from optical communications to spectroscopy. Because of their advantages over traditional FabryPerot semiconductor lasers, signicant resources have been directed towards active research and development of both laser structures. The single-longitudinal-mode operation of DBR lasers is a major advantage over FabryPerot lasers. In addition, because the cavity is dened by gratings rather than a cleave, DBR lasers can be monolithically integrated with other optoelectronic components. The ability to tune the output wavelengths by current injection makes DBR lasers excellent candidates for wavelength-division multiplexing (WDM) applications. Compatibility with current integrated-circuit fabrication technologies, the possibility of two-dimensional array congurations, and on-wafer testing enable inexpensive fabrication and packaging of VCSELs. As VCSEL technology matures, the numerous advantages of the VCSEL design will undoubtedly allow it to displace lightemitting diodes and edge-emitting lasers in some extisting applications as well as to foster new applications.
16
BIBLIOGRAPHY
1. H. Kogelnik C. V. Shank, Coupled-wave theory of distributed feedback 011lasers, J. Appl. Phys., 43: 23272335, 1972. 2. F. K. Reinhart, R. A. Logan, C. V. Shank, GaAsAlx Ga1 -x As injection lasers with distributed Bragg reectors, Appl. Phys. Lett., 27: 4548, 1975. 3. S Murata, I. Mito, K. Kobayashi, Spectral characteristics for a 1.5 m DBR laser with frequency-tuning region, IEEE J. Quantum Electron., QE-23: 835838, 1987. 4. S Murata, I. Mito, K. Kobayashi, Over 720 GHz (5.8 nm) frequency tuning by a 1.5 m DBR laser with phase and Bragg wavelength control region, Electron. Lett., 23: 403405, 1987. 5. Y. Kotaki H. Ishikawa, Wavelength tunable DFB and DBR lasers for coherent optical bre communications, IEE Proc.-J, 138: 171177, 1991. 6. Y. Tohmori,, et al. Broad-range wavelength tuning in DBR lasers with super structure grating (SSG), IEEE Photon. Technol. Lett., 5: 126129, 1993. 7. V. Jayaraman,, et al. Extended tuning range in sampled grating DBR lasers, IEEE Photon. Technol. Lett., 5: 489491, 1993. 8. R. M. Lammert,, et al. MQW wavelength-tunable DBR lasers with monolithically integrated external cavity electroabsorption modulators with low-driving voltages fabricated by selective-area MOCVD, IEEE Photon. Technol. Lett., 8: 797799, 1996. 9. H. Soda,, et al. GaInAsP/InP surface emitting injection lasers, Jpn. J. Appl. Phys., 18: 23292230, 1979. 10. Y. Ohiso, et al. T. Kurokawa, Long-wavelength (1.55-um) vertical-cavity lasers with InGaAsP/InP-GaAs/AlAs DBRs by wafer fusion, IEEE J. Quantum Electron., 34: 1904711913, 1998. 11. W. W. Chow,, et al. Design, fabrication, and performance of infrared and visible vertical-cavity surface-emitting lasers, IEEE J. Quantum Electron., 33: 18101824, 1997.
READING LIST
G. P. Agrawal (ed.), Semiconductor Lasers: Past, Present, and Future, Woodbury, NY: American Institute of Physics, 1995. J. Buus, Single Frequency Semiconductor Lasers, Bellingham, WA: SPIE Optical Engineering Press, 1991. H. Casey M. Panish, Heterostructure Lasers, New York: Academic, 1978. K. D. Choquette, Vertical-cavity surface emitting lasers: moving from research to manufacturing, Proc. IEEE, 85: 1730 1739, 1997. L. A. Coldren S. W. Corzine, Diode Lasers and Photonic Integrated Circuits, New York: Wiley, 1995. K. S. Giboney, L. B. Aronson, B. E. Lemoff, The ideal light source for datanets, IEEE Spectrum, 35: 4353, February 1998. T. L. Koch U. Koren, Semiconductor photonic integrated circuits, IEEE J. Quantum Electron, 27: 641653, 1991.
S. D. ROH R. B. SWINT J. J. COLEMAN University of Illinois
}{{}}
HOME
q
ABOUT US
q
CONTACT US
q
HELP
q

Distributed Feedback Lasers Standard Article Radhakrishnan Nagarajan1 1SDL, Inc., San Jose, CA Copyright 1999 by John Wiley & Sons, Inc. All rights reserved. DOI: 10.1002/047134608X.W6302 Article Online Posting Date: December 27, 1999 Abstract | Full Text: HTML PDF (210K)
q q
Browse this title

q

r
r r
Abstract
The sections in this article are DFB Device Structure and Material Characteristics Wavelength Division Multiplexing Analytic Treatment Device Characteristics About Wiley InterScience | About Wiley | Privacy | Terms & Conditions Copyright 1999-2008John Wiley & Sons, Inc. All Rights Reserved.
DISTRIBUTED FEEDBACK LASERS
705

Distributed feedback (DFB) lasers are a special class of semiconductor diode lasers. They have found widespread application in ber optic telecommunication systems, where they are essential for the operation of long-haul ber links. DFB lasers have a much narrower wavelength emission spectrum compared to the conventional diode lasers, and they emit light essentially at a single wavelength. For this reason they are also referred to as single frequency lasers. Diode lasers operate on the same amplication of stimulated emission principle as other laser systems. To achieve this light amplication, lasers are composed of a gain medium inserted between two mirrors. The mirrors provide the positive feedback needed to initiate laser action, as external excitation is applied to the active medium. This conguration of an active region and mirrors is referred to as the laser cavity. The mirrors are generally not completely reecting, so some amount of light leaks out and is collected as output from the cavity. As the active medium is excited, or pumped, the excitation is converted to light by the gain medium. The light begins to propagate within the cavity formed by the mirrors and the optical eld starts to build up in intensity. Laser action begins once there is enough light to overcome the cavity and mirror losses. In a typical laser cavity, the feedback from the mirrors is broadband and is not wavelength selective. A passive cavity, that is, in the absence of the gain region, is a resonator which, in principle, will support an innite number of oscillating modes. In a laser, the wavelength of operation depends on the range of wavelengths over which the active medium can provide useful gain. Diode lasers of this type are referred to as FabryPerot (FP) lasers. They generally operate at several different wavelengths or longitudinal (the direction along the cavity) modes. This type of laser is acceptable for many applications except in those where the dispersion in the optical ber becomes detrimental. The mode index (which is a combination of the material refractive index and contributions from the waveguiding structure) of the optical ber varies as a function of the wavelength of light propagating in it. This variation in index, which is commonly referred to as ber dispersion, causes different wavelengths to propagate down the optical ber at different speeds. When the laser signal, which is composed of several different wavelengths from a FP laser transmitter, reaches the receiver, after traveling some distance in the optical ber, it is spread out in time. This results in signal distortion, called the intersymbol interference, and severely limits the transmission distance of ber optic systems. To limit dispersion-induced distortion, one needs a laser source with a narrow emission spectrum.
Emission spectrum of FP lasers can be considerably narrowed by providing wavelength-selective feedback. In DFB lasers, such a wavelength-selective feedback is provided within and throughout the laser cavity. This type of feedback can also be provided by wavelength-selective mirrors. Such a laser is called the distributed bragg reector (DBR) laser. The DBR laser is a diverse subject in itself and will not be discussed here. FP lasers that are externally stabilized using a grating to provide wavelength-selective feedback also fall into this category. The external grating may also be written on the ber used to couple light out of a diode laser in a package. Frequency-tunable lasers can be made using external gratings. The wavelength of the feedback into the laser cavity is adjusted by changing the orientation of the grating with respect to the laser cavity and, by continually changing the orientation of the grating, the laser output can be tuned over a wide range of frequencies.
DFB DEVICE STRUCTURE AND MATERIAL CHARACTERISTICS Electrically, semiconductor lasers are equivalent to pn junction diodes. They are composed of a vertically (or laterally) stacked p and n heterojunction sandwich. The excitation is provided by injecting electrical current in a forward-biased conguration. The current ow is bipolar, that is, the current transport is composed of both electrons and holes. The heterojunction is necessary to conne the bipolar carriers in the same spatial location for efcient recombination, thereby reducing the threshold necessary to overcome cavity losses for laser action. An electron recombines with a hole to produce a photon. The rst-generation diode lasers were of the homojunction type and required very large pump excitation for laser action. Figure 1 shows a drawing of a modern buried heterostructure DFB semiconductor laser diode. Buried refers to the fact that the pn heterojunction gain region of the laser has been completely surrounded by another material. This minimizes the material index variation adjacent to the active region, and it has some desirable waveguide properties for the optical mode within the cavity. The wavelength-selective feedback in the cavity is provided by the mode index or gain/loss variations caused by the grating etched into the semiconductor material. InP/In1xGaxAsyP1y material alloy combination is typically used to make DFB lasers for telecommunication applications. Lasers made of this material combination emit light in the 1.2 m to 1.6 m wavelength. Although DFB lasers have been made from other material systems, most are from GaAs/AlxGa1x. As alloys emitting light in the 0.75 m to 0.85 m region, these have not found use in long-distance data transmission due to the loss and dispersion characteristics of the commonly used silica optical ber. This discussion will only be concerned with the InP based or, more commonly called, the long-wavelength DFB lasers. Figure 1 also shows the conduction band energy diagram of three possible types of active regions. In all three cases the light-emitting layer, the one in the middle (usually made of the InGaAs alloy) has the lowest bandgap energy. The outermost cladding regions are usually composed of InP and these layers have the largest bandgap energies. This combination of materials with different bandgap energies to form the pn junction is referred to as a heterojunction. In a homojunction
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
706
DISTRIBUTED FEEDBACK LASERS Figure 1. Cut-away drawing of a buried heterostructure DFB laser. The active region is the layer above the grating. The doping sequence for the laser structure is p-active region-n from the top to bottom. The sequence for the burying structure is the reverse. In addition to providing good waveguiding properties, the reverse doping sequence of the burying structure forms a current blocking region, thereby channeling the injected current, under forward bias, directly into the active region. The details of the active region conduction band energy structure are also shown. The material layer between two quantum wells is called the barrier. The width of the layer between the outer cladding and the rst quantum well is usually varied to provide maximum overlap between the quantum wells and the optical mode in the cavity. This type of design is called the separate connement heterostructure. The total width of the connement heterostructure (all the layers between the outer InP cladding layers) is about 0.2 m (of the order of the width of the bulk active region).
Buried heterostructure Dielectric layer p contact electrode
n-InP p-InP n-InP InGaAsP cap layer
DFB grating Active region

Increasing energy
n contact electrode Longitudinal direction

Increasing index
InP InGaAsP InGaAsP or InGaAs
Bulk
Single quantum well
Multiple quantum well
laser, the cladding and active regions have the same bandgap energies and no electric potential is present to conne the carriers and facilitate their recombination. In Fig. 1, it is easy to visualize the carriers tumbling down the energy potential of the active region to the lowest level before recombining to emit light. In the bulk active region, the layer width is typically between 0.1 m and 0.2 m. In this case, the carriers are unconned in all three dimensions. As the width of the active layer (the smallest bandgap layer in Fig. 1) shrinks to about 0.01 m, the carriers are quantum mechanically conned in the direction of the smallest dimension, but are free to move in the plane vertical to the paper. These are called quantum well lasers. Quantum well lasers can either have single or multiple wells. InP lasers, in general, tend to have multiple quantum wells (between 4 and 7). Although the cladding regions are p and n doped, the active region proper is nominally undoped. The active region is grown such that it is lattice matched to all the other layers. Doping and strain (by deliberate lattice mismatching of the active region) may be introduced into the active region. If done correctly, strained quantum well lasers and lasers with moderately doped active regions have a number of useful properties, like lower threshold current, narrower linewidth, and higher direct modulation bandwidth. Figure 2 shows the attenuation characteristics of the silica ber most commonly used in ber optic transmission. The minimum in the loss characteristics occur at about the 1.55 m wavelength, and hence, the relevance of DFB lasers emitting at this wavelength. The window at 1.3 m wavelength is traditionally signicant because the dispersion of the standard step index optical ber goes to zero at this wavelength (technically, it is the rst-order dispersion term that goes to zero at this wavelength, but dispersion has other higher-order terms that then become signicant). In modern bers this wavelength, also call the zero dispersion wavelength, can be tailored to match the loss minimum at 1.55 m. This type of optical ber is called the dispersion shifted ber.
The very strong interest in the 1.5 m region is also due to the ready availability of erbium-doped ber ampliers (EDFA) for boosting signals at this wavelength. Similarly, praseodymium-doped ber ampliers (PDFA) can be used to boost signals in the 1.3 m wavelength region. The amplication bands for both wavelength regions have been superimposed on the ber attenuation characteristics in Fig. 2.
100 1.3 m window

Loss (dB/km)
10
1.5 m window
0.1 0.9
1.0
1.1
1.2 1.3 1.4 Wavelength ( m)
1.5
1.6
1.7
Figure 2. Attenuation characteristics of the silica ber which is commonly deployed in the ground. The loss minimum occurs at around 1.55 m wavelength. The peaks in the absorption curve near the 1.3 m region is due to the hydroxyl ions (water), which are incorporated as impurities in the ber during fabrication. The optical transmission windows at 1.3 m and 1.5 m wavelengths are also shown. These lines are merely to show the bandwidth of the windows and are not indicative of any loss values.
707
WAVELENGTH DIVISION MULTIPLEXING The ber has a very large bandwidth for signal transmission. The advent of ber-based optical ampliers and other berbased devices has made it possible to realize this bandwidth over very large transmission distances. One way of utilizing this huge bandwidth is to use wavelength division multiplexing (WDM). Since it is impossible, at least for the present generation of electronics, to take full advantage of all the usable ber bandwidth, WDM systems employ lasers at several different wavelengths, each carrying a high-speed data signal. This is analogous to the subcarrier division multiplexed systems in the microwave domain. For instance, the conventional amplication band in the EDFA is about 32 nm wide. Current commercial transmission systems can accommodate signals spaced at 100 GHz or 0.8 nm apart for a total of 40 channels. Each of these channels run at the SONET (Synchronous Optical Network) OC-48 standard data rate of 2.48832 Gbit/s for an aggregate data rate close to 100 Gbit/s. There are proposals to halve the channel spacing and quadruple the data rate for an eightfold increase in data throughput to 800 Gbit/s in a single silica ber. This data throughput can be further enhanced with the new generation EDFAs, which, in laboratory tests, have demonstrated as much as 80 nm bandwidth in the 1.5 m wavelength region. As the wavelengths are packed together for higher and higher data throughput, the term dense WDM (DWDM) systems is coming into common usage. Modern-day DWDM systems increasingly need DFB lasers with tighter wavelength control and higher spectral purity (this translates to a requirement for narrow linewidth or low phase noise DFB lasers) for proper implementation. The ultimate limit to DWDM systems is the coherent transmission system. A review of the current state-of-the-art in components for optical ber telecommunication systems may be found in the two volume set edited by Kaminow and Koch (1). Two good textbooks in the area of semiconductor lasers are by Coldren and Corzine (2), and Agrawal and Dutta (3). ANALYTIC TREATMENT Distributed Feedback Model Detailed analysis of a DFB laser is complicated, and is only possible using numerical techniques. We present an analytic model which explains all major properties of DFB lasers without having to use numerical techniques. We follow the analysis used in the seminal paper on this subject by Kogelnik and Shank (4). The idea is not to replicate their work, but to provide an overview of the analysis and also supply a number of missing steps in the derivation that may prove useful to the reader. Starting point of the analysis is the scalar wave equation for the electric eld. E2 + k2 E = 0 z2 (1)
2 where k0 200 is the propagation constant in the vacuum and the complex permittivity, tot, of the medium has been written as a sum of its real and imaginary parts. Consider a nonmagnetic dielectric medium, 0. If the refractive index of the medium is n and the gain in the medium is (in units of inverse length), then the complex refractive index of the medium, ntot, can be written as
ntot = n + j
(3)
Assuming that the gain is small over distance of the order of a wavelength, /2 n,
tot 2 = n2 tot n + j
(4)
The expression for k2 in Eq. (2) can then be written as

2 k2 = k2 1+ j 0n
2 = k2 1+ j 0 n (z )
2(z ) k0 n (z )
(5)
where k0 2 / . In a DFB structure, both n and vary periodically, and they are taken to be a function of the z coordinate, that is, the longitudinal direction in the laser. The periodic spatial variation of the index and gain along the z direction in a DFB laser cavity can be written as n (z ) = no + (z ) = o + n cos(20 z ) cos (20 z ) (6)
where 0 is the propagation constant of the waves at the Bragg condition. If 0 is the period of the distributed feedback structure, 0 / 0. At the Bragg condition, 0
0
2 /n0
(7)
which implies that the spatial periodicity, 0, is equal to half the wavelength of the light in the medium, ( /2)/ n0. This is an important result for all devices that depend on some form of a distributed reector for their operation. Although this result has been assumed here, it can shown to be true using Fourier analysis of wave propagation in periodic structures (2). Substituting Eq. (6) into Eq. (5), one obtains the following expression for the propagation constant: k2 2 + 2 j 0 + 4 n +j cos (20 z ) 2 (8)
In deriving the expression for k in Eq. (8), k0n0 and is assumed that n n0, 0, and 0 0. Equation (8) can be rewritten in terms of a coupling constant as
k2 2 + 2 j 0 + 4 cos (20 z ) = 2 + 2 j 0 + 2(e 2 j 0 + e2 j 0 ) n +j r + j i = 2 (9)
where E is the complex amplitude of the electric eld. This eld varies with angular frequency . The propagation constant, k, can be written as k 2 = 2 = 2 0 0 (
r
+ j i ) = k2 0(
+ j i)
(2)
The coupling constant denes the strength of the feedback provided by the gratings in the DFB laser. The expression for k2 can then be substituted into the scalar wave, Eq. (1).
708
Coupled Wave Description The scalar wave equation, in principle, will have an innite set of solutions each corresponding to a certain diffraction order of the propagating wave. Consider the lowest-order solution to the equation close to the Bragg frequency. This corresponds to a forward -and-backward traveling wave. In the absence of any perturbation, these basic modes of the waveguide are orthogonal and do not couple, but in the presence of index and/or gain variations in the laser cavity they scatter into one another and form the basis of the coupled wave description of the DFB laser. The sum of the complex amplitudes of the forward-and-backward traveling waves, which will form the trial solution to the wave equation, is written as E ( z ) = R ( z )e
j0 z
conditions at the facet play a large role in the steady-state evolution of the optical eld within the laser cavity. To simplify the analysis here, assume that both facets are anti-reection coated, that is, the forward-propagating wave is not reected at the right-hand-side facet (thus not contributing to the initial value of the backward-propagating wave) and the backward-propagating wave is not reected at the left-handside facet (thus not contributing to the initial value of the forward-propagating wave). These boundary conditions can be written as R(L/2 ) = S(L/2 ) = 0 (15)
+ S ( z )e
j0 z
(10)
Substituting Eq. (9) and Eq. (10) into Eq. (1),
2R R 2 0 2 j 0 R + 2 R + 2 j 0 R + 2 S e j 0 z 2 z z + 2S S 2 0 + 2 j 0 S + 2 S + 2 j 0 S + 2 R e j 0 z z2 z + 2 Re 3 j 0 z + 2 Se3 j 0 z = 0 (11)
Since it has been assumed that the perturbations in the gain and index of the medium are small, 2R / z2 and 2S / z2 can be neglected. If the coefcients of each of the harmonic components are independently set to zero, one obtains a pair of coupled-wave equations:
Here it has been assumed that the total cavity length is L extending from z L /2 to z L /2. In a FP laser, the cleaved, uncoated, semiconductor crystal facets provide about 30% power feedback, which initiates and sustains laser action by overcoming the losses with the cavity. For all practical purposes, this feedback is uniform over all frequencies and such a laser is not wavelength selective. In the DFB structure, only frequencies at or close to the Bragg frequency will be supported by the cavity. If there is additional feedback from the facets (cleaved and uncoated) of the DFB laser, then the natural FP modes of the laser cavity will not be completely suppressed, leading to poor single-mode oscillation characteristics. The wave equations in Eq. (14) can be rewritten as
R 0 + R j z 0 S 0 + S j z 0
2 2 0 2 0 2 2 0 2 0
R= j S= j
S 0 R 0
2R [ 2 + (0 j )2 ]R = 0 z2 2S [ 2 + (0 j )2 ]S = 0 z2
The general solution of these equations is of the form: R = r1 e z + r2 e z S = s1 e z + s2 e z where the complex propagation constant is given by 2 = 2 + (0 j )2
(16)
(12)
(17)
When the deviation from the Bragg frequency is small, the coupled wave equation can be simplied by setting / 0 1. A normalized frequency deviation parameter, , is then dened as = 2 0
2 2 0
(18)
0 =
no ( 0 ) c
(13)
With these simplications, the coupled wave equations reduce to
R + (0 j )R = j S z S + (0 j )S = j R z
If is real then R and S will be purely evanescent waves and if is imaginary then R and S will form a standing wave within the cavity. Since it has been assumed that the device is symmetric, the solutions will be such that E(z) E(z) and E(z) E(z). Using this and the boundary conditions, the solutions may be written as
(14)
R(z ) = sinh[ (z + L/2 )] = (e (z+ L/2 ) e (z+ L/2 ) )/2 S(z ) = sinh[ (z L/2 )] = (e (z L/2 ) e (z L/2 ) )/2
(19)
where is the deviation of the oscillation frequency from the Bragg frequency 0. At the Bragg frequency, 0. The coupled wave equations describe a forward-propagating wave that is rst amplied by the medium. This wave is then scattered by the grating at frequencies close to the Bragg frequency into the backward-propagating wave. This scattered wave reinforces the backward-propagating wave in the cavity. Likewise, some of the backward-propagating wave is scattered into the forward-propagating wave. The boundary
These equations describe the longitudinal distribution of the optical modes within the laser cavity. The forward-traveling wave, R(z), builds up from zero at the left-hand end of the cavity at z L /2 to its maximum at the right-hand end of the cavity at z L /2, and likewise the backward-traveling wave, S(z), from the opposite end of the cavity. Now to determine the set of eigenvalues for the cavity structure: This can be done by substituting Eq. (19) [taking the negative solution for S(z)] into Eq. (14). The sum and dif-
709
ference of the resulting equations are taken and the common terms are eliminated. The results is as follows:
[e L/2 + e L/2 ] + (o j )[e L/2 e L/2 ] = j [e L / 2 e L / 2 ] [e L/2 e L/2 ] + (o j )[e L/2 + e L/2 ] = j [e L / 2 + e L / 2 ] (20)
Equation (20) can be again simplied by taking their sum and difference, to obtain Eq. (21):
The phase condition is central to the operation of the various classes of DFB lasers. The implications of the phase condition are listed below. 1. The cavity resonances are spaced approximately c /2n0L apart. This is like any other two mirror, FabryPerot laser cavity of length L. 2. Most conventional DFB lasers are purely index coupled, that is, , i 0. The lowest-order solution occurs for q 1, 0 where v1 vo c /4n0L, vo vo c /4n0L. There is no solution at the Bragg frequency (vo), and hence one has the problem of two degenerate modes in a conventional DFB structure, which are both equally likely to dominate unless something is done to break this degeneracy. 3. One way around the problem of two degenerate modes is to introduce a /4 additional phase shift within the cavity. For the index-coupled case, that modies the phase condition as follows: v vo = q+1 (c/2n0 L ) Now there is a resonance at the Bragg frequency, v1 v0. This shift is introduced in the grating structure, and for symmetry reasons, it is usually done in the middle of the laser cavity during fabrication. 4. The second solution is to fabricate a gain (or loss) coupled DFB laser instead of the conventional index coupled one. In this case, n, r 0. This modies the phase condition as follows: v vo = p+q+1 (c/2n0 L ) where p and q are integers such that p, q . Again there is a resonance at the Bragg frequency, v0,1 v1,0 vo. Generally, most gain-coupled DFB lasers also have some amount of index coupling. 5. The phase condition has been derived for a symmetric cavity. In practice, lasers with cleaved facets are seldom symmetric, and there is a good chance that one of the two degenerate modes will have a more favorable phase condition. Although one of the modes will dominate and lase, it is not possible a priori to determine the lasing wavelength, and this particular mode may not have a high discrimination under all operating conditions. The second mode is usually not completely suppressed, and may dominate under a slightly different operating condition, for instance, a different bias current or temperature. Cleaved facets thus lead to poor single-mode yields in DFB lasers. Commercially, the front facet of a DFB laser is usually anti-reection coated and the rear facet is high reection coated. This breaks the mode degeneracy leading to a better single-mode performance. This also results in a higher front facet output power (compared to the cleaved facet case, where both the front and back facets both have equal reectivities), which is essential for practical applications.
(0 j ) = j e L + (0 j ) = j e L
(21)
These equations can then be combined into one to obtain the complex transcendental equation for , which can then be numerically solved for the modes of the DFB structure. Each of these modes has its own threshold and lasing frequency corresponding to a particular cavity length and coupling strength of the grating. Approximate Solutions Several important results can be obtained without having to resort to a numerical solution of Eqs. (21). Invoke what is known as the high gain approximation to obtain these results. The expressions for given in Eq. (18) can by simplied by using the high gain approximation, that is, 0 ( r ji n / j /2). 0 j (22)
Substituting Eq. (22) into the second expression in Eq. (21), 2 (0 j ) = j e( 0 j )L (23)
Although the right-hand side of Eq. (23) is strictly negative, if one were to repeat the analysis starting at Eq. (19), taking the positive solution for S(z), the result would be the positive solution for the right-hand side of Eq. (23). Equation (23) can then be solved to obtain the approximate solutions of the modes of the DFB structure. First derive the phase condition that must be satised by the lasing modes in the cavity. This can be done by comparing the phase of both sides of Eq. (23). tan 1 0 = tan 1 i r L (24)
Near the Bragg frequency, one can assume 0. After substituting for from Eq. (13), Eq. (24) can be simplied to
q+
i 1 2 n 0 ( v v o ) = tan 1 L 2 r c i 1 1 v vo = q + + tan 1 (c/2n0 L ) 2 r
(25)
where 2v and q is an integer such that q .
710
Similar to the phase condition, the absolute value of Eq. (23) is used to obtain the threshold condition for the DFB laser.
2 4 (0 + 2 ) = e2 0 L
TEC
(26)
GRIN lens
From Eq. (26) it can be seen that for a xed value of , as the frequency deviation from the Bragg frequency increases, the threshold gain o also increases. This indicates that a larger gain is required for the higher-order modes to lase and, hence, the mode selectivity of the DFB lasers. DEVICE CHARACTERISTICS Light/Current Characteristics Figure 3 shows the static light/current (L / I) and voltage/current (V / I) characteristics of a packaged DFB laser. The general form of the curves is similar to other semiconductor lasers. This particular DFB laser is meant for high-power ber coupled applications and has a threshold current of 60 mA and an operating voltage of about 1.5 V. The V / I characteristics is similar to an electronic diode, and shows an exponential dependence of the injected current on applied voltage. The DFB laser in Fig. 3 is capable of operating at ber coupled output powers as high as 50 mW. DFB laser packaging styles vary from one manufacturer to another and the details are often trade secrets. Although the details may be different, there are three essential goals in any DFB package. The rst is temperature stability. Single-mode characteristics of DFB lasers are very sensitive to temperature variations. The parameter of major concern is the variations in the emission wavelength with temperature. Most high-end DFB lasers are packaged with a thermoelectric cooler (TEC) for stabilizing the temperature. The second goal is high coupling efciency. Output power from the laser diode is expensive and careful attention is paid to maximize the amount of light that is coupled into a singlemode optical ber pigtail. In manufacturing, ber coupling efciencies in the range of 60% can be obtained. This is achieved by a combination of laser diode design (to obtain a
Strain relief
Back facet monitor Ball lens Isolator

Figure 4. Cut-away drawing of a DFB laser in a buttery style package. The laser diode itself is a small speck to the rear of the ball lens. The ball lens makes the spatial emission pattern of the DFB laser more symmetric. This is followed by the isolator and GRIN lens before the ber. Strain relief prevents the misalignment of the bercoupling mechanism when the package pigtail is stressed during handling.
50 40 30 20
3.0
Operating voltage (V)
2.5 2.0 1.5 1.0
10 Laser threshold 0 0 100 200 300 Current (mA) 400
0.5 0.0 500
Figure 3. The L / I and V / I characteristics of the diode laser. At high drive currents, the series resistance of the diode dominates and the V / I curve tends to become more linear. The output power does not continually increase with injected current. The L / I curve tends to roll over at high bias levels. This is due to thermal effects and is a common phenomenon in diode lasers.
more uniform output spatial emission) and coupling lens design. Typically, a combination of a ball lens next to the laser facet (to correct any residual asymmetry in the emission proles in the lateral and transverse directions to the facet) and a graded index (GRIN) lens at the entry point to the output ber is used to couple light from the diode into the optical ber. The third goal is to minimize the reection of light back into the laser. As seen in the previous section, the wavelength stability of the DFB laser is governed by the wavelength-selective feedback provided by the grating structure. Any other spurious reections, from the laser facet or from an external component in the ber optic link, will lead to poor single-mode performance. Back reection is minimized by properly antireection-coating the lens surfaces and including an isolator in the package. Isolator is an optical device that allows light to be transmitted in one direction with very low loss, and essentially prevents light transmission in the reverse direction. The isolator may be placed after the ball lens and before the GRIN lens. Figure 4 shows a drawing of a packaged DFB laser. This is the AT&T (now Lucent Technologies) Type 246 isolated laser module. The TEC is to the rear of the package, and the laser is mounted on an L bracket which is cooled by this TEC. There is a ball lens followed by an isolator and then a GRIN lens before the ber pigtail, to couple the light output from the laser. The package also incorporates a laser back facet monitor, a pin photodiode, which measures the output power form the back facet of the laser. The photocurrent output from this detector can be appropriately scaled (if the relative facet coating levels are known) to accurately obtain the front facet or ber coupled power. The feedback from back facet monitor is used to operate the DFB is a constant output power mode. Optical Spectrum and Side Mode Suppression Ratio Figure 5 shows the output spectrum of a DFB laser (for the same device whose static L / I and V / I characteristics are given
Fiber coupled optical power (mW)
711
in Fig. 3) at various current levels from below threshold. The output spectrum essentially builds up from noise below threshold to a single-mode output with an acceptable side mode suppression at about 20 mA to 30 mA above threshold. The multimoded output spectrum of a FP laser is shown for comparison. A number of competing optical modes are supported by the optical gain medium in a FP laser, and any one of the modes may dominate depending on the operating conditions, that is, the bias current and temperature. Side mode suppression is a measure of the spectral purity of a DFB laser. Side mode suppression ratio (SMSR) or simply the mode suppression ratio (MSR) is the ratio of the power in the main oscillation peak to the power level in the most intense side mode (or the second most dominant mode). SMSR requirements are application specic, but a value in excess of 30 dB ( 1000) is considered desirable in a single-mode laser. As shown in Fig. 5, SMSR of a DFB laser improves as the power in the main mode increases. Figure 6 shows the SMSR as a function of main mode power. It can be seen that SMSR in excess of 35 dB have been achieved at a ber coupled power of 1 mW. This shows that DFB lasers are highly wavelength selective, and it takes very little power above threshold to achieve essentially single mode operation. Figure 6 also shows that SMSR increases linearly with main mode power
48 46
Side mode suppression ratio (dB)
44 42 40 38 36
10 Power (mW)
100
Figure 6. Dependence of the side mode suppression ratio (SMSR) on optical power. It takes very little output power to obtain a good SMSR. At high power levels, the SMSR begins to degrade due to nonlinear effects caused by high photon densities in the laser cavity.
FabryPerot laser SMSR = 47 dB
1530 1540 1550 1560 1570 1580 Wavelength (nm) 400 mA 1530 1540 1550 1560 1570 1580 1590
200 mA 1530 1540 1550 1560 1570 1580 1590
100 mA 1530 1540 1550 1560 1570 1580 1590
70 mA 1530 1540 1550 1560 1570 1580 1590
60 mA 1530 1540 1550 1560 1570 1580 1590
50 mA 1530 1540 1550 1560 1570 1580 1590 Wavelength (nm)
Figure 5. Spectral evolution of the DFB laser output at different bias current levels. The output builds from noise, and the single-mode characteristics are only well established at current levels above the threshold. The multimoded FP spectrum is provided for comparison.
(Fig. 6 is plot of log of SMSR vs. log of main mode power) up to a point before leveling off and eventually degrading at very high power levels. SMSR values up to 50 dB are possible in modern DFB lasers. The eventual degradation of the singlemode properties at very high power levels is a common characteristic of most DFB lasers. The grating structure in the DFB laser causes the optical eld within the laser cavity to be nonuniform. This spatial nonuniformity leads to what is known as spatial hole burning (SHB) in the laser. These are localized areas in which the optical density is much higher than the average eld in the cavity. The nonlinear effects due to SHB are complicated to analyze and are outside the scope of the treatment here. The output spectrum in Fig. 5 shows a main oscillating peak on a background with some ne structure. There are two interesting features here. One is the ne structure itself and the other is that the main peak has been pulled to the lefthand side (shorter wavelength side) of the background, which gradually peaks at a much longer wavelength. The ne structure is the natural FabryPerot modes of the laser cavity, which have been suppressed by the presence of the grating. Second, the gradually increasing background is the natural gain spectrum of the active region. By adjusting the pitch of the grating structure, one can selectively control the oscillation wavelength of the DFB laser. This is called detuning. There are physical limits as to how much detuning may be used in a DFB laser. Forcing the DFB to operate at the extreme wavelength ends of the material gain may lead to unacceptable increases in the threshold current. Without going into the details, among other things, DFB lasers detuned to the shorter wavelength side generally have better high-speed modulation properties (5). Great care is taken to suppress the natural FabryPerot modes of the cavity in a DFB laser. Figure 7 shows the L / I curve and the output spectrum of an as cleaved DFB laser. Laser facets or mirrors essential for laser action are formed by breaking the device along its crystal planes. These are natural cleavage planes and this process, referred to as cleaving, forms reectors of outstanding optical quality. The facets of the DFB lasers are then coated to break the mode degener-
712
12
100 mA 10
8 1.53 1.54 1.55 1.56 1.57 Wavelength (m)

Power (mW)
60 mA
6 1.53 1.54 1.55 1.56 1.57 Wavelength (m) 4
15 mA 2 1.53 1.54 1.55 1.56 1.57 Wavelength (m) 0 0 20 40 60 80 Current (mA) 100 120
bias current. Maintaining constant power at a higher temperature requires a higher bias current. There are two effects that lead to this wavelength increase. The dominant one is the variation of the mode index (material properties) with temperature. There is also a small carrier-induced contribution to the mode index. As the temperature increases, the larger bias current required for DFB operation at constant power, causes a corresponding increase in the carrier density within the laser cavity (6). These two effects are differentiated in the two curves presented in Fig. 8. The wavelength tuning (slope of the plot) with temperature of the DFB laser in Fig. 8, at constant bias, is about 0.09 nm/ C. For constant power operation, it is about 0.10 nm/ C. The slope is higher for the constant power operation, since it also includes the effect of the increased carrier density on the mode index. As is obvious by now, DFB lasers can also be current tuned. Figure 9 shows the variation is operating wavelength with bias current at room temperature. Although the relationship is not linear, the wavelength increases (approximately) at a rate of 0.1 nm for every 10 mA increase in bias current. Tuning range of the order of 1 nm to 2 nm is possible with either technique. A combination of current and temperature tuning is used in practice to operate DFB lasers at precisely dened wavelengths. Linewidth Another measure of spectral purity of a DFB laser is the laser linewidth. This is usually dened as the full width at half maximum (FWHM) power of the main oscillating mode. It is expressed in frequency units of kHz. Under steady-state constant bias (also called the continuous wave or CW) operation, the linewidth is a measure of the laser phase noise (frequency noise). There is additional broadening of this intrinsic linewidth under modulation due to other cavity mechanisms. Linewidth of a few hundred kHz (1 MHz) is desirable in good DFB lasers.
Figure 7. Output spectrum of an as-cleaved DFB laser. Although this laser is operating in a single longitudinal mode, its SMSR is poor compared to the case in Fig. 5. Since it is not possible to predict a priori which of the two degenerate modes will dominate, the singlemode yield, to a xed wavelength specication, will be poor during manufacturing. This also shows that the single-mode operation of the DFB laser will degrade with feedback from an external surface (in this case it is the uncoated laser facet that is providing the unwanted feedback).
1560
acy (see previous section). If the DFB lasers are operated uncoated, that is, as cleaved (Fig. 7), in addition to the competition between the two degenerate modes, one can also see the remnants of the natural FP modes of the cavity (the more pronounced spectral structure in the background compared to Fig. 5). SMSR for this structure is poor and will be strongly dependent on the operating conditions. Figure 7 also illustrates another problem with DFB laserstheir susceptibility to back reections from an external source. As discussed earlier, care must be taken to eliminate back reections in packaging these lasers. Wavelength Stability and Tuning In operation, both the temperature and bias current affect the output wavelength of a DFB laser. Figure 8 shows the effect of temperature on the operation wavelength of a DFB laser. This effect is also referred to as the temperature tuning of a DFB laser. One of the curves shows the wavelength variation with temperature at constant power and the other at constant
1559
Wavelength (nm)
10 mW 200 mA
1558 1557 1556 1555 1554
10
20 30 40 Temperature (C)
50
60
Figure 8. Temperature tuning curves at constant current and constant output power. Additional bias current is required to maintain constant output power from the DFB laser at high temperatures. At these higher bias levels, the additional index variation caused by the increased carrier density causes the constant power tuning curve to have a higher slope.
713
1560
1559
Wavelength (nm)
1558
1557
1556
laser can be further reduced by using external feedback techniques for noise reduction. Direct current modulation of the DFB laser will broaden its linewidth. The current modulation of the active region modulates both the photon and carrier density in the cavity. Modulation of the carrier density modulates the mode index, which results in varying the effective cavity length. This leads to a variation in the resonant oscillation frequency or the broadening of the laser linewidth. This process is called laser chirp. An expression can be written for the transient (time dependent) chirp that occurs during current modulation of the laser as in Eq. (29) (10). v(t ) = 4 1 P0 (t ) dP0 (t ) dt (29)
1555
100
200 300 Current (mA)
400
500
Figure 9. Wavelength tuning of the DFB laser with bias current at constant temperature.
Theory of noise in semiconductor lasers is complicated (7). The major contribution to noise in all laser systems is spontaneous emission. This is the result of adding the spontaneous emission eld which has random phase to the coherent oscillating eld of the laser cavity. The second contribution, which is peculiar to semiconductor lasers, is the index variations in the laser cavity caused by carrier density uctuations. Without going into the details of the mathematical formulation worked out by Henry (8), the proportionality relationship for the linewidth of a semiconductor may be written as in Eq. (27). f (1 + ) P0
2
Chirp in a laser is proportional to the linewidth enhancement factor and the rate of change of optical power (equivalently the modulation or data rate). Laser chirp under current modulation is a general property of all semiconductor lasers and is not peculiar to DFB lasers alone. Since laser chirp combined with ber dispersion, will limit the practical transmission distance in ber optic systems, great care is taken to design DFB lasers with minimum amount of chirp under modulation, that is, with small parameters. Since laser chirp is unavoidable, single-frequency DFB lasers are not directly modulated in long-distance transmission systems. Instead, an external modulator, which can be independently optimized for low chirp, is used to modulate data onto the optical carrier. Intensity Noise In addition to phase noise, there is also intensity noise in a diode laser. At constant bias, the laser output power uctu-
(27)
Equation (27) shows that the laser linewidth is inversely proportional to the output power P0. The (1 2) factor is referred to as the enhancement to the modied Schawlow Townes expression for the laser linewidth. This enhancement is the result of carrier density uctuation, and , which is called the linewidth enhancement factor is dened as in Eq. (28): = (28)
Linewidth (MHz)
1400 1200 1000 800 600 400 200 0 0.00
dn/dN dg/dN
The linewidth enhancement factor is proportional to the ratio of the index variation with carrier density to the gain variation with carrier density (also called the differential gain). (which should be confused with the notation used for the material gain in the previous section) is a material parameter and is inuenced by the design (dimensions and doping levels) of the active region. It is desirable to keep this parameter as small as possible. Typical values for in DFB lasers are less than 5 with good DFB lasers having values less than half that number. Linewidth as a function of inverse optical power for a commercial DFB module is shown in Fig. 10. The same plot gives the best linewidth data of 3.6 kHz reported to date for a solitary DFB laser (9). The point is made about the solitary diode because the linewidth of the semiconductor
0.05
0.10
0.15
0.20
Inverse power (1/mW)

Figure 10. Linewidth measured on a commercial DFB laser package (solid triangles) compared to what has been reported for some of the best DFB laser diodes in the world (solid circles).
714
DISTRIBUTED MEMORY PARALLEL SYSTEMS
110
Relative intensity noise (dB/Hz)
high-speed lasers, because the intrinsic modulation bandwidth of the laser is proportional to the resonance frequency. BIBLIOGRAPHY
1. I. P. Kaminow and T. L. Koch (eds.), Optical Fiber Telecommunications IIIAIIIB, New York: Academic Press, 1997. 2. L. A. Coldren and S. W. Corzine, Diode Lasers and Photonic Integrated Circuits, New York: Wiley, 1995. 3. G. P. Agrawal and N. K. Dutta, Semiconductor Lasers, 2nd ed., New York: Van Nostrand Reinhold, 1993. 4. H. Kogelnik and C. V. Shank, Coupled-wave theory of distributed feedback lasers, J. Appl. Phys., 43 (5): 23272335, 1972. 5. R. Nagarajan, D. Tauber, and J. E. Bowers, High Speed Semiconductor Lasers, in T. P. Lee (ed.), Current Topics in Electronics and Systems, Vol. 1Current Trends in Integrated Optoelectronics, Singapore: World Scientic, 1994, pp. 144. 6. S. Akiba et al., Temperature dependence of lasing characteristics of InGaAs/InP distributed feedback lasers in 1.5 m range, Japan. J. Appl. Phys., 21 (12): 17361740, 1982. 7. K. Petermann, Laser Diode Modulation and Noise, Dordrecht: Kluwer, 1988. 8. C. H. Henry, Theory of linewidth of semiconductor laesrs, IEEE J. Quantum Electron., QE-18 (2): 259264, 1982. 9. M. Okai et al., Corrugation-pitch-modulated distributed feedback lasers with ultra-narrow spectral linewidth, Japan. J. Appl. Phys., Part 1, 33 (5A): 25632570, 1994. 10. T. L. Koch and J. E. Bowers, Nature of wavelength chirping in directly modulated semiconductor lasers, IEE Electron. Lett., 20 (25/26): 10381040, 1984.
120 FP laser 130 140 Resonance frequency 150 DFP laser 160 170
2 3 Frequency (GHz)
Figure 11. Comparison of the relative intensity noise (RIN) for a FP and DFB laser. The humps in the curve are due to the resonant enhancement of noise, which is a consequence of the nonlinear interaction of electrons and photons in the laser cavity.
ates with time about its steady-state value. The random carrier recombination and photon generation events produce instantaneous time variations in the carrier and photon densities, even in the absence of current modulation or other external disturbances. This uctuation in the laser output intensity is called the relative intensity noise (RIN) and is written as in Eq. (30): RIN P(t )2 P0 2 (30)
RADHAKRISHNAN NAGARAJAN
SDL, Inc.
where P0 is the time averaged output power and P(t) is the instantaneous variation in the average output power. RIN is normalized to per unit bandwidth and is commonly expressed in log units as decibels per hertz. DFB lasers have a much lower level of RIN than FP lasers. The major source of RIN enhancement in FP lasers is the mode competition between multiple longitudinal modes of the cavity, which leads to intensity uctuations. This is known as the mode partition noise. Unlike the DFB lasers, the carriers that recombine in a FP laser can generate a photon in any one of the modes supported by the laser cavity. Figure 11 shows the RIN for a DFB and a FP laser. A good DFB laser designed for analog applications, like the cable television transmission systems, must have RIN at or below 160 dB/Hz. FP lasers generally have 20 dB to 30 dB higher RIN, but these levels are tolerable for most digital applications. The RIN spectrum for both the DFB and FP lasers have a distinct hump, called the resonance peak, which occurs at what is known as the resonance frequency. The resonance phenomenon is a result of the nonlinear carrierphoton interaction in the laser cavity. The resonance frequency is larger in cavities with larger photon densities. It is also larger in active regions with a larger differential gain. For low levels of intensity noise, the laser should have as large a resonance frequency as possible and be operated at high output power levels. It is also good to have a large resonance frequency for
}{{}}
HOME
q
ABOUT US
q
CONTACT US
q
HELP
q

Dye Lasers Standard Article Kamalesh Dasgupta1 1Bhabha Atomic Research Center, Mumbai (Bombay), India Copyright 1999 by John Wiley & Sons, Inc. All rights reserved. DOI: 10.1002/047134608X.W6303 Article Online Posting Date: December 27, 1999 Abstract | Full Text: HTML PDF (250K)
q q
Browse this title

q

r
r r
Abstract
The sections in this article are General Characteristics of Dye Lasers Physics and Technology of Dye Lasers Design Considerations Application of Dye Lasers Future Scope About Wiley InterScience | About Wiley | Privacy | Terms & Conditions Copyright 1999-2008John Wiley & Sons, Inc. All Rights Reserved.
DYE LASERS
77
DYE LASERS
Dye lasers are the most versatile class of lasers. They are based on intense optical excitation of dye molecules, commonly dissolved in organic solvents. The characteristic feature of laser dye molecules that makes them so attractive as laser media is that they exhibit strong absorption and emission (uorescence) of light over large-wavelength bands. Excitation of dye molecules for laser action is induced by optical radiation from other xed-wavelength lasers or from ashlamps emitting radiation over large spectral bands. The kinetics of these photophysical processes is such that the uorescence from excited dye molecules can be converted into a highly directional, coherent laser beam with ultranarrow spectral linewidth that exhibits wavelength variations less than 1 ppm. The wavelength of this highly monochromatic laser beam can be widely tuned over several tens of nanometers. The entire spectrum from near ultraviolet (UV) (310 nm) to near infrared (IR) (1.3 m) and up to 1.7 m with restricted performance is covered by using different dyes. The high intensity of this fundamental output enables further extension in the vacuum UV (VUV) or IR region using nonlinear optical techniques for frequency conversion. Several references (1,2,19) provide curves showing wavelength ranges covered by different dyes. On the other hand, the broad spectral emission can be exploited for generation of a laser beam consisting of ultrashort optical pulses (1010 s to 1014 s). With such diverse output characteristics, it is not surprising that dye lasers have found extensive use in all areas of basic and applied sciences related to energy, environment, health care, and industry as well as for elucidating fundamental processes in nature. Since its discovery by Sorokin and Lankard in 1966 at IBM, this awesome application potential has ceaselessly engaged the attention of the science and technology community and steadily driven a rapid cooperative growth of new dye laser congurations, molecular engineering of dyes, development of high-power pump sources, and novel applications. GENERAL CHARACTERISTICS OF DYE LASERS Lasers based on emission of light from atoms or molecules make use of the discrete quantum energy levels that these particles occupy and transitions between these energy levels, which result in absorption and emission. For laser action, the dye molecules are rapidly excited by absorption of light to a higher energy level such that the population density is more than that in a lower energy level to which the excited moleJ. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
78
DYE LASERS
cules may relax by emission of optical radiation. The frequency of the radiation emitted is related to the energy difference between the levels, E, by the relation h E. Long before the invention of laser, Einstein had classied the emission transitions between quantum energy levels into two different processesspontaneous and stimulated emission. In spontaneous emission, the radiation may be emitted in any direction, with arbitrary phase, polarization, and with a frequency distributed within the spectral width determined by the width of the energy levels. In stimulated emission, radiation incident on the excited medium from any source, with a frequency lying within the emission band, forces the excited atoms or molecules to emit radiation at the same frequency, polarization, and direction and with the same phase as that of the incident radiation. With more molecules in the higher energy level ( population inversion), the rate of stimulated emission exceeds that of absorption, leading to net amplication of the incident radiation. A laser resonator usually consists of curved and/or plane mirrors on opposite sides of the active medium to feed the emitted radiation repeatedly to and fro through the active medium. This serves to enforce amplication of the spontaneously emitted light predominantly along the axis of the resonator, thus helping in the formation of a directed laser output. The divergence of the laser beam is a measure of its directionality and is determined by the transverse eld distribution within the cavity. Detailed analysis and experimental evidence show that the transverse eld distribution of the laser at any plane inside the resonator exhibits one or more of certain steady-state forms (transverse modes) that repeats itself after every round trip in a specic class of resonators called stable resonators. The stability of the resonators is determined by the curvature of the mirrors, their separation and alignment, and the presence of limiting apertures. The minimum divergence is exhibited by the simplest eld distribution (lowest-order transverse mode) that is Gaussian in shape. The fundamental features that distinguish dye lasers from other lasers is the extent of the width of the energy levels participating in the laser action and the physical processes that lead to such broadening of the levels. These radiative absorption and emission processes involve transitions between well-separated energy levels of some of the electrons in the molecule that are relatively mobile. Dye molecules are large organic compounds having a complex structure with several quantied modes of vibration and rotation. This results in association of a densely spaced distribution of vibrational and rotational energy levels with each electronic state. Therefore, optical transitions take place between the vibrational-rotational levels of different electronic energy states. Perturbation of the vibrational-rotational motion caused by electrostatic interactions and collisions with the molecules of the host medium broadens the energy levels. As a result, dyes exhibit broad absorption bands and an emission bandthat is, they are red shifted (vide infra) with respect to the longestwavelength absorption band (Fig. 1). Amplier gain exists over a large part of the uorescence band and allows the wavelength of the dye laser to be tuned over the gain band. The broad absorption band of the dye molecules enables excitation of the same dye by a variety of pump lasers with xed frequencies, and this adds to the versatility of the dye laser. Intense optical excitation that is possible with such sources, or by intensely driven ashlamps, is necessary to pump dye
molecules to the excited state at a rate sufciently faster than the spontaneous emission rate, in order to create and maintain population inversion. If both the mirrors forming the resonator cavity have uniform reectivity over the emission spectrum of the dye, the laser output has a broad spectral widthon the order of several nanometers. To produce a narrowband and tunable output, dispersive optical elements such as gratings and prisms are introduced between the gain medium and one of the end mirrors, often replacing the end mirror itself. When the pump radiation is switched on, initially there is emission of broadband uorescence from the excited molecules. Gratings and prisms disperse the wavelength contents of this broadband radiation in different directions such that a spectrally narrow band of radiation is fed back, along the resonator axis, to interact with the excited molecules. This effect of highly dispersive elements may be depicted as a high resonator loss at all wavelengths except in a narrow band at the laser wavelength, 1 (Fig. 2). As radiation intensity at 1 starts growing, the gain or population inversion near 1 starts reducing due to higher rate of stimulated emission. Fast thermal interactions between the dye molecules and the solvent ensure that all the excited dye molecules interact with the radiation. In this condition the gain is said to be homogeneously broadened, and gain reduction (or gain saturation) occurs all over the gain prole. The growth of laser intensity at 1 and gain saturation continues until the gain is just sufcient to overcome the loss at 1 and a steady-state condition is reached. Thus, homogeneous broadening in dye lasers allows most of the energy available as broadband gain to be extracted within a narrowband laser output. This process of spectral condensation is a key feature that leads to efcient operation of narrowband dye lasers. Even the spectral width of the output of a broadband laser is substantially narrower than the uorescence width, because emission near the peak of the gain spectral prole grows faster during the repeated transits through the gain medium and saturates the gain. Tunability of the output wavelength is achieved by changing the orientation of the dispersive element, which changes the center wavelength of the radiation fed back into the cavity. Gratings are more dispersive than prisms. Although somewhat more lossy than prisms, reective gratings with ruled and specially shaped groove proles, referred to as
Absorption, emission (arbitrary units)
Flashlamp Abs. Fluor. T-Tabs.
0.3
0.4
0.5 0.6 0.7 Wavelength ( m)
0.8
0.9
Figure 1. Absorption (solid line) and uorescence (short dashed line) spectrum of a dye such as Rhodamine 6G in ethanol. Triplet absorption (T-T) often overlaps with uorescence spectrum. Also shown is a spectrum resembling emission spectrum of ashlamps (dotted line).
100
Gain, loss (rel. units)
10
10
Figure 2. Conceptual comparison of loss management and gain saturation in narrowband short-pulse (a) and CW (b) dye lasers. The high loss (dashed lines) in pulsed dye lasers would not allow laser action in CW laser pumped dye lasers. The low loss in CW dye lasers sufces to restrict lasing to a narrow band due to continued gain saturation, but in pulsed lasers it would allow broadband laser emission to persist for a large fraction of the pulse duration.
;;; ;;; ;;; ;;; ;;;

Loss Gain Tuning range Wavelength (a) Gain Loss Wavelength (b)
DYE LASERS
79
Types of Dye Lasers The diverse family of dye lasers is classied by the nature of the corresponding laser output. Continuous wave (CW) dye lasers (1,2) have a continuous, unperturbed output and are usually low-power lasers pumped by CW lasers like argon ion or krypton ion lasers. In comparison to the high-power shortpulse dye lasers, CW dye lasers possess two major advantages. The rst is that the minimum output linewidth can be much smaller as it is not limited by the Fourier transform of the output pulse shape. Second, homogeneous broadening of the gain prole of the dye is exploited to the best extent in CW dye lasers. Different wavelength components within the gain spectrum compete for the same overall gain available, and continue to do so, until the wavelength component that sees the maximum net gain (or minimum loss, as determined by the wavelength selective element in the cavity) grows faster and saturates the gain (Fig. 2). Then other wavelength components see a loss more than the saturated gain and are extinguished. With suitable cavity designs it is thus possible to obtain laser oscillation on a single longitudinal mode. The frequency of this cavity mode undergoes small variations due to changes in the optical length of the cavity caused by uctuations in the gain medium as well as by thermal and mechanical disturbances. Advanced frequency stabilization techniques by servo locking on a stable reference have resulted in commercial lasers with a frequency stability of 0.002 ppm, whereas laboratory systems with linewidth of better than 1 ppb have been demonstrated. The output linewidth is comparable with that of the narrowest atomic transitions in nature. Thus, CW, single-longitudinal-mode, frequency-controlled dye lasers have found immense applications in opening up a new area of ultra-high-resolution spectroscopy. Some of the major achievements in this area of applications are ultraprecision measurements of the fundamental constants in nature, development of high-precision standards of length and time, ascertaining the validity of quantum theory of particles with increasingly higher precision, cooling and trapping of atoms to ultracold ensembles with temperatures well below a microKelvin, and industrial applications like enrichment of nuclear isotopes for generation of nuclear power. Flashlamp pumped dye lasers (FLDL) form a different class of lasers with output pulse duration of a few hundred nanoseconds to a few hundred microseconds and have the highest overall efciencies because the efciency of ashlamps is much higher than that of lasers (35). Xenon-lled ashlamps of linear or coaxial design are generally used for pumping dye lasers. Linear ashlamps have a tubular quartz envelope with tungsten electrodes sealed at the ends. A shortduration high-voltage pulse applied to the electrodes results in intense emission over a large spectral range (Fig. 1). This is coupled to the dye solution by suitably designed enclosures with specular reectors for imaging the lamp on the dye solution. The dye solution is own through a cell made of fused silica at a fast speed for removal of heated dye solution between pump pulses. An important design consideration for these dye lasers is to minimize the risetime of the discharge pulses, which is limited by the impedance of the discharge. Preionization produced by a low dc simmer current, or by prepulsing with 10% of the main pulse, helps in reducing the discharge risetime and also enhances lamp life. Linear ashlamps are used in high-average-power applications. Sys-
blazed gratings, have high efciency and are commonly used as primary dispersive elements in some classes of dye lasers pumped by pulsed lasers or ashlamps. In these lasers the gain is high enough to tolerate the losses introduced by the gratings. Interference of electromagnetic waves plays an important role in determining the laser spectrum. Irrespective of the overall bandwidth of the laser, if the spectrum is recorded with high resolution, one observes a comblike pattern of sharp, closely spaced spectral lines, spanning the spectral envelope. This line structure arises because constructive interference occurs at discrete wavelengths at which the intracavity electromagnetic wave at any plane inside the resonator adds exactly in phase with itself after every round trip within the cavity. These cavity resonances or longitudinal modes appear at frequencies m mc /2L(m integer), where L is the optical length of the cavity. Due to constructive interference of multiple waves reected repeatedly by the end mirrors, the intracavity radiation intensity at these frequencies builds up rapidly. Even slightly away from the resonant frequencies, the waves are superposed with increasing phase lag after every round trip, causing rapid reduction of intensity when a large number of multiply reected waves participates. Thus, radiation at the resonant frequencies grows at a faster rate and saturates the gain.
Gain, loss (rel. units)
80
DYE LASERS
tems with 200 W average power at a pulse repetition rate of 50 Hz and an overall efciency as high as 0.6% have been reported. For high pulse energy but low repetition rate applications, coaxial ashlamps are used. The dye solution ows through the central tube while gas discharge takes place in the surrounding annular channel. Triaxial and quadraxial tubes are used to provide an intermediate cooling water channel and an evacuated acoustic barrier between the dye tube and the plasma tube. Thermal and shock-wave-induced optical inhomogeneities degrade dye laser beam quality and give rise to wavelength uctuations. High pulse energies of 400 J have been obtained from coaxial ashlamp pumped dye lasers with an efciency of 0.8. FLDLs have found limited use in highresolution spectroscopic applications due to difculties in eliminating rapid photodegradation of the dye caused by absorption of UV emission from the lamps and because of the short life of strongly driven ashlamps (106 to 107 pulses). Pulsed laser pumped dye lasers are capable of providing the best combination of high peak power ( several megawatts), repetition rate (10 Hertz to several kilohertz), and minimum linewidth that is limited by the Fourier transform of the pulse (2,46). Short-duration (several nanoseconds to a few tens of nanoseconds), pulsed lasers, such as nitrogen lasers, excimer lasers, copper vapor lasers, and fundamental and harmonic outputs of Q-switched solid-state lasers, are used as pump sources. Because of high intensity of the pump lasers, larger resonator losses than for CW lasers can be tolerated, making these lasers much simpler to set up and use. The high pump intensity also permits the use of transverse pumping schemes in which the pump laser is focused by a cylindrical lens onto a dye cell, forming a thin (100 m to 500 m) pencil-like (10 mm to 20 mm long) gain region. The depth of the gain region inside the solution is adjusted by changing the dye concentration. The dye laser axis is collinear with the elongated gain region. This design makes the gain length independent of the concentration of the dye and its absorption characteristics at the pump wavelength and allows more exibility in optimizing the performance of the system. The large aspect ratio of the gain region also helps in restricting the loss of excitation through amplication of spontaneous emission in directions other than the dye laser axis. Since feedback of narrowband radiation along the laser axis favors its buildup, albeit with a little delay, amplied spontaneous emission (ASE) in this direction is suppressed after an initial growth. In a temporal waveform, ASE shows up as a small hump at the beginning of the pulse, especially when the laser is tuned to the ends of the operating wavelength range, where the gain for the narrowband radiation becomes small. However, at very high pump intensities with resultant high gains, ASE in the direction of the gain length may compete better for the gain and increase to the extent where the gain is predominantly saturated by the ASE itself, signicantly reducing the efciency for the narrowband laser output. The presence of ASE in the laser direction degrades the spectral purity of the laser. Because of inadequate time for spectral narrowing to evolve to its full extent, strongly dispersive elements like gratings are commonly used. Figure 3 shows the two widely used pulse dye laser congurations that are modied versions of those due to Ha nsch (7) and Littman and Metcalf/Shoshan et al. (8). In the former, the grating is used in Littrow cong-
;;;;
=
Pump laser
Cyl. lens
Grating
Dye cell
Etalon
Beam expander (a)
Mirror
LG
Pump laser
C Grating Preexpander (b) Dye cell
Figure 3. Schematic diagram of a current version of Ha nsch (a) and grazing-incidence-grating (b) pulsed dye lasers pumped transversely by short-pulse lasers. Coutput coupler, LGLittrow grating.
uration in which it acts as a spectrally selective end mirror. The laser linewidth is determined by the grating dispersion and the divergence of the beam incident on the grating. The divergence is large (few milliradians) because of diffraction of the radiation at the narrow aperture formed by the pencillike gain region. Current versions employ inclined prisms, instead of an inverted telescope in the old design, to magnify the beam in the plane of grating dispersion. Magnication of a beam reduces its divergence and illuminates a larger number of grating grooves so that the linewidth is reduced. Typical laser linewidths are 0.6 cm1 to 0.2 cm1 ( / 104 to 105) with these systems. For narrower linewidths, FabryPerot interferometers are used between the prism beam expander and the grating. A FabryPerot interferometers consists of two parallel, highly reecting mirrors with a constant separation between the mirrors. Either air-spaced or solid interferometers (etalons) made of a single piece of fused silica glass, and coated with dielectric material for reectivity 85%, are used in pulsed dye lasers. When placed in the path of a collimated beam, they exhibit a series of closely spaced spectrally narrow passbands (Fig. 4) due to interference of multiply reected light waves passing through. The spectral separation, termed the free spectral range, and the width of the passbands are inversely proportional to the optical path length between the mirrors. Frequency tuning of the passband is easily achieved by tilting the etalon about a suitable axis. Making a judicious choice of the grating and interferometer parameters, it is possible to allow only one passband to be located at the peak of the spectral prole determined by the grating, with the other passbands lying outside the prole. Laser linewidths of
DYE LASERS
81
0.03 cm1 (0.9 GHz, / 106) in commercial systems and even single longitudinal mode operation ( 60 MHz) in laboratory systems have been demonstrated with such resonators (9). Grazing-incidence-grating (GIG) resonators (Fig. 3) make use of a single grating at a large angle of incidence. An expanded and dispersed beam diffracted by the grating is fed back by a mirror, or by another grating in Littrow conguration, for additional dispersion and narrower linewidth. Tuning is achieved by rotating the mirror or the Littrow grating. Commercial systems with linewidth of 0.05 cm1 are available, and single longitudinal modes of operation have been demonstrated. In comparison to other resonators with same linewidth, GIG resonators are less expensive, easier to align, compact, and offer advantages in terms of continuity of tuning over large-wavelength ranges (which is of particular importance in recording and interpreting spectra). The major drawback of GIG resonators is their lower efciency due to low diffraction efciency of the grating at the large angles of incidence at which they are used.
1.0
0.5
0.0 Frequency
Ultrashort pulse dye lasers are broadband lasers in which all the frequency components (longitudinal modes) making up the broad spectrum are forced to oscillate with xed phase relationships by techniques known as modelocking (10). At a point inside the laser, at some instant of time, all these oscillations arrive in phase and add up to a strong eld. Thereafter, the oscillations at different frequencies move out of phase and the resultant eld strength reduces. The larger the number of modes, the faster is the reduction in intensity. Since adjacent modes are separated by a frequency difference of c /2L, the modes again arrive in step after the cavity roundtrip time of 2L / c. Modelocking ensures that the phase relationships do not change with time. The process is treated mathematically as a Fourier sum of the spectral components. The result is a train of short-duration pulses separated by the cavity round-trip time. Each pulse has a minimum duration limited by the Fourier transform of the laser spectrum. Hence, a large bandwidth as well as a long resonator with a large number of modes (small mode separation) are both desirable. The former enables production of short-duration pulses, and the latter ensures larger separation between pulses with a cleaner baseline. Dye lasers, with broad spectral widths, have played a central role in the development of ultra-short-duration light pulses and investigation of ultrafast processes in nature. Amplication of ultrashort (100 fs) dye laser pulses in high-energy excimer laser ampliers, with prior frequency conversion when necessary, has produced few 100 fs pulses with 1 TW (terawatt) to 4 TW output power in the blue UV region (11). Such high-power lasers have opened up a new eld of high-intensity laser-matter interaction, with emerging applications in the area of laserinduced fusion technology, compact high-energy electron accelerators, sources of tunable, coherent XUV beams by high-order-harmonic generation, generation of ultrashort Xray sources, and investigation of fundamental processes in nature. PHYSICS AND TECHNOLOGY OF DYE LASERS Photophysical Properties of Dye Molecules
Transmission
(a)
1.0
0.5
0.0 Frequency (b)

Figure 4. Laser spectrum narrowing with grating and FabryPerot interferometer. (a) Solid line shows Fabry-Perot transmission as a function of frequency of transmitted beam. Dashed line shows spectral prole of laser selected by grating alone, allowing only one FabryPerot passband. (b) Expanded view of selected passband (dashed line) and longitudinal modes of laser (lled).
Laser dyes are organic molecules with a large two-dimensional planar structure that is characterized by a chain of conjugated double bonds (alternate single and double bonds), often with benzene-ring-like linkages that tend to rigidize the planar structure (12). Rigidization constrains internal vibration and rotation in the dye molecules, which act as nonradiative (i.e., nonuorescent) pathways for relaxation of the excited dye molecules. Figure 5 shows the molecular structure of the most commonly used, efcient, and stable laser dye, Rhodamine 6G. Each carbon atom is linked with the neighboring atoms with a single bond or a double bond. One of the two bonds in the double bond (referred to as the bond) is formed by a transverse overlap of the valence electron orbitals of two carbon atoms such that an electron distribution exists symmetrically above or below the center line joining the two nuclei. The other bond, referred to as the bond, is formed by overlap of two other valence orbitals along the axis. In a conjugated molecule, the electrons participating in the formation of the bond are not attached to any specic pair of carbon atoms but may shift between neighboring pairs. The electrons are thus relatively mobile. Hence an external elec-
Intensity
82
DYE LASERS
O C OC2H5 CH3 CH3 Cl H N H 3C 2 O N+ C 2H 5
the dye molecule involves raising an electron from the highest occupied (N /2)th level to the lowest unoccupied ((N /2) 1)th level. This corresponds to the longest-wavelength absorption band of the dye molecule at a wavelength, m, derived from Eq. (1): m = 8mcL2 /h(N + 1 ) (2)
Figure 5. Molecular structure of the most commonly used, efcient, and stable dye, Rhodamine 6G, illustrating the presence of conjugated double bonds.
tromagnetic eld of suitable wavelength can easily induce a strong oscillating dipole in the electron cloud. This results in the strong absorption and emission properties exhibited by dye molecules. A simple model, called the free-electron model, is often used for elucidating the photophysical properties quantitatively. In this model, the electrons are assumed to move freely in a one-dimensional box extending over the unfolded length L of the chromophore (the chain of conjugated bonds). Within the box the electrons see an averaged constant potential that rises steeply at the ends of the box. This is a standard exercise in quantum mechanics and predicts that the electrons shall occupy discrete energy levels with energy En = n2 h2 /8mL2 (1)
In the excited state the two electrons in the (N /2)th and ((N /2) 1)th level may have parallel or antiparallel spins resulting in the formation of triplet or singlet states, respectively. Figure 6 also shows the PE schematic curves for the excited singlet and triplet states useful for understanding laser action in dyes. In the electronically excited state, S1, the electron cloud is slightly expanded and exerts less binding force on the nuclei. As a result, the equilibrium separation of the nuclei corresponding to the minimum of the potential well is larger than in the ground state. The triplet states have lower energy than the corresponding excited singlet states. According to the Pauli exclusion principle, no two electrons can have the same spatial, momentum, and spin quantum numbers simultaneously. Hence, on average, wavefunctions of electrons in triplet state with parallel spins have less spatial overlap than those with antiparallel spins in the singlet state. As a result, the Coulomb repulsion component of the electron-electron interaction is less for the electrons in the triplet state, which are, therefore, slightly more bound (lower potential energy). The vibrational level spacings are 1200 cm1, which is much less than the electronic level spacings (14,000 cm1 in the visible). Rotational motion is also quantized, with level
where n is an integer, m is the electron mass, and h is the Plancks constant. In a dye molecule with N number of electrons the ground electronic state of the molecule is formed by lling up the energy levels up to the (N /2)th level, with two electrons having opposite spins in each level. For a stable dye molecule N is an even number as otherwise, even in the ground state, the molecule would have an unpaired electron and would hence be more reactive. The ground state of a laser dye molecule is therefore a singlet state with zero total electronic spin. A change in the internuclear separation from the equilibrium separation during internal vibration of the molecule leads to an increase in the electronic energy. Figure 6 schematically depicts this dependence in the form of potential energy (PE) curves as a function of the internuclear separation. Vibrational motion is also quantized and depicted as horizontal lines bounded by the PE curve, with the endpoints indicating the location of the turning points. For simple diatomic molecules the coordinate, R, is simply the separation between the nuclei. In a large dye molecule there are several modes of vibration, resulting in multidimensional PE surfaces. The representation in Fig. 6 is therefore schematic in nature, and R may be considered as a generalized vibrational coordinate. The minimum energy for exciting an electron in
Figure 6. Conceptual depiction of potential energy curves and physical processes inuencing laser action. Abbreviations and symbols are explained in text. Cross sections associated with the processes involved in the model discussed are designated by with appropriate subscripts. denotes the lifetime for spontaneous emission from the rst excited singlet state S0. TRthermal relaxation.
;; ;; ;; ; ;; ;;
S2 T2 IC TR ESA, 12 T-T Abs S1 IC TR Abs. Emission, 0 e,
ISC
Potential energy
T1
TR
S0
DYE LASERS
83
spacings typically one order of magnitude lower. Perturbation by solvent molecules further broadens these sublevels to form a quasi-continuum that gives rise to the broad absorption and emission spectra due to transition between the sublevels of different electronic states. At room temperature (kT 200 cm1) there is a Boltzmann distribution of dye molecules in the lowest vibrational-rotational levels of the ground electronic state S0. Optical transitions between singlet and triplet states have very low probability (referred to as being forbidden transitions in quantum mechanical description) compared to those between singlet-singlet and triplet-triplet states. Depending on the spectrum of the pump radiation source, optical excitation of dye molecules proceeds from the bottom of S0 to various high-lying sublevels of excited singlet states in accordance with the Franck-Condon principle (i.e., along a vertical linea result of the fact that, electronic transitions being much faster than nuclear motion, there is no change in nuclear coordinates during an electronic transition). For almost all laser dyes it has been found that nonradiative relaxation from S2 to S1 (internal conversion) in time scales 10 ps to 0.1 ps, and nonradiative vibrational relaxation (in time scales 1 ps), enhanced by thermal collision with solvent molecules, rapidly brings down the dye molecules to the lowest levels of S1. Subsequently, the molecules de-excite vertically down to the higher empty sublevels of S0 by uorescence emission at slower time scales ( few nanoseconds), followed again by rapid nonradiative relaxation down to the bottom of S0. Note that, due to the shift in the equilibrium separation in S0 and S1 the emission band is shifted considerably to longer wavelengths. Due to fast thermal relaxation in comparison to emission rates and small overlap between absorption and uorescence bands, a population inversion is easily created between the levels taking part in the emission process. Stimulated emission between these levels by a narrowband feedback in the dispersive laser resonator and homogeneous broadening caused by the fast thermal relaxation generate the spectrally condensed narrowband laser output. Fast relaxation (in comparison to excitation rates) to lowest levels of S1 also allows excitation of all the dye molecules to S1. This conguration of pumping levels is referred to as a four-level laser scheme, where the population dynamics allow laser action to be achieved at low threshold pump intensity and facilitate CW laser action by recycling the molecules through the different steps. The parameters that are used to describe the aforementioned processes (Fig. 6) are the transition cross sections (), the quantum yield of uorescence Qf , and the radiative lifetime of the S1 state. Qf is dened as the ratio of the number of photons emitted to that absorbed and may be expressed in terms of dye parameters as / r, where is the total lifetime of S1 and r is the radiative lifetime. For an efcient laser dye, Qf should be close to 1. (e.g., 0.93 for Rhodamine 6G). An important nonradiative de-excitation process that reduces Qf is intersystem crossing (ISC) of dye molecules from singlet to triplet states. Because the triplet to singlet radiative transitions are forbidden, the lowest triplet state is a metastable state that relaxes slowly to the ground state (T 10 s to 100 s). Thus, ISC removes dye molecules from the lasing cycle, accelerates photochemical degradation of dyes (since molecules with unpaired electrons are more reactive), and, more important, introduces a loss for laser photons by T1 T2 absorption, which may have a substantial overlap with the
emission spectrum (Fig. 1). Due to the relatively slow ISC rate (3.4 106 s1 for Rh6G), ISC starts interfering only for long pulse (ashlamp pumped) or CW dye lasers in which rapid ow of dye solution and presence of triplet quenchers (e.g., oxygen or cyclo-octatetrene dissolved in the solvent) help in reducing the adverse effects. Other processes that inuence laser action are groundstate absorption (GSA from S0 to S1) at the laser wavelength, l, and excited-state absorption (ESA from S1 to S2) at l and at pump wavelength, p. The strength of GSA at l depends on the tail of the absorption band extending into the emission band. Although weak, the large concentration of dye molecules in the ground state makes its inuence signicant. GSA is responsible for a red shift of the peak of the gain spectrum from the peak of the uorescence spectrum. The shift increases with increase in dye concentration and with decrease in pump power. Although a wide variety of pump sources with different wavelengths or with broad spectra may be used for pumping the same dye, use of pump sources with p close to the longest-wavelength absorption band reduces loss of photon energy (hp hl) by internal conversion and vibrational relaxation. Nonradiative relaxation of excitation results in heating of the laser active region and generates refractive index gradients and inhomogeneities, which, in turn, leads to distortions in the laser spectrum and beam prole. ESA at either p or l results in a loss for pump or laser photons, which also heats up the medium because of subsequent nonradiative decay to S1. Theoretical Model A rudimentary model of a transversely pumped dye laser consists of a cylindrical region of dye solution of length L with a concentration of N molecules/cc, bounded by two parallel mirrors of reectivity R1 and R2, and pumped uniformly with pump intensity Ip averaged over the absorption depth of the pump radiation in the active medium. The model also assumes that the triplet population is rendered negligible by quenching and removal of dye molecules in the triplet state by owing. This simplication is more appropriate for shortduration ( few tens of nanoseconds) lasers because of the slow rate of ISC. Due to rapid internal conversion, populations in higher excited states are also neglected. Then distributions of molecules in S0 and S1 are adequate for describing the process through the population densities N0 and N1, such that N N0 N1. The incremental amplication of a light beam of intensity I (photons cm2 s1) propagating along z direction is expressed as a differential equation: dI = e IN1 0 IN0 12 IN1 dz (3)
where the cross section (as shown in Fig. 6) are taken at the laser wavelength. The different terms describe, in the same sequence, stimulated emission from S1, GSA from S0, and ESA from S1. The population densities are determined by the rate equation dN1 N = 0p Ip N0 e IN1 + 0 IN0 1 dt (4)
Here the last term describes the spontaneous decay of population from S1. For most practical systems with pulse duration
84
DYE LASERS
large compared to , the time variations of the laser intensity are slow compared to the spontaneous or stimulated processes and a steady-state approximation is justied. Under this condition, Eq. (3) may be expressed as (13) 1 dI g0 SI = I dz 1 + I/IS 1 + I/IS Here, g0 eff N (0pIp 0 /eff ) 0p Ip + 1 (5)
spectroscopic investigations. The intracavity laser beam at wavelength incident on the grating at an angle is diffracted at angle according to the equation (sin + sin ) = m (6)
where a is the groove spacing and m 0, 1, 2, . . . is the integer order of diffraction. In the Littrow conguration ( ), commonly used in the Ha nsch resonator, one has 2a sin = m (7)
with eff e 12, is the small-signal gain coefcient that describes the single-passage amplication of a weak light beam (I IS) incident along the z direction according to I(z) I(0) exp(g0z). Moreover, IS 0p Ip + 1 (e + 0 )
is the saturation intensity that determines the extent of gain saturation by the laser, as is clear from the rst term in Eq. (5). The increase of IS with pump intensity represents the fact that at higher pump intensities the molecules brought down to S0 will be recycled back to S1 at a faster rate; a higher laser intensity is then required to saturate the gain. Finally, the last term in Eq. (5), with N S 0 12 0p Ip + 1 describes a nonlinear loss arising out of absorption of two laser photons by molecules from S0 and subsequently from S1 (13). This nonlinear loss due to ESA turns out to be important at shorter wavelengths in the gain spectrum where the absorption from S0 and hence 0 increases. When the round-trip gain in the cavity is sufcient to compensate for cavity losses arising out of mirror transmissions, absorption, scattering, diffraction at apertures, and losses at other optical elements inside the cavity, laser oscillation sets in and an intense directed laser output is obtained. This threshold gain condition, where small signal gain approximation is not unjustied, is expressed as R1R2 exp(2g0L) 1, with all the losses clubbed into mirror reectivities. Typical threshold pump intensities predicted by these equations are few tens to hundreds of kW/cm2 in agreement with experimental results. The equations just described serve as a starting model for description of the dye laser. Additional equations describing the absorption of pump radiation, especially in longitudinally pumped lasers or ampliers, evolution of broadband ASE, and triplet state effects, are necessary for more rigorous treatments and are best solved by numerical methods (4). Approximate analytical expressions that are adequate for preliminary analysis of dye oscillators and ampliers are also available in the literature (2,1315). DESIGN CONSIDERATIONS Pulsed Laser Pumped Narrowband Dye Lasers Figure 3 shows the two congurations of pulsed narrowband dye lasers commonly used today for linear and nonlinear
For example, using a grating with 600 grooves/mm at 45, the Littrow condition will be satised for a wavelength of 3.33 m in rst-order diffraction, for 1.667 m in second order, for 666.7 nm in fth order, and so on. Thus, the same grating may be used for covering the tuning range of all dyes by using the diffraction at different orders. Currently, gratings fabricated by holographic techniques are commonly used because of the absence of ruling errors present in replica gratings made from mechanically ruled master gratings. In the absence of intracavity beam expanders, the divergence, , of the beam incident on the grating is / w, which is largetypically a few milliradians, mainly due to diffraction at the narrow gain region (diameter w). The dispersion of the grating, derived from Eqs. (6) and (7), is given by d = d 2 tan (8)
and determines a passive bandwidth for the resonator given by G = 2 tan (9)
The laser bandwidth is somewhat reduced by gain narrowing and repeated dispersion at the dispersive elements due to a number of round trips in the cavity during the existence of gain. As discussed earlier, the gain narrowing process in these pulsed dye lasers is limited by the short pulse duration. A prism, used at a large angle of incidence and at near normal incidence at the exit face (Fig. 3), expands the beam by a factor MP cos /cos , where is the angle of refraction inside the prism. The divergence of the beam, in the plane of grating diffraction, and hence the passive bandwidth is thereby reduced by the same factor. However, the reectivity loss at the entrance face of the prism increases rapidly with as it approaches 90. Hence, multiple prism beam expanders are used to obtain large magnications with reduced loss. For identical prisms, the overall loss is minimum when the angles of incidence on all the prisms are same. In commercial narrowband pulsed dye lasers that use such Littrow grating and about 40 four-prism beam expanders, output laser bandwidths of 0.3 cm1 to 0.6 cm1 (9 GHz to 18 GHz) are available. Duarte (16) provides a detailed discussion on design considerations for prism beam expanders in narrow linewidth dye laser systems. This linewidth is still too large compared with linewidths of atomic and molecular transitions in vapor samples, especially when such samples are available in the form of collimated beams. In vapor samples, the frequency or wavelength of the directed laser source as seen by the randomly moving atoms
DYE LASERS
85
is Doppler shifted to different extents depending on the distribution of velocity components in that direction. The absorption or excitation spectrum is then Doppler broadened, with the broadening increasing with vapor temperature and decreasing with the mass of the particles. In high-resolution spectroscopy, Doppler broadening is reduced to a less than a few gigahertz by steering the probing laser in a direction perpendicular to a well-collimated atomic or molecular beam. As mentioned earlier, to reduce the laser bandwidth further, intracavity FabryPerot etalons (FPE) are used. The FPE passbands are centered at wavelengths given by 2t cos = n (10)
where is the angle of refraction inside the etalon and n is the integer-order of the passband. The free spectral range (FSR) of the etalon is given by (2t)1 (in wave numbers, cm1) or by 2 /(2t) in wavelength units. The sharpness, or the FWHM, of the passband is related to the etalon parameters by F FSR / F, where the nesse, F, is determined (1) by the reectivity, R, of the coated surfaces, FR R /(1 R); and (2) by the atness and parallelism of the surfaces in the illuminated region, FF M /2, where / M is a measure of the deviation from atness and parallelism across the illumi2 nated region. The overall nesse is given by F 2 F R 2 F F . In practice, using the equations given previously, designers of narrowband dye lasers select grating, beam expander, and etalon parameters such that etalon FSR is slightly larger than the spectral width determined by the grating and beam expander. Other considerations that play important roles in laser design are (1) maximum transmission of etalon, given by Tm (1 R A)2 /(1 R)2 (where A stands for the fractional absorption and scattering losses introduced by the etalon and its surface coatings), which increases rapidly with reectivity R and thus reduces the laser efciency drastically when attempts are made to decrease laser linewidth by increasing the reectivity nesse; and (2) the divergence, , of the intracavity beam transmitted through the etalon, which contributes to an increase of the etalon passband width, given by / 2, and which increases as the etalon is tilted with respect to the beam for tuning the laser wavelength. Further, as the etalon is tilted, spatial overlap between successive reected waves from the etalon surfaces decreases and the multiple beam interference process is impeded. This leads to a reduction in nesse and peak transmission. The resultant effect of these factors is to increase laser bandwidth and reduce its efciency. To minimize these adverse effects, the etalon is placed in the expanded part of the intracavity beam and tilted in the plane of the expanded beam. Another complication arising out of using an etalon is that, due to the very different dependence of the center wavelengths selected by the grating and etalon on the respective tilt angles, specially designed tilting mechanisms for both are needed such that the center wavelengths are tracked together during wavelength tuning. Degradation of the etalon nesse restricts such synchronized tuning ranges to 1 nm. After completing one such synchronized scan, the grating is held xed while the etalon is tilted back to the starting orientation such that a suitable passband is again centered precisely on the wavelength peak currently selected by the grating. Working systems almost invariably employ computerized control
for such piecewise continuous wavelength scan using mechanical tilting operations. In some systems, sealed pressure chambers are constructed around the grating and an airspaced etalon combination with a suitable window for beam entry and exit, to enable synchronized wavelength scanning by changing the pressure of a suitable gas (nitrogen, sulfur hexauoride, etc.) inside the chamber. Changing the pressure changes the refractive index, gas, of the gas and hence the wavelength according to air airgas / gas (6). The actual laser spectrum, as shown in Fig. 4, consists of the longitudinal modes of the resonator separated in frequency space by c /2L (or 2 /2L in wavelength units), where L is the optical length of the cavity. The frequency of the modes is extremely sensitive to external inuences. A change in the cavity length by a small as /2 causes the frequency of a mode to be changed by the separation between the modes themselves. While tuning the wavelength of the laser one actually scans the envelope of the etalon-selected spectral prole over the modes, which may themselves also shift due to change in L during tilting of optical elements. For common applications it is preferable to pack several modes inside the spectral prole by choosing an appropriate cavity length. This is necessary because the intensity of individual modes in such multilongitudinal-mode lasers has been found to uctuate considerably from pulse to pulse, often with only small subsets of all the longitudinal modes oscillating in individual pulses. Such uctuations do not show up in time-averaged laser spectra but manifest themselves as randomly uctuating beat modulations within the temporal prole of successive pulses recorded with a large-bandwidth oscilloscope. When only a few (say, two or three) modes are allowed to oscillate, such uctuations lead to considerable uctuations in the output pulse energy. In several applications it is also necessary to maintain a good overlap of the laser spectrum with the absorption spectrum. Fluctuation in mode content of the laser leads to variation in spectral overlap with a narrow atomic transition, or inadequate overlap with an inhomogeneously broadened transition, when the absorption width is similar to, or large compared to, the mode separation, respectively. A better option, although technologically more complex, is to allow laser oscillations in only a single longitudinal mode by designing a compact cavity with sufcient dispersion so that only one mode lies within the allowed spectral prole. The minimum linewidth of such single-longitudinal-mode dye lasers is dictated by the Fourier transform of the pulse shape. Frequency tuning of these lasers turns out to be technologically more complex than for multilongitudinal-mode narrowbandwidth lasers. The frequency of the mode is tuned by changing the cavity length, with the required submicron precision provided by a piezoelectric transducer that translates a cavity mirror. Simultaneously, the center frequency selected by the intracavity etalon is forced to follow the mode frequency by active servo control. Otherwise, as the etalonselected prole strays away from the chosen mode and verges on an adjustment mode, mode hopping takes place, with the output frequency switching discontinuously to the latter. Tracking is achieved in very high repetition rate pulsed lasers by introducing a small oscillation in the tilt of the etalon, driven by a piezoelectric actuator (9). The dither in the etalon spectral prole modulates the average output power. The amplitude of modulation increases with frequency detuning between the etalon prole peak and the mode, and its phase
86
DYE LASERS
with respect to the drive signal depends on the sign of detuning. Phase-sensitive measurement of the average laser power with the drive signal as a reference provides an error signal that is amplied and fed back to the etalon piezo in the correct phase such that the error signal (and hence detuning) is driven to zero. Although the pressure-scanning technique may be employed as a simpler alternative, frequency controlsuch as precise tuning to a prespecied frequency or stabilizing the laser frequency by servo controlis very slow in comparison. The GIG conguration, described earlier, features easy and continuous wavelength tuning by tilting a single element (i.e., the tuning mirror or another grating in Littrow orientation). To achieve laser bandwidths (1 GHz) comparable to that with Littrow grating etalon conguration, the grating in the GIG resonator is used at large angles of incidence close to 89, where the grating efciency is very low (10%). With the incorporation of prism pre-expanders that allow the use of the grating at smaller angles of incidence, laser efciency comparable to that of the Ha nsch conguration is now possible. Although single-mode operation of GIG resonator and mode-hop free tuning of the output frequency over large ranges (100 GHz) have been demonstrated in specially designed laboratory systems (17), the need to design a compact cavity precludes the use of prism expanders. As a result, the efciency of such lasers is limited to about 1%. CW Narrowband Dye Lasers Design considerations for CW dye lasers (1,2) differ considerably from those for pulsed dye lasers. Most commonly used CW pump sources are the argon ion and krypton ion lasers, with output powers of 5 W to 25 W distributed over a number of discrete lines in the blue-green (Ar laser) and red (Kr laser) wavelength range. Because of the low output power of CW pump lasers, they are focused tightly (20 m to 50 m diameter) into the dye solution to generate sufcient excitation intensities ( MW/cm2) required for pumping CW dye lasers. This tiny excitation region is located within a small, optically at area of a thin stream (0.1 mm 3 mm) of slightly viscous dye solution that is ejected as a free jet from a at slotted nozzle at velocities of several meters per second. An open tube catches the jet and returns the dye solution to a pump that keeps the dye owing. The choice of the pump and ow system, consisting of a lter that removes insoluble contaminants, bubbles, and also acts as a buffer to reduce ow uctuation, is important in narrowband dye laser systems, where uctuation in the jet thickness or the presence of bubbles lead to wavelength and power uctuations. Commonly, positive displacement pumps such as gear pumps are used with magnetically coupled drives to prevent contamination of solution and minimize re hazard arising from leakage of inammable solvent vapor. The use of free jets, instead of dye cells through which the solution is own, was motivated by the need to avoid dye cell damage caused by the high focused intensity of the pump laser, especially at the low velocity boundary layers near the cell walls. Figure 7 shows the schematic of a three-mirror folded resonator in a near-collinear longitudinal pumping geometry used for CW dye lasers. This geometry was a consequence of the need to focus the dye laser radiation at the dye jet for saturating the gain and for maximizing overlap with the gain region.
Pump
;
Jet Pump
Piezo actuator Thin etalon BRF C
Thick etalon (a)
Faraday rotator
Polarization rot.
M (b)
Figure 7. Schematic diagrams of CW dye laser illustrating (a) standing wave resonator with spectral ltering and tuning elements; (b) unidirectional ring resonator (lter and other optical elements for astigmatism compensation and cavity length change are not shown). Coutput coupler. Mmirror. BRFbirefringent lter.
The resonator consists of a short arm in which the dye laser beam is focused and a long arm in which the beam is collimated and coupled out through a at mirror. Optical elements for spectral narrowing and wavelength tuning are placed in the long arm. The choice of these dispersive elements is dictated mainly by the need to minimize the losses inside the cavity so that reasonable efciency and large tuning ranges are obtained. For coarse spectral narrowing and wavelength tuning, a birefringent lter (BRF) plate is used. The BRF is oriented at Brewsters angle with respect to the dye laser beam in the long arm of the cavity such that the reectivity loss of the electric eld vector in the plane of incidence vanishes. It consists of crystalline quartz plates with an axis of symmetry, called the optic axis, lying in the plane of the plates and oriented at 45 with respect to the preferred polarization of the electric eld vector. Inside the plate, the electric eld components of the light wave, polarized along (ordinary) and orthogonal (extraordinary) to the optic axis, travel with different velocities (birefringence) and recombine at the exit face. This results in formation of linearly polarized light in the plane of incidence only for specic wavelengths, at which the phase difference between the ordinary and extraordinary waves is an integral multiple of . These wavelengths determine the center of the passbands of the BRF at which the losses are minimum. At other wavelengths, components of electric eld that are orthogonal to the plane of incidence exist and suffer reective losses at the surfaces of the BRF. A stack of BRF plates with increasing thickness is used. Only one selected passband of the thinnest BRF lies within the dye gain prole, while the other plates produce progressively narrower passbands (albeit with closer separation) such that a passband of each plate is exactly centered on the main passband. This combination results in the lowest loss in
DYE LASERS
87
a narrow wavelength range (FWHM few tens of cm1) centered on the main passband and higher losses at all other passbands of the thicker plates that lie within the dye gain prole. For wavelength tuning, the BRF stack is rotated in its own plane such that the wavelengths of all the centered passbands track together. Typical laser linewidths of less than 0.1 cm1 containing a large number of longitudinal modes are obtained with the BRF. For restricting the laser oscillation to a single mode, a pair of low-loss and hence lownesse Fabry-Perot etalons are used. For obtaining the same effective FSR and passband width as that of two etalons in tandem with only one etalon, its surface reectivities must be increased that would result in higher losses. As a result of unrestricted spectral narrowing, in sharp contrast to pulsed laser pumped dye lasers, less than 1% extra loss for the unwanted modes just away from the peak of the thick-etalon transmission prole is sufcient to restrict lasing in the selected mode located at the peak. Such survival of the ttest mode is, however, interrupted by the formation of standing wave patterns due to superposition of the counterpropagating light waves in the linear cavity. The standing waves formed by the selected mode at wavelength have high intensities at antinodes and low intensities at nodes separated spatially by /4. Consequently, the gain is actually saturated only close to the antinodes in the gain mediuma process aptly referred to as spatial hole burning. A second mode in the neighborhood, in spite of a slightly higher loss, may have antinodes at the location of the nodes of the rst mode. Spatial hole burning results in unwanted multiple frequency operation, especially at high pump laser powers. In addition, incomplete utilization of the gain by the selected longitudinal mode restricts the laser efciency in single-mode operation. To overcome the adverse effects of spatial hole burning, current CW dye lasers are also congured as a traveling-wave ring resonator in which a small relative loss (0.5%) is introduced for one of the two possible counterpropagating waves. This is achieved by using the Faraday effect in a piece of crystal or glass placed in a magnetic eld applied parallel to the direction of the propagating light wave. The Faraday effect rotates the plane of polarization of light in a direction independent of the direction of propagation through the material. A polarization rotator (a birefringent crystal plate of appropriate thickness) placed on one side of the material reverses the rotation of the emerging light. The counterpropagating light has its polarization rotated by both the rotator and Faraday effect in the same direction, and therefore suffers higher loss at the several Brewster angle surfaces in the cavity. The small differential loss again sufces for producing unidirectional traveling-wave operation because of continued saturation of gain by the wave in the favored direction. For efcient dyes, unidirectional ring lasers produce almost twice the single-mode output power as that from linear cavity lasers. However, with low gain dyes the increase in threshold due to the insertion loss of the unidirectional device results in lower efciency than in the linear cavity. The capability of the CW single-frequency dye laser in ultra-high-resolution spectroscopy is essentially dependent on the monochromaticity (i.e., the time-averaged linewidth and its tunability). The linewidth is determined by uctuations in the cavity length caused by vibration, uctuations in the dye jet thickness, ambient air pressure uctuations, and drifts in
the cavity length caused by thermal expansion. With standard procedures like vibration isolation, thermally stable construction and environment, and suitable enclosures, effective time-averaged linewidth of several megahertz is achieved. Various methods of active stabilization techniques have been described in the literature. Most of these make use of the transmission of a small part of the laser output through a highly stabilized, high-nesse (narrow passband) tunable Fabry-Perot interferometer. Transmission at the falling (or rising) edge of the interferometer changes with change in laser frequency and produces the requisite frequency discrimination error signal for feedback control of the cavity length of the laser through a piezoelectric transducer. Commercial stabilized lasers provide effective short-term linewidth of 1 MHz (about 2 parts in 109). The narrowest linewidth of less than 50 Hz (1 part in 1013) has been demonstrated by using sophisticated heterodyne techniques and fast electro-optic modulators that change the cavity length by varying the refractive index of an intracavity crystal by changing an applied electric eld. Tuning the frequency of the laser requires synchronized scanning of the cavity length along with the passbands of the multiple frequency-selective devices. Commercial lasers employ complex electronically controlled drive mechanisms, such as rotating galvo-drives (for the thin etalon and for a nearBrewster plate for cavity length change) and piezoelectric servo-tracking the thick-etalon passband to the scanning cavity mode. This provides a continuous tuning range of 30 GHz, after which the devices are reset to the initial status, the BRF is stepped to the center of the next 30 GHz range, and the scanning continues. Computer-controlled software and data acquisition can stitch together such piecewise continuous scans to produce an effectively continuous scan over the 10 THz range. Pulsed Dye Laser Ampliers Several applications of narrowband dye lasers, like investigation of nonlinear optical effects, remote sensing of pollutants in the environment by laser-induced uorescence, and ultrasensitive detection of trace elements with isotopic selectivity, requires moderate to high laser powers. It is simpler to generate high instantaneous powers in pulsed operation. Amplication of narrowband dye laser oscillator outputs in pulsed laser pumped dye laser ampliers is a standard technique used for this purpose. Such master-oscillator-power-amplier (MOPA) concepts resulted from the realization that attempts to obtain high power by pumping narrowband oscillators far above threshold led to an increase of bandwidth and broadband ASE and also generated spurious unwanted spectral components in the laser output. The solution is to constrain the master oscillator operation to produce a low-power, highquality output that is amplied in one or more dye amplier stages. The ampliers usually have a similar transverse pumping conguration that offers design exibility, but longitudinally pumped ampliers are also used when beam quality is a prime concern. Frequently, a small fraction (10%) of the pulsed pump laser is split out, using coated glass plates to pump the oscillator, and the rest is distributed optimally between the following amplier stages with an optical path delay for synchronization of the oscillator pulse and the pump laser pulse at each stage. The delay is introduced to avoid
88
DYE LASERS
amplication of the ASE that exists at the beginning of the oscillator output. The primary design consideration for pulsed ampliers is to saturate the gain of the amplier by a strong signal (oscillator output). Ineffective saturation leads to reduced efciency and the presence of undesired broadband ASE in the amplier output coming from single-pass amplication of the spontaneously emitted radiation within the amplier itself. In other words, the signal should extract the gain in the amplier efciently without allowing any spontaneous emission to compete for the gain. High-power dye laser systems are therefore congured in a MOPA chain, such that each amplier boosts the signal sufciently to saturate the gain in the following higher-power amplier. Excessively high signal intensities are, however, detrimental due to increasing nonlinear absorption losses arising from ESA. Integrating Eq. (5) with appropriate initial conditions for an amplier, one nds that the extraction efciency of the amplier exhibits a maximum at an optimum signal intensity (13,14). Extensive recent reviews on capability of high-power dye laser oscillator-amplier systems may be found in Ref. 18. Extending the Wavelength Range of Dye Lasers The short wavelength limit (310 nm) of dye laser operation arises from photodissociation of dye molecules when the excitation energy due to absorption of UV photons becomes comparable to the bond energy. The long wavelength limit (1.7 m) is determined by the short operation life of the molecules because of enhanced chemical reactivity in triplet states to which the molecules are thermally excited, because the triplet states are closer to the ground state for infrared dyes. The fundamental tuning range of abut 310 nm to 900 nm is extended into the VUV and IR regions by nonlinear optical processes, such as frequency doubling and mixing in crystals (19). At the high intensity available with lasers, the oscillating polarization induced in the material shows signicant nonlinear dependence on the electric eld. If two lasers with electric eld components E1 cos 1t and E2 cos 2t and total eld E E1 cos 1t E2 cos 2t are incident, then the rst nonlinear term (proportional to E2) in a power series expansion of the polarization in terms of the total eld contains oscillating components at frequencies 21, 22, and 1 2. These components radiate, respectively, at the second harmonics and the sum or difference frequencies. Since there is no net energy or momentum transfer to the medium, the process becomes signicant only when the momentum conservation condition for the photons participating in the process is satised. The momentum of a photon propagating in a transparent medium with refractive index is given by k, where k is the propagation wave vector of magnitude k 2 / / c, in the direction of propagation of light. The momentum conservation condition (also referred to as the phase matching condition to signify that the two waves propagate with the same phase velocity when they are colinear) for frequency doubling is given by k(2) k() k(). Although, as expressed here, the fundamental photons combining to form one second harmonic photon may, in principle, propagate in different directions, they should be colinear for efcient frequency conversion. The colinear phase matching condition cannot be satised in isotropic media due to normal dispersion (monotonic dependence of refractive index on wave-
length) at wavelengths away from absorption features in the medium. Fortunately, birefringent crystals such as KDP, KTP, ATP, LiNbO3, and, more recently, -BaB2O4 (BBO) and LiB3O5 (LBO), with a high nonlinear response and high damage threshold, have been developed in which efcient frequency conversion is possible. In these crystals, the electromagnetic wave with polarization components parallel (ordinary component) and perpendicular (extraordinary component) to a symmetry axis (optic axis) travel with different velocities. In other words, the refractive indices for the two components are different. In addition, while the ordinary refractive index does not depend on the direction of propagation with respect to the optic axis, the extraordinary refractive index does. As a result, it is possible to nd a propagation direction in the crystal such that the phase matching condition is satised. Tuning of the second harmonic output is achieved by simultaneously tilting the crystal to vary the angle between the incident beam and the optic axis as the fundamental frequency is being tuned. This requires precise angle tilting drives for a few components and is best done under computer or servo control. With frequency doubling and sum frequency mixing, coherent radiation in the wavelength range of about 200 nm to 500 nm has been generated. On the other hand, with difference frequency mixing in LiNbO3 and AgGaAs2, tunable IR radiation over 2.2 m to 4.2 m and 4 m to 9 m, respectively, has been generated. Ultrashort Pulse Dye Lasers Modelocked, ultrashort pulsed dye lasers are essentially broadband CW lasers in which the longitudinal modes are forced to oscillate in phase at precisely dened frequency separation of c /2L between adjacent modes. In ordinary broadband lasers the longitudinal modes corresponding to the different transverse modes oscillate at different frequencies. Even when forced to operate in the lowest order (Gaussian) transverse mode by placing a small aperture on the axis (which is provided by the gain medium itself), the frequency separation varies slightly from the nominal value of c /2L due to minute pulling of the longitudinal modes toward the center of the gain spectral prole. The relative phases and amplitude of the modes uctuate in time due to random spontaneous emission within these modes. As a result, the output of broadband CW dye lasers shows transient spikes and uctuations over a continuous emission level. Two different techniques for modelocking are commonly employed (10). In passive modelocking, a CW laser is used for pumping the dye laser with a folded, three-mirror, linear resonator conguration similar to that for narrowband CW dye lasers. A saturable absorber dye jet is introduced at a second focus inside the cavity constructed by incorporating two additional curved mirrors. The saturable dye displays increasing transmission with incident intensity due to saturation of the transition (equilization of population in ground and excited states so that there is no net absorption). The modelocking process may be visualized in the spectral domain as follows. When two adjacent modes start growing, the absorber sees an incident radiation with its intensity modulated at the beat frequency. The transmission of the absorber and hence the loss in the resonator is also modulated at the same frequency. Such forced amplitude modulation of the radiation generates sidebands precisely separated by the beat fre-
DYE LASERS
89
Prism

Absorber
ing. Subsequently, pairs of gratings and prisms producing an opposite chirp compress the pulse to shorter duration. Measurement of the duration of such short pulses is done by employing nonlinear optical techniques to determine the spatial overlap of two components of the same pulse as a function of a precisely variable path delay between the two. The nonlinear interaction (such as frequency doubling in crystals) is congured specially so that a signal is detected only when both pulses are overlapping. The resulting autocorrelation trace is deconvoluted, assuming trial pulse shapes, to determine the spatial length of the pulse and hence its duration. APPLICATION OF DYE LASERS The majority of applications of dye lasers make use of their diverse properties, such as (1) tunability of the laser, for selective interaction with a particular species; (2) high brightness, for increasing sensitivity and enhancing the response of the application process; (3) focusability, for spatial resolution; and (4) pulse compressibility, for temporal resolution. A few current representative examples are given in this section to illustrate the scope of applications of dye lasers in basic sciences, in industry as diagnostic tools or as process drivers, in environmental monitoring, and in health care. More information may be found in Refs. 19, 2224. Laser-Induced Manipulation of Atoms Dye lasers, and later tunable diode lasers, have been used for cooling and trapping atoms to temperatures less than 20 nK and densities 1011 /mL (25,26). At these ultra low temperatures the quantum nature of the atomic motion is manifested through the wavelike behavior of atoms, described by the thermal de Broglie wavelength DB / p, where p is the momentum of the atom. When the average interparticle separation becomes comparable or less than DB, atoms that behave as bosons gather in the lowest-energy quantum state, resulting in BoseEinstein condensation (BEC). BEC is a central feature in understanding the collective behavior of particles, such as superuidity and superconductivity. Tunable lasers have made it possible, for the rst time, to produce and demonstrate BEC in a collection of atoms that are weakly interacting unlike superuidity in strongly interacting liquid He4. The laser technique that led to this fundamental application was recognized by awarding the 1997 Nobel Prize in physics to some of the scientists who had pioneered the eld (S. Chu, W. D. Phillips, and C. Cohen-Tannoudji). The basic light-induced process that forces atoms to slow down is the momentum transfer, (h / c), to a moving atom, every time it absorbs a counterpropagating photon from a source tuned to a resonance transition from the ground state. The randomly directed atomic recoil due to spontaneous emission in between successive excitations is canceled out over many absorptionemission cycles because spontaneous emission can be in any direction. For instance, Na atoms moving at average velocity of 900 m/s from an oven heated to 600C would slow down by about 3 cm/s per absorption of a photon (on an average) from a counterpropagating light beam tuned to the 589 nm transition. About 30,000 scatterings would be required to stop the atoms, and the high brightness of the laser helps in achieving this in a short time. To compensate for the chang-
Gain Pump
Figure 8. Schematic diagram of a colliding-pulse modelocked (CPM) ultrashort dye laser with intracavity compensation of group-velocity dispersion by prism pairs.
quency and in phase with the driving elds. The coherent sidebands force the evolution of neighboring modes in phase, separated exactly by c /2L, and thus initiate modelocking. With increasing numbers of modes evolving in phase, the pulses become sharper and more intense, leading to more effective saturation (and hence lower loss) of the absorber. Irregular low-intensity uctuations that see a high loss are suppressed. While simple, passively modelocked dye lasers have led to the generation of picosecond pulses, the shortest pulse from a dye laser has been produced in the now famous colliding-pulse modelocked (CPM) ring laser (Fig. 8) (20). In the CPM laser the absorber jet and the gain jet are placed with a separation of one-quarter of the path length around the ring cavity. This is done to ensure that the most favored pulse formation will correspond to counterpropagating (clockwise and anticlockwise) pulses colliding in the absorber jet and arriving at the gain jet with maximum possible time separation (half the round-trip time) such that each pulse of this pair sees the gain recovering to the same value after saturation by the other. These pulses, therefore, have equal amplitudes that result in high-contrast interference when they collide at the absorber jet. This results in more effective saturation by pulses that collide exactly at the absorber. Pulse lengths of 90 fs have been demonstrated from CPM ring dye lasers. Further reduction in pulse duration (30 fs) is obtained by using one or two pairs of prisms inside the cavity, with orientation and separation chosen to compensate for dispersion in the cavity. Material dispersion (dependence of refractive index on wavelength) in the gain jet and the absorber jet leads to variation of group velocity of the pulse with the wavelength that broadens the short pulses. The shortestduration pulses (6 fs) were obtained by further compression outside the cavity (21). Pulses with 50 fs duration were amplied and transmitted through a short length ( mm) of single-mode silica ber. Due to nonlinear optical effects at the high intensity in the ber, the refractive index of the ber material changes during the pulse. As a result, the velocity of light in the ber and the phase of the light transmitted are also modulated. The process, referred to as self-phase modulation, leads to a time-dependent spectral broadening, or chirp-
90
DYE LASERS
ing Doppler shift of the laser frequency as seen by the decelerating atoms, either the laser frequency is slowly chirped or the atomic transition frequency is tuned by using Zeeman effect in a spatially varying magnetic eld. Using such lightinduced tunable forces at the intersection of three orthogonal pairs of counterpropagating beams, a slowed-down beam of Na atoms was cooled to a temperature estimated to be 240 K. Tunability of the laser again plays a crucial role in the cooling process, for which the laser frequency is tuned slightly below resonance (red detuned). Atoms moving in any direction see counterpropagating light Doppler shifted closer to resonance, whereas the co-propagating light is shifted away from resonance. Thus, the atoms absorb more counterpropagating photons per second, resulting in a frictional force that is proportional to the Doppler shift and hence to the velocity. As a result, in addition to slowing down, the velocity spread is also reduced, and a cold ensemble of atoms remain conned in the viscous optical molasses eld produced by the laser beams. The atoms, however, cannot be trapped by these momentum transfer forces alone. Lower temperatures and BEC were achieved by innovative cooling techniques and the use of atomic traps constructed with specially shaped magnetic elds and tunable lasers. One compelling signature of BEC was the coherent wavelike motion of the atoms, which manifested itself as an interference pattern between two overlapping condensates released from the trap. Absorption of a laser tuned to the atomic transition was used for observing the pattern. Laser-cooled and trapped atoms are ideal candidates for precision measurement of frequency because Doppler shift and collision-induced errors are reduced and long measurement times are possible. It is anticipated that the accuracy of the frequency standard, presently based on the Cs atomic clock, may be increased from one part in 109 to 1015. Manipulation and focusing of atomic beams using tunable lasers have important applications, such as depositing sub-micronsize structures on surfacesa technique referred to as atom lithography. In an electric eld E, particles with polarizability have a potential energy E2. For a laser tuned below (red), the transition from the ground state of the atom is positive. The atom is therefore attracted toward the focus of a nonuniform laser beam by this tunable optical dipole force. The standing wave pattern of counterpropagating laser elds blue detuned from the atomic transition forces cooled atoms in an orthogonal beam to collect at the nodes. By placing a silicon substrate close to the standing wave pattern, parallel line structures of 100 nm width and half-wavelength separation have been deposited (19). Diagnostic Applications Several diagnostic applications of dye lasers in science, industry, environmental monitoring, and health care can be found in Refs. 19, 2224 and other references cited. Laser-induced uorescence (LIF) is one of the most widely used ultrasensitive techniques for detection and estimation of atoms, ions, or uorescent molecules in vapor or condensed phases. Use of pulsed dye lasers offer the advantage of threefold selectivityselectivity of preferentially exciting a particular species among others; selectivity of the uorescence spectrum that carries the signature of the species and its concentration; and time-resolved measurement to eliminate
interfering uorescence spectra or Raman scattering from the ambient material. In the microelectronics industry, LIF has been used for monitoring the distribution and temperature of gas phase reactants, in processes such as chemical and physical vapor deposition, plasma etching, and chemical etching during semiconductor processing. Measurement of uorescence lifetime helps in assessing the sample maturity of crude oil and coal. Time-resolved LIF is being used as a diagnostic tool for ultrasensitive detection and analysis of trace quantities of uorescent actinides and lanthanides in solution, which are commonly encountered in the nuclear fuel cycle (27). For example, uranium as the uranyl ion (UO2 2 ) and other similar ionic forms of Cm, Am, Eu, Tb, Gd, and so on are uorescent in solutions. Strong and broadband prompt uorescence from organic chemicals used in spent fuel reprocessing, or present in ground water, severely restricts detection and reliable measurement of these uorescent species. The delayed uorescence spectrum suppresses these interferences and makes it possible to detect these elements with ppb (106 g/L) to ppt (109 g/L) level sensitivity. The rapidity, selectivity, and sensitivity of these diagnostic techniques are acquiring importance in nuclear waste management for monitoring potential environmental contamination, in monitoring spent fuel reprocessing, and in the health care of uranium mine workers. Laser-induced photoacoustic spectroscopy (LIPAS) has emerged as another ultrasensitive trace element detection and assay technique, especially when the element is present in weak or nonuorescent form (19). In LIPAS a pulsed dye laser tuned to an absorption band of the analyte molecule produces a thermal and hence a pressure pulse due to nonradiative (thermal) dissipation of the excitation energy. A pressure transducer, such as a piezoelectric material, is used for detecting the acoustic disturbance as a function of the laser frequency to generate the absorption spectrum. Different valence states of the same element present in different molecular forms in solution exhibit absorption peaks at different wavelengths and are detectable at micromolar to nanomolar concentration levels. The high intensity at the focus of laser beams gives rise to several nonlinear optical processes, among which coherent anti-Stokes Raman scattering (CARS) has emerged as an important tool for gas phase diagnostics (e.g, in combustion research aimed at improving the thermodynamic efciency of combustion engines and reducing pollutant emission) (28). Here, two laser beams with frequencies p and s are focused at a common location in the sample, and s is tuned such that the difference (p s) coincides with a Raman active vibrational-rotational mode in a constituent molecule. Since the incident wave frequency coincides with that of the Stokes shifted Raman scattered wave, it creates (together with the pump wave) a coherent excitation of the vibrational-rotational mode of the molecules. In turn, the pump radiation is coherently scattered by the excited molecules to produce a coherent anti-Stokes component at the frequency as p p s. Since CARS does not involve any net energy or momentum transfer to the molecule, the efciency is strongly enhanced only when the momentum conservation condition for the photons is satised, as expressed by the phase matching condition, kas kp kp ks. The preferred conguration for satisfying the phase matching condition is a converging angular combination of the incident laser beams that results in conve-
DYE LASERS
91
s
(a) Pump ks kp (b) k s kp
differential delay at times t and t t allows determination of absorption due to molecules at a distance of ct /2 extended over the path length c t /2. Rapid lateral scanning of the laser beam generates a complete air-pollution map at different heights in a short time, and pollution sources are localized for adopting environmental control measures. High-power dye laser oscillator-amplier systems pumped by the harmonics of Nd : YAG laser, excimer laser, or ashlamp pumped dye lasers are commonly used. Pollutants such as NO2, SO2, or NO can be detected with better than ppm level sensitivity at distances up to a few kilometers. Similar techniques have been demonstrated for monitoring the thickness and spreading of oil slicks in oceans. DIAL is also used for monitoring variations in the atmospheric temperature prole with time. Measurement of the broadening of the absorption line of atmospheric sodium, due to thermal-velocity-dependent Doppler shift, by DIAL with narrowband dye lasers provides the temperature information. Industrial Applications Since laser radiation is a costly form of energy, the use of dye lasers as industrial process drivers demands high process yields for each photon used and/or high value addition. As an example of the former, Klick (22) has carried out a detailed analysis of photoassisted curing of pigmented polymers using dye lasers instead of UV lamps. Polymers are used as protective coatings in the automobile and appliances industry. The study favors the use of relatively broad-band dye lasers not only for replacement of existing capacity but also for expanding the scope of applications. The polymerization proceeds in a mixture of monomers, oligomers, a photoinitiator (PI), a tertiary amine, and pigment as follows. Absorption of a photon by the PI leads to a proton pick-up from the amine, thus producing a free radical that initiates the polymeric chain reaction. Powdered TIO2 is used as a white pigment that is nearly opaque. However, the pigment transmission increases with wavelength, whereas the PI absorption decreases, rapidly for both, in the 400 nm to 420 nm region. Since the radiation has to reach the substrate through the coating for uniform curing, the coating efciency and speed exhibit a maximum at an optimum wavelength. For example, using DETX (2,4-diethylthioxanthone) as PI in a 35 m thick coating containing 24% by weight of TiO2 and dye laser pulse energy uence of 0.05 mJ/cm2 at 50 Hz repetition rate, the total exposure required for curing showed a minimum (24 mJ/cm2) at a wavelength of 422 nm. The required exposure was doubled for a laser detuned by 8 nm, which justies the use of tunable sources. UV lamps or lasers are not suitable for curing pigmented coatings because UV radiation cannot penetrate sufciently and produces wrinkles due to faster curing at the surface. Enrichment of uranium in the ssile isotope U235 using high-average-power dye lasers is being pursued actively in the United States, Japan, and France (29) for the production of fuel for nuclear reactors. It is projected that the process will be economically more attractive compared to existing technologies based on gas diffusion and high-speed centrifuge and offers scope for better utilization of nuclear fuel resources, because of the high selectivity of the laser-based process. Laser isotope separation (LIS) is an offshoot of the versatile technique called resonance ionization spectroscopy
Stokes
Sample CARS Signal (c)
Figure 9. Schematic diagram illustrating CARS. (a) Transitions between molecular energy levels for CARS. (b) Wave vector diagram showing phase matching in non-collinear geometry. (c) Experimental scheme for achieving similar phase matching condition.
nient separation of the CARS beam from the lasers (Fig. 9). The emission of the signal as a separate coherent beam, and the fact that it is at a higher frequency than the incident beams, helps in discriminating against background scatter and LIF at lower frequencies. Measurement of the vibrational-rotational spectra of the constituent molecules and the broadening of spectral lines provides information on species concentration, temperature, and distribution during chemical reactions in the gas phase. CARS has also been applied in determining the temperature and concentration of gases in metallo-organic chemical vapor deposition. Fairly good spectral (0.1 cm1), spatial (1 mm), and temporal (ns) resolution is possible with detection limits below 0.1 mbar. Pulsed dye lasers are being used in environmentally important applications such as remote detection and monitoring of pollutants in the atmosphere and on water bodies. The method, termed LIDAR (light detection and ranging), consists of sending a pulsed laser beam out to the remote target zone and detecting the signal backscattered from absorbing molecules after a time delay that determines the distance of the sampled region. To minimize spreading of the laser beam at a large distance, the beam divergence is reduced by expanding and collimating with a telescope. The same telescope is used for collection of the backscattered signal. Among different variations of LIDAR, differential absorption lidar (DIAL) is a widely used technique that uses two synchronized dye lasers (or other tunable lasers), tuned to the peak and base of an absorption line of the molecule. Some systems use a high pulse repetition frequency dye laser with the facility for rapid wavelength switching between successive pulses. Taking the ratio of the scattered signal at the two wavelengths eliminates unknown contributions to signal attenuation (such as scattering). Measuring the signals with a gated
92
DYE LASERS
IP Energy (a) Atomic beam TOF-MS E E Ions Detector
Vapor source Combined laser beam (b)

Figure 10. Schematic diagram illustrating multistep resonance ionization process (a) and experimental setup (b). In the diagram shown in (a), three photons of different wavelengths are absorbed successively for ionization of an atom. Alternative schemes are available in Ref. 23. The dashed lines indicate neighboring energy levels of an isotope of the same element or another element that is left unaffected to a large extent. As shown in (b), instead of detection of ions by biased electrodes (E) only, a time-of-ight mass spectrometer (TOFMS) is often used for identication of isotopes and increasing selectivity. This technique is termed as resonance ionization mass spectrometry (RIMS).
process stems from the fact that for pulsed dye laser excitation, atomic transitions can be saturated by relatively small uence (few J/cm2 to few mJ/cm2). The pulses from the different dye lasers are synchronized for efcient excitation and ionization. However, high repetition rates on the order of several kilohertz are necessary to avoid loss of atoms ying through a nite-sized interaction zone in between successive laser pulses that would otherwise reduce the sensitivity for trace detection. The high repetition rate (4 kHz to 20 kHz) copper vapor laser, its current hybrid variants (which aim to improve the performance by producing low-vapor-pressure copper halides through chemical reactions), and futuristic, high-power diode-laser-array pumped, Q-switched, high-repetition-rate, solid-state lasers are attractive choices as pump lasers for dye lasers used in RIS applications. The dye lasers themselves are technologically more complex, requiring highspeed (several m/s) dye solution ow through specially designed dye cells to remove pump laser-induced thermal gradients between successive pump pulses. In Lawrence Livermore National Laboratory in the United States, one of the most powerful (2500 W average power), monochromatic ( / 108), high-repetition-rate (26 kHz) dye laser facilities has been constructed for demonstration of LIS on a plant scale. The facility consists of several chains of copper vapor lasers and dye lasers congured as MOPA systems. Medical Applications Dye lasers are nding increasing application in medical treatment and diagnostics. Flashlamp pumped dye lasers are being used for laser lithotripsybreaking of stones in the ureter or gall bladder by laser radiation. The laser is tuned to the peak absorption wavelength of the stone so as to reduce the power requirement and thus minimize the risk of causing peripheral injuries to the patient. This optimum wavelength depends on the stone composition and is determined by LIF using the same laser at lower power. The laser as well as the LIF signal are carried via an optical ber that allows less invasive treatment procedures to be adopted. An important application of dye lasers is in the treatment of cancer by photodynamic therapy (PDT). The patient is injected with a drug, HpD, a derivative of hematoporphyrin (C34H38O6N4) that is preferentially retained by the malignant tissue. After two to three days the tumor sites are irradiated by laser radiation tuned to the peak of a red absorption band of the drug at 630 nm, either directly or through an optical ber, depending on the access to the tumor. Absorption of light initiates a photochemical reaction that destroys the malignant tissue containing HpD and leaves the healthy tissue unaffected. The use of lasers is dictated by the high-energy exposure required ( few hundred J/cm2) and the ease of efciently coupling the well-directed laser light to an optical ber.
(RIS)the most sensitive and selective technique for detection and analysis of trace quantities of atoms and isotopes in the vapor phase (23). In practice (see Fig. 10), an atomic beam of the material is produced in a vacuum enclosure by vaporizing the sample and collimating the emerging vapor. The heating is done by electric current, ion beam sputtering, laser melting, and vaporization or by a focused electron beam, depending on the material and the application. Two or more narrowband dye lasers are tuned precisely to transitions of the target isotope or element for excitation from the ground state to successively higher energy levels, resulting in ionization. The ions are collected by biased electrodes placed away from the atomic beam. The combined laser beam is incident at near normal angles on the atomic beam. In this arrangement, the Doppler broadening of the transitions for most elements, including uranium, is small compared to the isotope shiftthe small difference in transition frequency between energy levels is due to differences in nuclear mass, size, and charge distribution between the different isotopes. The difference in the transition frequencies of isotopes or of different elements acts as the spectroscopic discriminator that allows narrower-linewidth dye lasers to be used for detection and collection of trace quantities of the target isotope/element with high selectivity in the presence of high concentrations of other isotopes and elements. Multistep sequences enhance selectivity by multiplication of selectivity in each step, provided, of course, that the spectroscopic discrimination in cumulative excitation is not compromised. The efciency of the
FUTURE SCOPE With the arrival of tunable solid-state lasers such as Ti : Sapphire (TiS), semiconductor diode lasers, and the optical parametric oscillator (OPO), replacement of dye lasers in the future is being discussed, debated, and perhaps aggressively projected by commercial manufacturers. The relative merits
DYE LASERS
93
of the different systems in the narrowband tunable output regime are discussed here. Narrowband (v 3 GHz) TiS lasers pumped by CW argon ion laser or second harmonic of Q-switched pulsed Nd : YAG laser can be tuned over very broad wavelength ranges, about 710 nm to 800 nm in the CW or 680 nm to 980 nm in the pulsed mode of operation. Broadband (450 GHz) operation over 670 nm to 1100 nm is possible using different mirrors with optimized reectivity in different wavelength ranges. High-average-power operation at a high repetition rate is more difcult than in dye lasers due to limitations in power dissipation and nonlinear optical effects. When pumped by pulsed lasers, the long lifetime (3 s) of the upper laser level in TiS results in long buildup times (several tens of nanoseconds) for laser action and exhibits large jitter (5 ns to 10 ns). This is not acceptable in applications requiring nanosecond synchronization of different lasers. Extension of the spectral range by frequency doubling leaves a substantial gap in the useful blue-red wavelength region. Additional frequency mixing techniques are employed to cover this range, but at the cost of further reduction in efciency, increased complexity in synchronized tuning of more optical elements, and requirements for high-power and costly pump lasers to drive all the nonlinear processes efciently. TiS crystal replacement costs are higher, and tailoring the photophysical properties, as is possible in dye lasers (such as changing dye concentration or solvent), is not feasible. Thus, in spite of comparative drawbacks of dye lasers, such as the use of owing liquid media that is also toxic and hazardous and hence restrains airborne, submarine, or other eld applications, dye lasers will continue to attract users and developers because of their larger wavelength coverage, ease of operation, and lower cost. Tunable semiconductor diode lasers currently cover wavelength ranges of 15 nm or more around discrete center wavelengths in the red (635 nm) near IR region (2 m). CW single longitudinal mode operation at power levels in the milliwatt range is possible. Emission in the bluegreen range has been demonstrated with groups II to VI semiconductors like ZnSe, or by waveguided frequency doubling techniques, but these are in developmental stages. An attractive hybrid conguration would be pulsed amplication of the output of semiconductor diode lasers in pulsed dye ampliers. Since the diode laser power is too small to saturate the gain in pulsed dye ampliers, carefully designed regenerative (multipass) ampliers, or oscillators seeded by the diode laser output, will be necessary to achieve efcient saturation of the amplier gain and to suppress ASE. With the development of efcient, high-damage-threshold nonlinear crystals such as -BaB2O4 (BBO) and LiB3O5 (LBO), optical-parametric oscillators and ampliers have emerged as a serious rival to dye lasers due to the unique capability of producing a continuously tunable coherent output over a very wide spectral range [e.g., 410 nm to 3 m in BBO pumped by third harmonic of Nd : YAG laser at 355 nm (30)] from a single device. In OPO, when a nonlinear crystal is placed inside a resonator and irradiated with an intense pump laser with frequency p directed along the resonator axis, the nonlinear polarization induced in the crystal generates two outputs, the signal and idler, at frequencies s and t, respectively, such that p s t. These new elds grow at the expense of the pump eld provided that they are co-propagat-
ing and satisfy the phase matching condition, for which the birefringence of the crystal is again exploited. The resonator provides the feedback necessary for efcient generation of a coherent output. For a given orientation of the optic axis of the crystal with respect to the resonator axis, the energy conservation and phase matching conditions together determine the frequencies of the signal and the idler uniquely. Tuning is then simply achieved by tilting the crystal so that the optic axis orientation is changed. In the preceding example, the signal is tunable from 410 nm to 710 nm while the idler, which can be separated from the signal by specially coated beamsplitters, simultaneously tunes from 3 m to 710 nm. When pumped three times above threshold, the bandwidth of the signal wave varies from few angstro ms at 410 nm to several nanometers close to degeneracy at 710 nm (s i). For narrowband operation, GIG architecture borrowed from the dye laser congurations has been applied successfully. However, the threshold pump power for OPOs is high in comparison to dye lasers; for the example cited (30), a 12 mm long BBO crystal placed in a 3 cm long cavity consisting of mirrors of reectivity 96% and 70% (output coupler) at the signal wavelength range, and transmitting for both idler and pump wavelengths, requires a threshold pump power of 20 MW/cm2 to 40 MW/cm2. Focusing a low-power pump laser tightly for generating threshold intensities is counterproductive because in anisotropic birefringent crystals the direction of energy ow (ray direction) is usually different from that of the propagation vector (normal to the wavefront); as a result, the signal wave walks off from the pump wave, faster for a smaller-diameter beam, impeding energy extraction from the pump. For convenient broadband operation above threshold, where optical conversion efciencies greater than 30% may be obtained, at least 50 mJ/pulse at 355 nm is desirable. In narrowband OPOs the threshold increases substantially due to insertion losses of dispersive elements and increase in cavity length (smaller number of round trips). Further, since the optical parametric process responds almost instantaneously to the changes in the pump laser, multi-longitudinal-mode pump lasers that show strong, intermode beat modulation within the pulse prole are unacceptable for reliable and efcient OPO operation. State-of-the-art pump lasers, injection seeded with single-mode output of diode-array-pumped solidstate laser, are necessary for pumping useful OPOs. The narrower bandwidth of the pump laser in single-mode operation is also important for restricting the available OPO gain within a narrower spectral band that enables better extraction efciency with narrowband feedback or injection. The pump laser also needs to have exceptionally good beam quality and beam-pointing stability. Due to these several restrictions on the pump laser performance characteristics, the OPO turns out to be a much costlier alternative. Beam quality, pulse shape, and especially linewidth of pump laser are minor issues in dye lasers. In addition to the high capital cost, chances of damage and costly replacement of OPO crystals or other optics are discouraging aspects that will continue to favor the use of dye lasers for common laboratory-based applications. The requirement of high pulse energy can be relaxed by injection seeding an OPO with a narrowband output from either a dye laser (pulsed or CW) in the signal wavelength range or from tunable diode lasers in the idler range. With the former scheme, a commercial OPO system working at 1 kHz repetition rate with a linewidth of less
94
DYE LASERS
than 1 GHz has been marketed, while single-mode operation has been demonstrated in the laboratory with both schemes. Molecular engineering of laser dyes and improvement of solvent characteristics continue to be challenging areas for improving laser efciency, arresting photochemical degradation processes, and improving solvent compatibility. An example (31) is the use of biuorophoric laser dyes for increasing efciency of FLDLs. The optical energy conversion efciency in FLDLs is limited to 1% because of poor absorption of the wideband ashlamp output by the dye, which exhibits strong absorption over a much narrower band (Fig. 1). To improve the efciency, energy transfer form the excited singlet state of a shorter wavelength absorber dye (donor) to the laser dye (acceptor) has been investigated extensively. The donor helps in converting a larger part of the ashlamp output into useful excitation of the laser dye. Limited success has been achieved in some cases via radiationless energy transfer of the Foerster type when the uorescence and absorption band, respectively, of the donor and acceptor overlap. More commonly, singlet to triplet crossing in the donor and subsequent triplet-triplet absorption at the lasing band of the acceptor were found to reduce laser efciency. A biuorophoric combination of a donor (p-terphenyl) and an acceptor (dimethyl-POPOP) was synthesized in which, due to proximity of the molecules, the energy transfer rates are substantially enhanced compared to singlet to triplet crossing rates in the donor. The engineered molecule showed a reduction in threshold by 10% from that of POPOP alone and by 70% from that of the simple mixture. As a host medium, water and its isotopic analog, D2O, possess much better thermo-optical properties than organic solvents, such as large heat capacity and low dependence of refractive index on temperature. Additional advantages are noninammability, nontoxicity, and ease of disposability. However, due to the high dielectric constant of water, dye molecules aggregate to form nonuorescing dimers even at the low concentrations required for laser action. Several laser dyes also show poor solubility in water. Both of these drawbacks are alleviated to a large extent by adding detergents such as sodium dodecyl sulfate, cetyl-trimethyl-ammonium bromide, and Triton X-100. Above a certain critical concentration, the detergent molecules or ions are organized into cagelike assemblies or micelles. These micelles incorporate the dye molecules in the interfacial regions or in the surface layers. This helps in solubilization and provides an environment where dimerization is inhibited. In some cases, use of watersurfactant combination has been found to increase uorescence quantum yield and laser efciency by rigidizing the dye structure. Jones (32) provides a useful discussion of photochemical tailoring of laser dye properties. Detailed listings of laser dyes are available in Refs. 4 and 33. Attempts to use dye molecules in the vapor phase with direct discharge excitation have not been successful due to rapid dissociation of molecules. However, laser action in solidstate dye-doped materials is gaining attention due to fabrication of improved hosts, such as modied polymethyl methacrylate (MPMMA), organically modied silicates (ORMOSIL), and other nanocomposites. Narrow linewidth ( 1.2 GHz) operation in prism-expander Littrow grating congurations at 9% efciency, single-longitudinal-mode operation in GIG resonators, and efciencies exceeding 60% for broadband emission have been reported. Localized photochemical degradation and slow heat dissipation restrict operation to low rep-
etition rates. Laser output power has been found to reduce by 33% after irradiation with 20,000 pulses at 0.6 J/cm2 at a single location. For a discussion and references readers may refer to Ref. 34. A new approach consists of using two-photon absorption of infrared laser light by new synthesized dyes that leads to excitation of the singlet state, followed by direct upconversion lasing in the visible. Intracavity upconversion lasing in a dye with high two-photon absorption cross section has been demonstrated by placing a dye-solution-lled cell inside the cavity of a Q-switched Nd : YAG laser (35). It is conceivable that long lengths of MPMMA or ORMOSIL bers doped suitably with such new dyes and pumped by IR lasers launched into the ber would act as efcient sources with very low thresholds for frequency upconverted outputs. With feedback provided only at the pump end, the ASE generated at the entrance would travel down the ber lagging behind the pump pulse and grow in intensity to saturate the gain in the rest of the ber. Double-ended pumping with suitable delay and pump power distribution may improve efciency and offer scope for optimization of the device. BIBLIOGRAPHY
1. L. Hollberg, CW dye lasers, in F. J. Duarte and L. W. Hillman (eds.), Dye Laser Principles, New York: Academic Press, 1990. 2. T F. Johnston, Jr., Tunable dye lasers, in Encyclopedia of Lasers and Optical Technology, San Diego, CA: Academic Press, 1991. 3. P. N. Everett, Flashlamp-excited dye lasers, in F. J. Duarte (ed.), High Power Dye Lasers, Berlin: Springer, 1991. 4. L. G. Nair, Dye lasers, Prog. Quantum Electron., 7: 153268, 1982. 5. F. J. Duarte and D. R. Foster, Lasers, dye, technology and engineering, in G. L. Trigg (ed.), Encyclopedia of Applied Physics, vol. 8, New York: VCH Publishers, 1994, pp. 331352. 6. R. Wallenstein, in M. L. Stitch (ed.), Laser Handbook, vol. 3, Amsterdam: North-Holland, 1979. 7. T. W. Ha nsch, Repetitively pulsed tunable dye lasers for high resolution spectroscopy, Appl. Opt., 11: 895898, 1972. 8. (a) M. G. Littman and H. J. Metcalf, Spectrally narrow pulsed dye laser without beam expander, Appl. Opt., 17: 22242227, 1978. (b) I. Shoshan, N. N. Danon, and U. P. Oppenheim, J. Appl. Phys., 48: 4495, 1977. 9. A. F. Bernhardt and P. Rasmussen, Design criteria and operating characteristics of a single-mode pulsed dye laser, Appl. Phys., B26: 141146, 1981. 10. (a) W. Kaiser (ed.), Ultrashort Laser Pulsers and Applications, Berlin: Springer, 1988. (b) P. M. W. French, The generation of ultrashort laser pulses, Rep. Prog. Phys., 58: 169267, 1995. 11. (a) S. Watanabe et al., Multiterawatt excimer laser system, J. Opt. Soc. Am., B6: 1870, 1989. (b) F. K. Tittel et al., Blue-green dye laser seeded operation of a terawatt excimer laser amplier, in M. Stuke (ed.), Dye Lasers: 25 Years, Berlin: Springer, 1992. 12. F. P. Scha fer (ed.), Dye Lasers, 2nd ed., Berlin: Springer, 1990. 13. K. Dasgupta and L. G. Nair, Effect of excited state absorption at signal wavelength in pulsed dye laser ampliers, IEEE J. Quantum Electron., QE-26: 189192, 1990. 14. K. Dasgupta, S. Kundu, and L. G. Nair, Extraction efciency of saturated-gain high-power dye laser ampliers: Effect of nonlinear signal absorption, Appl. Opt., 34: 19821988, 1995. 15. R. S. Hargrove and T. Kan, High-power efcient dye amplier pumped by copper vapor lasers, IEEE J. Quantum Electron., QE16: 11081113, 1980.
DYNAMIC PROGRAMMING 16. F. J. Duarte, Narrow linewidth pulsed dye laser oscillators, in F. J. Duarte and L. W. Hillman (eds.), Dye Laser Principles, New York: Academic Press, 1990. 17. M. G. Littman, Single-mode pulsed tunable dye laser, Appl. Opt., 23: 44654468, 1984. 18. F. J. Duarte (ed.), High Power Dye Lasers, Berlin: Springer, 1991. 19. W. Demtro der, Laser Spectroscopy, 2nd ed., Berlin: Springer, 1996. 20. J.-C. Diels, Femtosecond dye lasers, in F. J. Duarte and L. W. Hillman (eds.), Dye Laser Principles, New York: Academic Press, 1990. 21. R. L. Fork et al., Compression of pulses to six femtoseconds by using cubic phase compensation, Opt. Lett., 12: 483485, 1987. 22. D. Klick, Industrial application of dye lasers, in F. J. Duarte and L. W. Hillman (eds.), Dye Laser Principles, New York: Academic Press, 1990. 23. V. S. Letokhov, Laser Photoionization Spectroscopy, Orlando, FL: Academic Press, 1987. 24. L. Goldman, pp. 419432 in Ref. 1. 25. M. H. Anderson et al., Observation of Bose-Einstein condensation in a dilute atomic vapour, Science, 269: 198, 1995. 26. C. S. Adams and E. Riis, Laser cooling and trapping of neutral atoms, Prog. Quantum Electron., 21 (1): 179, 1997. 27. C. Moulin et al., Time-resolved laser-induced uorescence in the nuclear fuel cycle, Proc. 6th Int. Symp. Adv. Nucl. Energy Res., Innovative Laser Technol. Nucl. Energy, March 2325, 1994, Ibaraki, Japan: JAERI, 1995. 28. W. Richter, In situ gas-phase diagnostics by coherent anti-Stokes Raman scattering, in M. Stuke (ed.), Dye Lasers: 25 Years, Berlin: Springer, 1992. 29. Proc. 6th Int. Symp. Adv. Nucl. Energy Res., Innovative Laser Technol. Nucl. Energy, March 2325, 1994, Ibaraki, Japan: JAERI, 1995. 30. A. Fix et al., Efcient narrowband optical parametric oscillators of beta-barium-borate (BBO) and lithium-triborate (LBO), in M. Inguscio and R. Wallenstein (eds.), Solid State Lasers: New Developments and Applications, New York: Plenum, 1993. 31. F. P. Scha fer, Dye lasers and laser dyes in physical chemistry, in M. Stuke (ed.), Dye Lasers: 25 Years, Berlin: Springer, 1992. 32. G. Jones II, Photochemistry of laser dyes, in F. J. Duarte and L. W. Hillman (eds.), Dye Laser Principles, New York: Academic Press, 1990. 33. M. Maeda, Laser Dyes: Properties of Organic Compounds for Dye Lasers, New York: Academic Press, 1984. 34. F. J. Duarte, Opportunity beckons for solid-state dye lasers, Laser Focus World, May 1995. 35. G. S. He et al., Intracavity upconversion lasing within a Qswitched Nd : YAG laser, Opt. Commun., 133: 175179, 1997.
95
KAMALESH DASGUPTA
Bhabha Atomic Research Center
DYNAMICAL SYSTEMS. See LINEAR DYNAMICAL SYSTEMS, APPROXIMATION.
DYNAMIC DATA STRUCTURES. See LIST PROCESSING.
}{{}}
HOME
q
ABOUT US
q
CONTACT US
q
HELP
q

Electronic Speckle Pattern Interferometry Standard Article Elliott Horch1 1Rochester Institute of Technology, Rochester, NY Copyright 1999 by John Wiley & Sons, Inc. All rights reserved. DOI: 10.1002/047134608X.W6305 Article Online Posting Date: December 27, 1999 Abstract | Full Text: HTML PDF (310K)
q q
Browse this title

q

r
r r
Abstract
The sections in this article are Laser Speckle Interferometry Stellar Speckle Interferometry Electronic Imaging Devices Used in Speckle Interferometry Speckle Interferometry Today About Wiley InterScience | About Wiley | Privacy | Terms & Conditions Copyright 1999-2008John Wiley & Sons, Inc. All Rights Reserved.
ELECTRONIC SPECKLE PATTERN INTERFEROMETRY
615

A speckle pattern is a complicated interference phenomenon that is observed in two main areas of research: laser optics and stellar astronomy. In the case of lasers, speckle patterns are generated when the coherent light from the laser is reected off an abraded or optically rough surface. The grainy or speckled appearance of the laser light when viewed after reection is known as laser speckle. In astronomy, it has been known for some time that highly magnied short-exposure images of stars have a similar speckled appearance when viewed through the atmosphere, and that each bright speckle has a size that is easily related to the diameter of the telescope. Figure 1 shows an example of both a laser speckle pattern and a stellar speckle pattern. The study of both kinds of patterns, with the aid of electronic imaging devices, is known as electronic speckle pattern interferometry. Laser speckle interferometry is a tool used in many areas of engineering for precise measurement of physical systems, whereas in astronomy, it is a way to compensate for the blurring of images caused by the atmosphere and to reconstruct extremely highresolution images of astronomical objects. Though they differ in some important ways, the two types of speckle patterns have a common origin. Light emanating from a laser is coherent, which means that it travels in phase in the form of a plane wave. After reection from an abraded surface, however, the light is no longer coherent. Each point in the beam may be thought to have a random phase, and when brought into focus, the random phases interfere with one another, creating a eld of speckles on the image plane. Likewise, if light from a distant star were monochromatic, it would arrive as a plane wave at the top of the earths atmosphere. (To approximate the monochromatic situation, stars may be observed through a narrow wavelength band pass lter.) As the light travels down through the atmosphere to the telescope, it encounters inhomogeneities in the refractive index of the air, so that when the light enters the primary telescope aperture, random phase delays are again present. After brought to a focus by the telescope optics, the phase delays interfere, again creating a speckle pattern. LASER SPECKLE INTERFEROMETRY While laser speckle had been observed since the rst use of lasers in the 1960s, the science of speckle interferometry began with the idea that the speckle elds could be used to make sensitive measurements of test samples, such as vibraJ. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
616
tion or displacement studies (1). The extraction of information from the test sample is nondestructive and does not require that the sample have a specular surface. For these reasons, speckle interferometry has found many uses in engineering today. Figure 2 shows a simple laser speckle interferometer system that uses a single laser as the light source. Collimated, coherent light from the laser strikes a beam splitter and then the double beam proceeds to the two arms of the interferometer, marked A and B in the gure. At the end of arm A are two mirrors, reecting the light toward the imaging detector as a reference beam. At the end of arm B is a test sample with an optically rough surface. Light is reected back to the beam splitter where it is recombined with light from arm A and imaged onto the imaging device. Images, like the one shown in Fig. 1(a), are then recorded by the imaging detector and transferred to computer.
Arm A Laser
Beam splitter Arm B
Detector Wedge Test object Lens Aperture
Computer
Figure 2. A simple laser speckle interferometry system.
In principle, many types of imaging devices could be used in a speckle interferometer. However, television camera systems (also known as vidicons) have typically been used in the past, and now charge-coupled device (CCD) cameras are common. These are relatively inexpensive and adequate for many laser speckle applications, in contrast to astronomical speckle interferometry systems, where the requirements for the imaging detector are much more stringent because of the intrinsic faintness of most target stars and the fast, random uctuations of astronomical speckle patterns. (These uctuations usually change faster than the standard video rate of 30 Hz.) Vidicons and CCDs also can interface easily with computers, so that digital images of the speckle patterns can be stored directly on computer for processing in various ways that enhance the contrast of the interference features. These and other imaging detectors are discussed in more detail in a later section. Basic Characteristics of Laser Speckle Fields The size of speckles in the speckle eld is determined by the optical design of the interferometer. When a lens is used, as in Fig. 2, the speckles at the focal plane of the imaging device are referred to as subjective speckles, since their size is determined in part by the magnication of the lens or lens system. Assuming that the lens images the surface of the object under study, the size of speckles at the imaging device is given by 1.2 (1 + M )F (1)
where M is the magnication of the lens, F is its f /number and is the wavelength of light being used. This can be related to the size of speckles on the surface of the object, , by 1.2 (1 + M )F /M = /M (2)
Figure 1. (a) Model speckle pattern that might be generated from a laser speckle interferometer, (b) model speckle pattern that might be generated by a star observed with a large telescope.
If the light reected from the surface of the object is combined with a uniform, coherent reference beam, which is a very common technique in laser speckle interferometry, the size of speckles will change. This is becuse the image formed at the focal plane of the detector may be thought of as the sum of the interference patterns generated by pairs of points in the aperture. Each such pair would produce a series of fringes whose spacing is determined by the distance between the two
617
point-apertures (as in Michelson interferometry), but taken together, the various fringes form the complicated interference pattern of the speckle eld. Without the reference beam, the size of the speckles is essentially determined by interference between light traveling through opposite sides of the lens, since pairs of points with the largest separation will have the smallest spacing between fringes. It is the diameter of the lens that therefore controls the speckle diameter in this case. Adding the reference beam has the effect of placing a strong central ray in the aperture, and then the most obvious interference effect generated by pairs of points is when one of the points is the central ray. The largest effective separation is then between the center of the aperture and the edge, which corresponds to a fringe spacing double that of the previous case, and the speckle size doubles likewise. The distribution of brightness of the speckle eld is another basic property of speckle patterns. Assuming that the test surface truly randomizes the phases from the incoming coherent laser light, and that many scattering points contribute to the image of the speckle eld at the detector, it can be shown that the intensity distribution of a single speckle eld follows the negative exponential distribution p (I ) = 1 exp I0 I I0 (3)
where I is the intensity, p(I) is the probability of a point in the image having intensity I, and I0 is the average intensity of the eld. It is interesting that in this case the most probable intensity at any point in the image is zero, since p(I) is a monotonically decreasing function of I. As with the size of speckles, the intensity distribution of speckles changes when a reference beam is introduced. In the case where the reference beam has intensity equal to the average intensity of the speckle eld, the distribution becomes p (I ) = 2 2I exp 1 + I0 I0 J0 (2 2I/I0 ) (4)
rst, the object under analysis is placed in the interferometer and a speckle pattern is recorded. The sample is then subjected to whatever test the operator desires, and then a new speckle pattern is recorded of the same region of the sample. The two patterns can then be combined electronically by taking the negative of the rst pattern and adding it to the second. If the two patterns are identical, which is to say that the two images are perfectly correlated, then the speckles will exactly cancel and the result will obviously be zero. If the two patterns are uncorrelated in some regions of the speckle eld, then the speckles will not cancel and the sum at a given speckle location will vary from zero in a random fashion in these regions. This can give rise to a speckled fringe pattern, where fringes contour lines of equal displacement in the sample. An example of a speckled fringe pattern is shown in Fig. 3. This image was formed as a result of the above analysis when the model sample experienced a shear in the direction perpendicular to the plane of the sample surface between recording the rst and second speckle patterns. The displacement of the sample was linear in the vertical direction, creating a fringe pattern following the contours of equal displacement in the horizontal direction. A dark fringe occurs when the displacement is equal to a multiple of the wavelength of the laser light. The technique described here is a powerful tool in strain analysis of mechanical pieces, allowing for measure of minute warping or shearing effects, but the contrast in the fringes is often poor. A second kind of correlation technique involves displacing speckle images slightly from one another. As in the previous technique, one exposure is typically recorded before some mechanical test and a second after it is complete. The images are displaced slightly by a row or column shift of the rst digital image and then added together. If the two images are highly correlated, there will be speckle pairs or doublets all over the eld separated by the shift given the rst image. If the elds are not correlated, the doublets will not be present. The occurrence of speckle doublets is easy to determine either
where J0 is the zero-order Bessel function of imaginary argument. Again the most probable intensity is zero, and in practice the two distributions are difcult to distinguish visually. However, if the pattern is recorded electronically, its statistics can be derived easily. A third important feature of laser speckle elds is the contrast C dened by C = I /I0 (5)
where I is the standard deviation of intensity values (the noise) in the speckle eld and I0 is the average intensity of the eld. In the case of an interferometer with a specular reference beam, as in Fig. 2, it can be shown that C 2, but in many cases C is a constant that depends on details of the optical system (2). This means that the noise in a speckle pattern is proportional to the local average intensity, which is sometimes referred to as multiplicative intensity noise. Two Common Methods of Speckle Correlation While several important methods have been used for correlating speckle elds, this discussion will be limited to two. In the
Figure 3. A model speckled fringe pattern, as generated in a displacement analysis.
618
by spatial ltering or by computing the spatial frequency power spectrum of the summed image, Q(u), given by Q(u ) =
x u I (x )e 2 jx 2
(6)
where I(x) is the summed image, x is the variable describing position on the image plane, and u is the spatial frequency variable conjugate to x. The fringe pattern formed in this way tends to have very high contrast, and fringes only form if the two elds are highly correlated. Both correlation techniques described here were rst used not with electronic detectors but with speckle interferometry done with photographic plates, or speckle photography. Each speckle pattern image had to be developed and studied later. Electronic detectors made possible real-time fringe pattern processing, which not only made the speckle experiments faster, but also made larger the number of correlation techniques that could be easily implemented. Motion and Vibration Analysis Two of the simplest kinds of speckle analyses can be done by visual inspection of the recorded speckle patterns. The rst is motion. If a test sample is slowly moving in the speckle interferometer, the speckle pattern it generates at the imaging detector will change. The speckles will begin to twinkle, since the intensity of individual speckles is very sensitive to the position of the sample. This type of analysis is often used to monitor holographic recording, especially when the sample may expand during the recording process because of change in humidity or some other variable. Vibration analysis is another example. If a sample is vibrating, and the amplitude of the vibrations is more than a fraction of a wavelength, the speckle pattern in the region of the vibration will begin to blur out and lose contrast as the speckles begin to change in intensity with the vibration. As a result, a speckle pattern recorded from a vibrating object will have a loss of contrast in regions that are moving. Nodal lines in the sample (areas of little or no motion) will still show the familiar high-contrast speckle pattern. In this way, speckle patterns can map the nodal structure of samples under vibration. Displacement and Misalignment Analysis Displacement of a test object can occur either in a direction in the plane of the surface of the object or in a direction perpendicular to the object surface (in the line of sight of the imaging system). For displacement in the line of sight of the imaging system, the previously described correlation techniques are often used. For localized displacement in the plane of the surface of the test object, several specialized types of speckle interferometers have been used and optimized for measure of in-plane displacement, such as the pioneering instrument of Leendertz (3). This instrument produced fringe patterns based on optical path length differences from two laser beams when the sample was moved. It is also true that small in-plane displacements of a test sample produce a speckle pattern that is merely shifted from the original pattern. This fact lends itself to a kind of double exposure analysis similar to the one described above. An exposure is taken before and after the test sample is shifted. These images are
then summed. Speckle doublets will then be present in the summed exposure, where the spacing between the matching speckles scales with the physical displacement. Once the spatial frequency power spectrum of the image is formed, a fringe pattern will be present where the spacing of the fringes is inversely proportional to the physical displacement. A misalignment, or tilt, in a physical system can also be measured with a speckle interferometer. It is possible to show that for laser light at normal incidence to the test surface, and viewing of the surface likewise near the normal, the test sample will act as a mirror under small rotations about any axis that lies in the plane of the surface of the sample. This means that if the sample is tilted by an amount , the speckle pattern will be rotated through an angle 2 . At the image plane, this rotation of the speckle pattern will be viewed as a translation of the speckles. If the camera system is defocused by a known amount U, and a double exposure speckle pattern is recorded before and after the tilt, the amount of tilt can be derived from the speckle displacement by = d/2M U (7)
where d is the displacement of the speckle pairs and M is the system magnication. STELLAR SPECKLE INTERFEROMETRY In conventional imaging with ground-based telescopes, astronomers must contend with the atmosphere, which blurs images signicantly. The problem is so severe that for virtually all research class telescopes in operation today, the resolution of images is not determined by the optical design of the telescope but rather by the atmosphere through which the images are viewed. Loss of resolution is caused by a combination of two effects: rst, inhomogeneities in the refractive index of the air above the telescope aperture break up stellar images and spread the light over a much broader region at the telescope focus than if the atmosphere were not present, and second, these inhomogeneities vary rapidly with time. Stellar speckle interferometry is a way to compensate for the blurring effects of the atmosphere and obtain very high-resolution information about astronomical objects from the ground. The method is usually performed on data taken at a large-aperture telescope, and high-resolution images are reconstructed after the observation. For this reason, it is referred to as a single-aperture passive technique, in contrast to other methods such as adaptive optics, where high-resolution images are built up during the observation, or Michelson interferometry, where more than one aperture is used to make observations. High-resolution images play a key role in several areas of astronomy. One of the most important areas is in deriving the orbits of binary stars. Binary star orbits are currently the only well-established way to obtain estimates of the masses of stars. High-resolution images also yield important information about the environment around young stars and the percentage of double and multiple star systems in young stellar associations. Speckle interferometry has made a signicant contribution to several other kinds of astronomy, such as imaging asteroids and planets from the ground, measuring stellar diameters, and obtaining high-resolution information about supernova remnants.
619
Large Telescopes and the Atmosphere In typical astronomical imaging applications the target is very faint, and a long exposure time is necessary to obtain a high-quality image. Over such long time scales, the effect of the atmosphere is such that a perfect source is blurred into a smooth quasi-Gaussian image with width ranging from a few tenths of a second of arc to several arcseconds, depending on observing conditions. It was rst realized in the 1960s that highly magnied short exposure images of stars are not simply smooth quasi-Gaussian images, but instead are speckle patterns that have quasi-Gaussian envelope. The speckles form as a direct consequence of the effect of the atmosphere on the light coming from the star. To see why this is so, consider an idealized astrophysical point source of monochromatic radiation. If this radiation were to enter the telescope in the absence of the atmosphere, it would be a plane wave and would therefore have uniform magnitude and phase across the telescope aperture. For the purposes of this discusison, it will be sufcient to use the model of a scalar electromagnetic eld, so that the aperture function may be dened as the scalar (complex-valued) function representing the electromagnetic eld entering the primary telescope aperture. An elementary theorem from optics states that the image of a perfect point source (known as the point spread function and henceforth called the PSF) of the telescope is the modulus square of the Fourier transform of the aperture function (4): S(x ) =
x z / z A(z )e 2 jx dz 2
(x )|2 = |A
(8)
where z is the variable describing the plane of the aperture, x is the variable describing the image plane, S(x) is the PSF, denotes the Fourier transA(z) is the aperture function, and form. Figure 4 shows a sample zero atmosphere aperture function and PSF for a large telescope. The width of the PSF denes the resolution obtainable at the image plane. Since in astronomy the image corresponds to the intensity distribution projected on the sky (where the position of objects are measured in angular terms), it is common to refer to the width of the PSF in terms of arcseconds, as shown in Fig. 4(b). The atmosphere changes the character of the aperture function. Because of motion and temperature uctuations in the air above the telescope aperture, inhomogeneities in the refractive index develop. These inhomogeneities break the aperture into cells with a typical length scale of about 10 cm that are translated by a prevailing wind (in the simplest model) across the telescope aperture on a typical time scale of about 10 ms. Within a cell, the phase of the aperture function remains roughly uniform but it takes on different values from cell to cell. Because of the relationship between the aperture function and the PSF, the PSF is likewise changed; in fact, the time-evolution of the aperture function implies that the PSF is also time-dependent. If the atmosphere is frozen at a particular instant in time, the resulting PSF may be described as the sum of interference patterns between pairs of subapertures dened by the cells arising from the inhomogeneities in the refractive index. Just as with the size of laser speckles, each such pair would produce a fringe pattern whose spacing is determined by the distance between the two subapertures (as in Michelson interferometry), but taken to-
Figure 4. (a) Gray scale plot of the aperture function of a point source for a large telescope in the absence of the atmosphere. Since the aperture function is complex values, only the phase is plotted. The central region is assumed to be occulted by a secondary mirror. (b) The resulting point spread function from the aperture function in (a). Except for the occultation of the central region of the aperture, this would be the familiar Airy pattern from diffraction theory.
620
of the broad quasi-Gaussian envelope, called the seeing disk of the star. For this reason, the resolution possible with conventional ground-based astronomy is much poorer than the diffraction-limited resolution of large telescopes. While the speckle patterns contain structure on the scale of the diffraction limit, these structures are lost in any exposure of an object longer than a few tens of milliseconds. Classical Speckle Interferometry The rst technique for obtaining high-resolution information in the presence of atmospheric turbulence was developed by Labeyrie (5) in 1970. He pointed out that there is a way to obtain the diffraction-limited spatial frequency power spectrum of an object from a collection of individual speckle patterns. For sources of sufcient simplicity, such as binary stars, the power spectrum is a powerful tool for deducing high-resolution features. This was an important initial step toward high-resolution image reconstruction. As mentioned above, the uctuations of the atmosphere occur on a time scale of typically about 10 ms, so a stellar speckle observation would consist of a sequence of many short exposures of this typical exposure length. Each frame is then a xed quasitime-independent speckle pattern like the one shown in Fig. 5(b). In this case, each frame preserves structure on scales much smaller than the seeing disk, as evidenced by the speckles themselves and to the extent that the speckles can be resolved by the imaging detector on this time scale, each frame contains high angular resolution about the source. Suppose that the object of interest is a binary star. Then the speckle patterns will exhibit speckle doublets at the vector separation of the two companions, as shown in Fig. 6. Of course, these speckle pairs will occur at various places on the
Figure 5. (a) Gray scale plot of a model instantaneous aperture function of a point source for a large telescope in the presence of atmospheric uctuations. Since the aperture function is complex valued, only the phase is plotted. The central region is assumed to be occulted by a secondary mirror. (b) The resulting point spread function from the aperture function in (a), which is known as a stellar speckle pattern.
gether, the various fringe patterns form a speckle pattern. The speckles ll a broad region, typically 1 to 2 arcseconds in diameter on the image plane, but have a characteristic width similar to that of the diffraction-limited PSF of the telescope. An example of an (instantaneous) aperture function and PSF of a large telescope in the presence of atmospheric uctuations is shown in Fig. 5. If Fig. 5 is now evolved in time, some of the phase cells pass out of the telescope aperture and new ones enter, and so the position and intensity of the speckles change on the image plane. On an exposure considerably longer than the time scale of atmospheric uctuations, the speckle nature of the PSF washes out to leave only the shape
Figure 6. A model speckle pattern of a binary star, exhibiting double speckles. In this example, the separation vector of the two components is about twice the size of a speckle and directed toward 1 oclock.
621
image plane, but they may be collected together by considering the autocorrelation function of the frame, which, up to a normalization factor, is equivalent to the histogram of the distance vectors between photon pairs. The autocorrelation function (x) of a frame is dened by (x ) = x1 I (x 1 )I (x 1 + x ) dx (9)
the function into two components, a seeing-limited (low-frequency) portion approximated by
Tlf (u ) = |Tt (u )|2 exp 6.88
u r0
5 /3
u r0
1 /3
(14)
and a diffraction-limited (high-frequency) portion approximated by Thf (u ) = 0.435 r0 D

2
where I(x) is the frame image and * denotes complex conjugate. The Fourier transform of the image autocorrelation is the image power spectrum (dened as the modulus square of the Fourier transform of the frame image). The frame image, on the other hand, is the convolution of the instantaneous point spread function S(x) with the actual intensity distribution of the object O(x), where the convolution integral is dened by I (x ) = x1 O(x 1 ) S(x x 1 ) dx (10)
Tt (u )
(15)
In the Fourier domain, a convolution becomes an ordinary product so that (u ) O (u ) (u ) = S I (11)
once again denotes the Fourier transform and the where spatial frequency variable conjugate to x is u. Taking the modulus square of this expression and averaging over many frames, the average image power spectrum is obtained: (u )|2 |O (u )|2 (u )|2 = |S |I (12)
where denotes the average over many frames. The function (u)2 is known as the speckle transfer function. If a speckle S (u)2 can be observation is performed on the binary star, I obtained by Fourier transforming the average autocorrelation of many frames of data. The same process on a point source (u)2. As (that is, an unresolved star) yields an estimate of S Eq. (12) shows, the true binary power spectrum can then be obtained by division, where such a division is valid in the re (u)2 is nonzero. In fact, the average point gion where S spread function power spectrum is nonzero out to the diffraction limit of the telescope, and the method works not only for binary stars but also for general objects. In the case of a binary system, the power spectrum is a fringe pattern. The spacing of the fringes is related to the separation of the stars, the orientation of the nges denes the relative orientation of the two stars (up to a 180 ambiguity), and the amplitude of the fringes is related to the relative brightness of the components. A variation on the autocorrelation method described here called the directed vector autocorrelation (6) can remove this ambiguity in the orientation, and is particularly useful in the study of binary stars. The Speckle Transfer Function As above, the speckle transfer function T(u) is dened by (u )|2 T (u ) = |S (13)
where u u, r0 is the Fried parameter describing the length scale of coherence cells in the aperture function, D is the telescope diameter, is the wavelength of observation, and Tt(u) is the (diffraction-limited) telescope transfer function. Though most of the power is in the low-frequency seeinglimited portion of the function due to the attenuation factor (r0 / D)2 in the high-frequency portion, a nonzero high-frequency shoulder can also be seen out to the diffraction limit of the telescope. As discussed above, for a binary star, a fringe pattern would appear on top of the speckle shoulder, but more complicated objects would exhibit other functional forms in this region. The low-frequency portion of the transfer function has a width that is determined by the Fried parameter. If r0 is large, the seeing peak in frequency space is large, and the seeing disk on the image plane is therefore small, but if r0 is small, then the low-frequency portion of the transfer function is small and the seeing disk is large. The quantity r0 / , which is a measure of the width of the low-frequency transfer function, is sometimes called the seeing cutoff in frequency space. The high-frequency portion of the transfer function is an attenuated version of the transfer function of the telescope, which means that the speckle transfer function manifestly extends to the diffraction-limit, roughly given by the ratio D / in the frequency plane. The attenuation factor involves the quantity (r0 / D)2; because of the relationship between the Fried parameter and the seeing, the value of the transfer function on the speckle shoulder is therefore a sensitive function of the seeing. An important consideration of the method of speckle interferometry is the expected signal-to-noise ratio (SNR) of the image power spectrum. In the high-frequency wing of the power spectrum (i.e., above the seeing cutoff), the SNR of a general object O(x) is given by
(u )|2 S nsp Tt (u ) |O (u ) = M (u )|2 N 1 + nsp Tt (u ) |O
(16)
and relates the image power spectrum to the true object power spectrum. Theoretical models of T(u) generally break
where nsp is the number of photons per speckle, and M is the number of frames averaged. When the number of photons per (u)2, the SNR increases speckle is small compared to Tt(u) O linearly with nsp, meaning essentially that the frames are limited by photon statistics, and when the number of photons per (u)2, the SNR becomes speckle is large compared to Tt(u) O asymptotically independent of nsp, and the frames are limited by the atmospheric process. This interplay between photon statistics and the atmosphere also determines the optimal frame exposure time. The frame time should be chosen so that the motion of the speck-
622
les within the frame is small, but if too short a frame time is chosen, frames will be photon-starved and this leads to a loss of SNR. On the other hand, if too long a frame time is chosen, speckle motion will lead to a loss of contrast in the frames, and again the SNR is reduced. It also depends on atmospheric (or equivalently, on seeing) conditions in the sense that shorter frame times are required when the seeing is poor. Image Reconstruction In principle, if one had a diffraction-limited estimate of the (u), it would be simple (generally complex-valued) function O to Fourier invert the function to obtain a reconstructed image. In order to do that, however, one needs to have estimate of (u). Classical speckle both the modulus and the phase of O interferometry as discussed above falls short of providing true image reconstructions because the recovered function is the (u)2, and clearly contains objects power spectrum, that is, O no phase information. This is known as the phase problem in speckle interferometry. Solutions to the phase problem began to appear in the literature starting in the early 1970s. Some of the most common methods that have been used to reconstruct images from speckle data since that time include speckle location rearrangement, the Knox-Thompson algorithm, and bispectral analysis. The rst method does not explicitly calculate the phase of the objects Fourier transform, but consists of a manipulation of the speckle patterns in the image plane that results in a high-resolution image. The last two methods do (u), which is provide a means of calculating the phase of O then combined with the modulus estimate, derived from the classical autocorrelation analysis. Speckle Location Rearrangement. One approach to obtaining diffraction-limited images from speckle data is to consider bright, clearly identiable speckles as estimates of the diffraction-limited PSF of the telescope. In general, speckle patterns contain speckles of varying brightness, and the fainter speckles are not clearly identied. For brighter objects, where many photons are detected in each frame, many speckles are clearly visible, as in Fig. 1(b), but for fainter objects, this may not be the case. Take, for example, the case where only the brightest speckle is easily identiable in each frame. In this case, a simple way to proceed is to shift each frame so that the speckle is in the center of the frame (or the brightest pixel is in the center of the frame), and then co-add all of the frames. This is known as the shift-and-add technique (7). A slightly more complicated approach is possible when there is sufcient contrast in each frame to identify several speckles (8). In this case, each frame of a general object may be written as
n
of p(x) O(x). The function p(x) can be obtained from a point source observation, and a further deconvolution can be performed to yield the desired function O(x). KnoxThompson Algorithm. Knox and Thompsons solution (9) to the phase problem involves the analysis of a generalized version of the frame image power spectrum called the frame cross-spectrum, dened by (u )I (u ) X (u1 , u2 ) = I 1 2 (18)
(The power spectrum is obtained when u1 u2.) Using Eq. (11), this can be rewritten in terms of the instantaneous point spread function and the true object intensity distribution. Averaging over many frames results in (u )S (u ) O (u )O (u ) X (u 1 , u 2 ) = S 1 2 1 2 (19)
(u1) S (u2) is known as the cross-spectral The function S transfer function. It can be shown that the cross-spectral transfer function takes on nonnegligible values only for frequencies u1 and u2 satisfying |u 1 u 2 | < r0 (20)
which is to say, their difference has to be smaller than the seeing cutoff in frequency space. Since this region is small compared to the region of the frequency plane inside the diffraction limit, it is convenient to dene u1 u and u2 u u, with u r0 / . Then the phase of Eq. (19) can be written in the form (u )I (u + u ] = (u ) (u + u ) + (u, u + u ) arg[ I XT O O (21) and XT is the phase of the crosswhere O is the phase of O spectral transfer function. In analogy with classical speckle interferometry, the function XT(u, u u) can be measured using a point source calibration object, then subtracted from the phase of the image cross-spectrum. The result is (u )I (u + u )]} (u , u + u ) = (u ) (u + u ) arg{[I XT O O (22) which is a difference equation for the function O. Using two nonparallel values for u and an appropriate initial condition, it is therefore possible to integrate this equation and obtain an estimate of O over the entire u-plane, out to the diffraction limit. A standard initial condition is to set O(0) 0, since the reconstructed image must be a real function, and therefore its transform must be Hermitian. Bispectral Analysis. The standard way to calculate the image bispectrum (10) is to start from the triple correlation of a frame image, which is dened by C(x 1 , x 2 ) = x I (x )I (x + x 1 )I (x + x 2 ) dx (23)
I (x ) =
i= 1
(x i ) p(x ) O(x )
(17)
where n is the number of speckles identied in the frame, (x) is the impulse function, the xi represent the locations of the speckles on the frame, p(x) is the prole of a single speckle, I(x) and O(x) are the frame image and the true object intensity distribution as previously dened, and denotes the convolution integral. The frame image can, for example, be iteratively deconvolved by the impulse eld, leaving an image
The Fourier transform of the triple correlation is called the image bispectrum and can be written in the form (u , u ) = I (u )I C 1 2 1 (u 2 )I (u 1 + u 2 ) (24)
623
Using Eq. (11), the right side of Eq. (24) may be rewritten in terms of the instantaneous point spread function and the true object intensity distribution, just as with the cross-spectrum. Averaging over many frames, this becomes (u , u )] = O (u )O (u )O (u + u ) S (u )S (u )S (u + u ) [C 1 2 1 2 1 2 1 2 1 2 (25) (u1)S (u2) S *(u1 u2) is known as the bispecThe quantity S tral transfer function. It can be shown that the bispectral transfer function has zero phase, so that the phase of Eq. (25) is simply (u , u )]} = arg[O (u )O (u )O (u + u )] arg{[C 1 2 1 2 1 2 (26)
reconstructed image: (1) phase reconstruction and (2) Fourier (u). inversion of O The phase is obtained from the bispectrum or cross-spectrum by integration, but it is possible to integrate the phase derivative map in several different ways, the simplest being via iteration, proposed by Lohmann et al. in 1983 (10). From the initial condition of zero phase at the origin (pixel 0) and the difference equation, the phase at pixel 1 is calculated, then from the value at pixel 1 and the difference equation, the value at pixel 2 is calculated, and so on. However, this method has the disadvantage that errors increase as the spatial frequency increase, making the phase estimate least reliable in the high-frequency region of frequency space. Another technique, used by Meng et al. (12), is a relaxation method, where zero phase or some initial guess at the phase map is assumed at the start and the phase at a particular pixel in the frequency plane is assigned by using the difference equation and the starting values of the phase at nearby pixels. This produces a new phase map, which replaces the initial guess and the process is repeated. After many iterations, the phases converge to the correct values. Still another method is to assume that the object has a particular intensity distribution (for example, assume that it is a binary star), and to do a best t to the bispectrum (or cross-spectrum) to arrive at the correct phase map (13). After the phase map is obtained and combined with the modulus estimate of an object, there is the matter of Fourier inversion to arrive at a reconstructed image. Essentially the problem is that to reconstruct the image unambiguously, one must know all of the Fourier components of the object, whereas in any real observation, only some of the components are obtained, and these all have some noise associated with them. Many techniques for Fourier inversion used so far in speckle imaging have been borrowed from radio astronomy. One example is the maximum entropy technique (14), which produces the smoothest map consistent with the data. Another approach is the clean algorithm (15), which assumes that the solution is a collection of point sources, and derives a map of their locations. Observing Stellar Speckle Patterns Several properties of stellar speckle patterns as observed through the atmosphere impose special requirements on the detection system. First, although a monochromatic speckle pattern was previously discussed and shown in Fig. 5, light from stars is not monochromatic. The character of speckle patterns is dependent on wavelength, in part because of a wavelength-dependent magnication of the image, and in part because of true differences in the optical path through the atmosphere as a function of wavelength. To avoid these problems, it is necessary to use a narrow-band pass lter to maintain a high degree of contrast to the patterns on the image plane. Using a narrow lter signicantly reduces the total amount of light in of the speckle pattern, and limits the range of magnitudes for which the method is useful. In fact, the larger the telescope diameter, the narrower the band pass one must use, so that the limiting magnitude is only a weak function of collection area of the primary mirror of the telescope. Another problem is that even with narrow-band pass lters, the color dispersion of the atmosphere (which is a function of zenith angle) is large enough to elongate speckles on
Consider the region of the bispectrum such that u1 u and u2 u with u some small xed increment in frequency space. The bispectrum is a four-dimensional function for a two-dimensional image, so the requirement u2 u xed denes a subplane of the bispectrum close to the plane u2 0. at u be denoted by O as in the previous Let the phase of O section and Eq. (26) may be rewritten as (u , u )]} = (u ) + ( u ) (u + u ) arg{[C 1 2 O O O (27)
If it were not for the term O(u), this would again be a difference equation for the function O(u) as with the cross-spectrum, and therefore an estimate of this function could be built up over the entire frequency plane using two nonparallel vectors u (or equivalently, using at least two subplanes from the bispectrum). The added term complicates this, but only slightly. If one chooses O(u) arbitrarily, the effect on the recovered function O(u) is to add a linear term to it. A linear phase term added to a function in frequency space merely translates the position of the inverse transform in image space, but image features are not affected. It is therefore sufcient to choose O(u) arbitrarily for the purposes of image reconstruction. As with the KnoxThompson algorithm, a standard initial condition to integrate the difference equation is to set O(0) 0. Nakajima (11) gave the rst general expression for the SNR of the bispectrum at arbitary light levels. At low light levels, the SNR is proportional to n3/2, where n is the number of photons per frame. This can be compared to the power spectrum, where in the low light level regime, the SNR is proportional to n. In the low light regime, only the so-called near axis subplanes (subplanes with u small) have high SNR. For simple images, these subplanes are usually sufcient to derive good reconstructed images in the way discussed above. For more complicated images, other regions of the bispectrum may play a more important role in determining the phase map. Nakajima found that the limiting magnitude for bispectral analysis was between 13th and 15th magnitude, depending on telescope aperture size and seeing conditions. Phase and Image Reconstruction. Bispectral analysis and the KnoxThompson algorithm share the feature that they produce from the speckle data an estimate of the phase deriva (u) and this function must be integrated before it is tive of O combined with a modulus estimate obtained from the standard autocorrelation analysis of classical speckle interferometry. From this point there are two steps before arriving at a
624
the image plane, and this effect must be removed. This is generally done with the use of two zero-deviation Risley prisms that can be rotated independently. The dispersions of the two prisms add vectorially, so it is possible to rotate the prisms to positions such that the resultant vector exactly counteracts the dispersion vector of the atmosphere in the direction of the target. The techniques discussed above for deriving high-resolution information from stellar speckle patterns depend on the convolution relation, that is, that the frame image is the convolution of the frame speckle pattern and the true object intensity distribution. However, this assumes that the optical path through the atmosphere is identical for all points in the eld of view of the telescope. For points very close together on the sky, this is a very good approximation, but as the separation between two points increases, their optical path through the atmosphere begins to diverge, and as they encounter different turbulent cells in the atmosphere, they generate different speckle patterns. This means that the deconvolution is only valid when all points of interest are contained within the region of the sky known as the isoplanatic patch. The isoplanatic patch describes the area of sky over which speckle patterns remain very much the same and is usually only a few arcseconds in diameter at visible wavelengths, and perhaps 30 arcseconds in the near infrared. Unlike laser speckle interferometry, where the speckle patterns can be controlled somewhat in the laboratory, a stellar speckle interferometer must have an imaging detector with several special characteristics. Most importantly, the imaging device must be capable of reading frames out at the rate of the evolution of the speckles; a minimum requirement is about 10 ms in the visible, and somewhat longer in the infrared. The detector must have relatively high quantum efciency so that speckles in a single frame have sufcient contrast. Low read noise and low dark current are important because the number of detected photons per frame is usually small by astronomical standards. Since the goal is to obtain very high-resolution information inside a comparatively large seeing disk, excellent geometric delity and fairly large formats are necessary. ELECTRONIC IMAGING DEVICES USED IN SPECKLE INTERFEROMETRY In both laser and stellar speckle interferometry, TV cameras and photography were popular methods for recording speckle
h
patterns in the 1970s, but in the 1980s and 1990s virtually all speckle pattern interferometry has been done with more sophisticated electronic imaging detectors. Current instrumentation can be divided roughly into two main categories: (1) photon-counting cameras (including intensied-CCDs), and (2) solid state detectors (i.e., CCDs and infrared arrays). Photon-Counting Devices Photon-counting imaging devices are cameras which make use of the photoelectric effect in such a way that they can detect individual photons. Several types of photon counters have been used for speckle imaging, including precision analog photon address (PAPA) detectors (16), resistive anode devices (13), wedge-and-strip detectors (17), multianode microchannel array (MAMA) detectors (18), and intensied-CCDs (19). By far the most popular of these is the intensied-CCD, but all use a microchannel plate (MCP) or MCP stack for signal amplication where the front end of the plate has a photocathode held at high voltage and the back end is held nominally at ground. A photon then strikes the photocathode and liberates one or more photoelectrons. As a photoelectron is accelerated through a microchannel, it contacts the channel wall several times and with each contact, secondary electrons are liberated. When the event exits the MCP, a total charge cloud of 105 to 107 electrons from a single input event can be produced. A microchannel therefore acts as a small, continuous photomultiplier tube when a photoelectron is accelerated through it. Figure 7 shows common types of high-gain MCPs. The photocathode can be made of a number of different materials, depending on what wavelength response is desired. Visible light photocathodes with reasonable quantum efiency have been available for many years, and the one used with most photon-counting speckle interferometers is the so-called S-20 or trialkali photocathode. The S-20 gives a quantum efciency of about 10% through the visible and out to the near infrared. Photon-counting devices have traditionally been used at low light levels, where CCDs cannot be used because of read noise. Because the same photocathode materials are available for all of these cameras, they all have about the same limiting quantum efciency. The distinguishing feature separating one photon-counting camera from another is the way that the output of the MCP is detected. This process affects not only the overall quantum efciency of the device, but also its linearity and timing resolution.
h h
Figure 7. Three types of microchannel plate arrangements common in photon-counting detectors. The left-most example is a single curved channel MCP or C-plate, the middle example is known as a chevron arrangement, and the right-most example is known as a Z-stack. Copyright 1992 by D. B. Kasle, reprinted with permission.
Charge amplifiers
Charge amplifiers
Charge amplifiers
625
Solid-State Detectors CCDs. CCDs are an obvious choice for speckle work today because of the large formats that are available, their outstanding imaging characteristics, and their ease of use, especially in conjunction with computers. They are probably the most common devices for laser speckle intererometry, and indeed they are attractive for astronomical speckle interferometry because of their much higher quantum efciency than the currently available photon-counting devices. Traditionally, they have had two drawbacks for stellar speckle interferometry. First, CCD readout speed has not been fast enough to take advantage of their quantum efciency. Typically, the desired frame integration time is obtained by some sort of shuttering operation, then the frame is read out and meanwhile the detector is effectively off, creating a large source of dead time. Second, their read noise also negates some of the advantage of the higher quantum efciency by making it impossible for frames to have photon-limited statistics. This is especially important if the objects of interest are faint. Nonetheless, the properties of CCDs are improving very rapidly. It is possible now to obtain CCDs that can read out one million pixels per second with root-mean-square (rms) read noise of less than 10 electrons, and that have quantum efciency of about 90% ). For a 128 128 pixel subarray, this at 700 nm (7000 A means that the device is capable of recording about 60 frames per second. The problems of read time and read noise do not affect laser speckle interferometry applications, since the light source and sample can be controlled in the laboratory. Infrared Arrays. Infrared array technology has become very important for astronomy in the last decade. In the 1980s some one-dimensional arrays were used for speckle imaging, but more recently, some outstanding two-dimensional infrared speckle imaging has been possible with the advent of highquality two-dimensional infrared arrays. InSb arrays have been used the most so far, which have a wavelength response from about 1 m to 5 m. A common format for the two-dimensional arrays used in the early 1990s in speckle imaging was 58 62 pixels, with larger arrays (256 256) available now. Like the CCDs discussed above, these arrays offer high quantum efciency (about 80%), but read out the data in full frames. Today, the frame rates possible tend to be of order 10 per second (20), so as with visible CCDs, some of the advantage of the high quantum efciency is sacriced due to the detector dead time while frames are being read out.
ference (or equivalently, the relative brightness) of the two stars in a binary system to better than 0.5 magnitudes, because of calibration problems and other issues (21). Since this is not very precise, and since a binary star system represents the simplest nontrivial photometric problem, this creates doubt about the photometric information of more complicated reconstructed images. Indeed, reconstructed images are generally assumed to have excellent astrometric information and imperfect photometric information. The problem of doing precise photometry with the speckle technique remains predominant in the eld at the moment. One of the most important aspects of speckle interferometry is that is has dramatically increased the awareness that many astronomers have about the power of optical interferometry and atmospheric compensation. This would not have been possible without some of the new instrumentation developed in the 1970s and 1980s, which has made speckle data taking more efcient and the observations themselves better. Now it is almost commonplace to measure objects in the sky with high angular resolution. In this sense, speckle interferometry has helped to pave the way for other kinds of high resolution astronomy such as adaptive optics and optical longbaseline interferometry, which are being pursued today. Impact on Precise Measurement Laser speckle interferometry provides a relatively easy way to make precise measurements, where in many cases the test object does not need to be prepared in any special way. This has brought the power and precision of interferometry to the ngertips of many engineers, but in addition, the use of electronic imaging has made the interpretation and manipulation of speckle data much easier. Though still limited in some ways, electronic speckle pattern interferometry offers the advantage of simplicity over holographic interferometry or speckle photography. With the advent of digital imaging devices, the number of analysis tools for use with speckle patterns has greatly increased, and continues to increase today, making electronic speckle pattern interferometry useful for a wider and wider range of measurement problems. BIBLIOGRAPHY
1. E. Archibold, J. M. Burch, and A. E. Ennos, Recording of in-plane surface displacement by double-exposure speckle photography, Opt. Acta, 17: 883898, 1970. 2. G. Slettemoen, First-order statistics of displayed speckle patterns in electronic speckle pattern interferometry, J. Opt. Soc. Am., 71: 474482, 1981. 3. J. A. Leendertz, Interferometric displacement measurement on scattering surfaces utilizing speckle effect, J. Phys. E, 6: 214 218, 1970. 4. D. J. Schroeder, Astronomical Optics, San Diego: Academic Press, 1987. 5. A. Labeyrie, Attainment of diffraction limited resolution in large telescopes by Fourier analysing speckle patterns in star images, Astron. Astrophys., 6: 8587, 1970. 6. W. G. Bagnuolo et al., Absolute quadrant determinations from speckle interferometry of binary stars, Astron. J., 103: 1399 1407, 1992. 7. R. H. T. Bates and F. M. Cody, Towards true imaging by wideband speckle interferometry, Opt. Commun., 32: 365369, 1980.
SPECKLE INTERFEROMETRY TODAY Astronomical Impact In making scientic conclusions from high-resolution images reconstructed from speckle interferometry data, a primary concern is the quality of the reconstruction. This can be judged in terms of the delity of the astrometry of the image (measuring the relative positions of objects) and the photometry of the image (measuring the relative brightness of objects). While there is ample evidence that speckle interferometry can give astrometric results of outstanding quality (from the study of binary star orbits, for example), the photometric accuracy of the method is not on such rm ground. A recent paper by the leading group in binary star speckle interferometry states that it is difcult to determine the magnitude dif-
626
ELECTRONIC WARFARE
8. S. P. Worden, C. R. Lynds, and J. W. Harvey, Reconstructed images of alpha Orionis using stellar speckle interferometry, J. Opt. Soc. Am., 66: 12431246, 1976. 9. K. T. Knox and B. J. Thompson, Recovery of images from atmospherically degraded short-exposure images, Astrophys. J., 193: L45L48, 1974. 10. A. W. Lohmann, G. P. Weigelt, and B. Wirnitzer, Speckle masking in astronomy: Triple correlation theory and application, Appl. Opt., 22: 40284037, 1983. 11. T. Nakajima, Signal-to-noise ratio of the bispectral analysis of speckle interferometry, J. Opt. Soc. Am. A, 5: 14771491, 1988. 12. J. Meng et al., Triple correlation subplane reconstruction of photon-address stellar images, J. Opt. Soc. Am. A, 7: 12431250, 1990. 13. T. Nakajima et al., Diffraction-limited imaging. II. Optical aperture-synthesis of two binary stars, Astron. J., 97: 15101521, 1989. 14. J. Skilling and R. K. Bryan, Maximum entropy image reconstruction: General algorithm, Mon. Not. R. Astron. Soc., 211: 111 124, 1984. 15. P. W. Gorham et al., Recovery of diffraction-limited object autocorrelations from astronomical speckle interferograms using the CLEAN algorithm, Astron. J., 100: 294306, 1990. 16. C. Papaliolios, P. Nisenson, and S. Ebstein, Speckle imaging with the PAPA detector, Appl. Opt., 24: 287292, 1985. 17. J. Barnstedt, R. Neri, and M. Grewing, The AIT-MCP-detector system: Spectroscopic and speckle tests, Mitt. Astron. Ges., 68: 240243, 1987. 18. J. S. Morgan, Speckle imaging with the MAMA detector, in F. Merkle (ed.), High Resolution Imaging by Interferometry, Garching bei Mu nchen: European Southern Observatory, 1989. 19. H. A. McAlister et al., ICCD speckle observations of binary stars. II. Measurements during 19821985 from the Kitt Peak 4-m telescope, Astron. J., 93: 688723, 1987. 20. D. W. McCarthy, Jr. et al., The low-mass companion of Gliese 22A: First results of the Steward Observatory infrared speckle camera, Astron. J., 101: 214219, 1991. 21. W. I. Hartkopf et al., ICCD speckle observations of binary stars. XIII. Measurements during 19891994 from the Cerro Tololo 4m telescope, Astron. J., 111: 936945, 1996. Reading List J. C. Dainty (ed.), Laser Speckle and Related Phenomena, Munich: Springer-Verlag, 1984. J. W. Goodman, Statistical Optics, New York: Wiley, 1985. M. C. Roggemann and B. Welsh, Imaging through Turbulence, Boca Raton, FL: CRC Press, 1996.
ELLIOTT HORCH
Rochester Institute of Technology
ELECTRONICS, SUPERCONDUCTING. See SUPERCONDUCTING MICROWAVE TECHNOLOGY.
ELECTRONICS, THERMAL. See THERMAL ANALYSIS AND

DESIGN OF ELECTRONIC SYSTEMS.
ELECTRONIC TRANSPORT, HIGH FIELDS. See

HIGH-FIELD EFFECTS.
}{{}}
HOME
q
ABOUT US
q
CONTACT US
q
HELP
q

Excimer Lasers Standard Article J. W. Nicholson1, W. Rudolph1, J. Kleinschmidt2 1University of New Mexico, Albuquerque, NM 2Lambda Physik GmbH, Goettingen, Germany Copyright 1999 by John Wiley & Sons, Inc. All rights reserved. DOI: 10.1002/047134608X.W6306 Article Online Posting Date: December 27, 1999 Abstract | Full Text: HTML PDF (254K)
q q
Browse this title

q

r
r r
Abstract
The sections in this article are About Wiley InterScience | About Wiley | Privacy | Terms & Conditions Copyright 1999-2008John Wiley & Sons, Inc. All Rights Reserved.
182
EXCIMER LASERS
Table 1. Table of Potential Active Media for Excimer Lasers and Their Emission Wavelengths
Type
Diatomic rare-gas halides
Active Medium
XeCl XeBr XeCl XeF KrI KrBr KrCl KrF ArBr ArCl ArF NeF
Ar2F Ar2Cl Kr2F Kr2Cl Xe2F Xe2Cl
Ar2 Kr2 F2 Xe2
HgCl HgBr HgIl
Wavelength (nm)
253 282 308 351 185 206 222 248 161 175 193 308
285 245 400 325 610 430 25 15 35 15 65 40
Triatomic rare-gas halides
Rare-gas excimers
126 146 157 172

558 503, 498 444
Metal vapor excimers
The examples in bold face have particularly high-gain cross sections and are used in commercial excimer lasers.
rare-gas halide laser was reported by Searles and Hart in 1975 (4), and many other lasing compounds were discovered shortly thereafter. Extensive summaries of the various excimer compounds including their chemistry and spectroscopy are in (5) and (6). Active Media
EXCIMER LASERS
Excimer laser refers to a large class of lasers that use excimer, exciplex, or trimer molecules as the active medium. These molecules form stable compounds only in the excited state, and once de-excited, they dissociate on time scales of the order of 1012 s, which means that the lower laser level is practically always empty. Thus high pump and laser efciencies are possible. Excimer lasers are pumped by electrical discharges, electron or proton beams, or optical techniques. Today, a large number of lasing excimer compounds are known, spanning a wavelength range from the visible to the vacuum ultraviolet spectral region. Of the many possible active media, rare-gas halide lasers are the most widely applied as powerful, tunable radiation sources in the ultraviolet. Their application elds range from fundamental science to manufacturing and to medicine. This class of excimer lasers is the focus of this article. The use of excimer molecules as active laser media was proposed as early as 1960. The rst excimer lasers, based on rare gas excimers, were demonstrated in the early 1970s by Basov et al. (1), Koehler et al. (2), and Hoff et al. (3). The rst
Excimers and exciplexes form the active media of excimer lasers. They are chemically stable compounds that exist only in the excited state. In a strict sense, excimers are excited dimers made up of two identical subunits, whereas exciplexes are formed by two different compounds. Often, the word excimer is used in the literature for both species. Examples are listed in Table 1. As the name suggests, rare-gas halide excimers consist of a rare gas R and a halide X atom. Typically, there are no stable ground-state RX compounds. The situation is different for the excited state. A reaction of the type R + X RX (1)
produces a stable molecule in an excited state. This reaction is feasible because R* has a completely different electronic conguration than ground-state R, leading to distinctly different chemical properties. Whereas an R atom has no free valences, R* does and undergoes chemical reactions. Ground-state, rare-gas atoms Ne, Ar, Kr, and Xe have closed electronic shells with no free (bonding) electrons to form a molecule RX. The electronic conguration of excitedstate rare-gas atoms, 3P0,2, is very similar to that of an alkali
EXCIMER LASERS
183
Structures of rare gas monohalides R+ + X R* + X Selfabsorption R + X* R +X ( 2 ) R +X ( 2 )

Energy
- Strongly bound, ionic like alkalai halide
Fluorescent spectrum of KrF (Ar + 4% Kr + 0.2%F AT 350 kPa) 2 2

2 2
Kr2F
Broadband ( 2
2 )
Laser transition (2-2)
X(2)
2 )
- Strongly repulsive, covalent R+X
240 250 260 270 280 290 300 Wavelength (nm)
248 nm Laser
350
400 450 500
R X(
- Weakly attractive, covalent thermally unstable
Figure 2. Fluorescence spectrum of KrF* and Kr2F. Adapted from (7) by permission.
Internuclear distance
(a)
12 10 8
Energy (eV)
thermally unstable. The -bond gives rise to a repulsive potential. The emission spectrum of an excited RX molecule has a distinct maximum and a broad continuum. The former is attributed to the 2 2 transition on which lasing is achieved. The bound molecule decays to the ground state emitting a photon RX R + X + h (2)
D23/2
Kr*(2P1/2)+F Kr*(2P3/2)+F C 23/2
6 4 2 0 2 1 2 3 X 21/2 B
2
1/2
A 21/2, 3/2 Kr + F (2P3/2, 1/2) 4 5 R () 6 7 8
Typical uorescent lifetimes are on the order of ns to 100 ns. The dissociation of the (unstable) ground-state molecules occurs on a time scale of ps making the lower laser level practically empty, which means that a population inversion between RX* and RX is relatively easy to achieve. The emission cross sections are on the order of 1016 cm2, comparable with laser dye molecules. The cross section F is estimated from the uorescent lifetime F and spectral width F (7) F 0.04 4 cF (3)
Figure 1. (a) Potential diagram of a rare-gas halide molecule [adapted from (7) by permission]. (b) Energy level scheme of KrF and KrF*, from (8).
metal, such as Na. In both species, a single electron orbits a positive core, and they subsequently undergo a reaction to form molecules with halide atoms. Depending on which electron is transferred, a molecule in a 2 or 2 state is formed. The corresponding bond is ionic. The excited rare-gas atoms transfer one charge to the strongly electronegative halide. Thus many properties of the RX* molecules are similar to those of alkali halides. A detailed summary of the binding and potential curves of rare-gas halides is in (7). Figure 1(a) shows a potential diagram of the electronic structure of the rare-gas halides. Figure 1(b) shows KrF as an example. The ground state comprises a weakly attractive and a repulsive potential which is a consequence of the P-type halogen and the S-type rare gas atom. The binding type is covalent leaving the molecule in either a RX(2) or RX(2) state. The potential curve belonging to the former is either at or exhibits a shallow minimum which makes the molecule
assuming lifetime broadening and Gaussian transition proles. Depending on the pump mechanism and gas composition, several different reactions are possible, leading to a rare-gas halide excimer. As an example, several processes leading to KrF* in a high-voltage, high-pressure discharge containing Kr, F2, and Ne are listed, following (8). These are the main reactions leading to KrF*: Kr+ + F + Ne KrF + Ne three-body excimer formation (ionic channel) Kr + F2 KrF + F harpooning reaction (neutral channel) (5)

(4)
Table 2. Spectroscopic Data of KrF* Internuclear equilibrium distance Vibrational frequency (KrF*) Transitional wavelength Fluorescent linewidth Spontaneous emission lifetime Stimulated emission cross section 0.23 nm 310 cm1 248.5 nm 2 nm 7 ns 2.4 1016 cm2
184
EXCIMER LASERS
HV CS
4 2 5 3
(a)
Thyratron
CP
Charge transfer (b)
Figure 3. (a) Schematic diagram of an excimer laser discharge chamber. 1-electrodes, 2-preionization pins, 3-heat exchanger, 4-blower, 5-internal cleaning. (b) Electrical discharge circuit (charge transfer circuit). Cs-storage capacity, Cp-peaking capacity.
The ions Kr and F and the metastable Kr* are formed by collisions with electrons: e + Kr e + Kr e + F2 F + F e + Kr 2e + Kr+ (6) (7) (8)
For such an electrical discharge, total pressures between 1 and 3 bar are typical with a relative abundance of the species of Ne:Kr:F 98.4:1.5:0.1. A manifold of excited KrF states exhibits a potential minimum, a prerequisite for a stable compound. The ground state KrF possesses a dissociative potential which makes KrF an unstable molecule. KrF* decays spontaneously into KrF emitting a photon with a typical lifetime of 5 ns to 10 ns. In contrast the dissociation time of KrF is only a few picoseconds. Figure 2 shows the uorescent spectrum of a high-pressure gas mixture. In addition, a three-body collision is likely to occur leading to a triatomic molecule RX + 2R R2 X + R (9)
This reaction gives rise to a broad uorescent band at longer wavelengths and quenches lasing on the 248 nm line. Table 2 summarizes some spectroscopic data of the transition at 248 nm. Laser Operation Pumping. Most commercially available excimer lasers used for industrial and scientic purposes are based on self-sustaining, transverse electrical discharges. To ensure a homogeneous excitation and to prevent lamentation, the active volume is preionized before the main discharge is triggered. Preionization is accomplished, for example, by UV photopreionization where the UV photons are generated by pin-arc or corona discharges. Another possibility is to use X rays to preionize the gas in the discharge volume. The electric discharge circuit is designed for fast discharges (voltage: 20 to 30 kV, current: 30 kA to 50 kA) of durations between 10 and 50 ns. A typical discharge chamber is shown in Fig. 3(a) and a typical discharge circuit in Fig. 3(b). First the storage capacitor Cs is loaded up to the loading voltage, the peaking capacitor Cp is uncharged. Next the thyratron is switched which transfers the charge to the capacitor Cp. When the voltage at Cp reaches
the breakdown voltage of the gas mixture, an electrical discharge takes place pumping the active medium. A low inductance of the peaking circuit (a few nanohenries for Cp and 100 nH for the thyratron) guarantees a fast rise time of the discharge. In rare-gas halide lasers, small signal gain coefcients, g0 gN, on the order of 0.1 to 0.2 cm1 are desired, requiring excited molecules at densities of N 1015 cm3. This inversion density determines certain requirements for the pumping process, summarized in Table 3. Electron beam pumping is applied for very high-power, low-repetition rate laser systems. The (pulsed) electron beam is injected into the laser resonator and is steered and guided with the aid of magnetic elds. Several excitation geometries are possible, comprising longitudinal, transverse, and coaxial pumping (9). The primary excitation process ionizes the buffer gas which distributes the energy in collisions to the rare-gas and halide atoms. Typically an electron beam generated by an accelerator enters the discharge region through a thin window of aluminum or titanium 5 m to 50 m thick. A lower limit of the electron energy is caused by losses of about 200 keV in the entrance window. The maximum energy is determined by the maximum efciency of the energy coupling into the gas. The main advantage of electron beam pumping is the possibility of exciting large gas volumes to produce large laser pulse energies. Output energies of KrF systems of tens of kilojoule are reported (10). Typical pulse durations are between 0.01 and 1 s. Such lasers are important for laser fusion research. Small-scale excimer lasers based on waveguide-type active regions are operated at a several kilohertz repetition rate (11). They consist of a radio-frequency source as pump (pump power 10 kW/cm3), an impedance-matching circuit, and two electrodes between which is placed a small insulating tube ( 1 mm diameter, 30 mm long) containing the gas. The output pulse energies are typically not larger than a few tens of mi-
Table 3. Pump Parameters for Rare Gas Halide Lasers Required to Reach Small Signal Gain Coefcients of 0.1 to 0.25 cm1 Discharge current density Discharge electric eld strength Pump power density Pump energy density Pump duration 103 A/cm2 5 103 V/cm 5 MW/cm3 0.20.25 J/cm3 4050 ns
EXCIMER LASERS
Table 4. Laser Beam Properties for Different Excimer Laser Resonators Resonator Type Plane mirrors Unstable cavity Injection seeded Beam Divergence (mrad) 37 0.1 diffraction limited Spatial Coherence Length (nm) 0.20.5 25 beam diameter
185
crojoules. The walls of the active volume (diameter of the order of millimeter or less) act as waveguides. Output pulse durations up to a few hundred nanoseconds are possible and linewidths as narrow as a few 10 MHz are reported. Resonator. For many applications excimer lasers with resonators formed by plane mirrors are used. The output beam is multimode, exhibits a strong divergence of several milliradians, and is relatively broadband. Table 4 summarizes the spatial characteristics of excimer laser radiation for various resonator congurations. The divergence is reduced to about 100 rad with unstable resonators. Because of the small number of cavity round trips (typical outcouplers have transmission coefcients of 80% and higher), the magnication of such unstable resonators must be high (typically 10) (12). Difculties associated with implementing unstable cavities of high magnication factors are (1) the cavity feedback decreases with increasing magnication ( 1/ M2); and (2) the coating of the reecting spot on the defocusing mirror must sustain high intensities. Rare-gas halide lasers have emission spectra with line widths on the order of a few nm because the excimer molecule exists in several vibrational states. Thus the laser output consists of a band comprising the strongest vibrational transitions. The vibrational relaxation (thermalization) times for typical gas pressures are on the order of 100 ps. Because laser pulse durations are on the order of ns, it is possible to extract most of the excitation energy in a narrow laser line. This is important for applications such as high-resolution spectroscopy and UV photolithography. Such laser operation can be achieved by implementing suitable frequency-tuning and selecting elements. For extreme spectral narrowing, prism beam expanders or etalons combined with gratings are applied. Line widths 0.2 cm1 ( 1 pm) are feasible and have been demonstrated in commercial systems (13,14) which provide mJ energies. By cascading such frequency lters, even narrower laser output is feasible, albeit at the expense
of output power. Laser spectra as narrow as 0.02 cm1 at J energies were observed (15). Figure 4 shows an example of a narrowband and injectionlocked unstable resonator for an excimer laser. Excimer Laser Output Characteristics The main output parameters of typical commercial excimer lasers are listed in Table 5 and illustrated in Figs. 5, 6, and 7. It should also be mentioned that the wavelength of excimer laser radiation can be shifted by nonlinear optical effects. Raman shifts in gases (16) are particularly attractive because of the UV transmissivity and the avoidance of damage effects. Interaction of UV Laser Radiation with Materials Fundamentals. The generation, handling, and application of excimer lasers involves interacting powerful UV radiation with matter. At low power the response of the material is completely described by a complex dielectric constant, () = or complex refractive index, n 2 () = () = [n() i ()]2 (11)
1 ()
+ i 2 ()
(10)
where n is the (real) refractive index and is the imaginary part of the refractive index responsible for absorption. A UV beam incident on a material surface is partly reected and partly transmitted into the substrate. Under normal incidence the reection coefcient is given by R() = n () 1 n () + 1
2
(12)
The propagation through the substrate is affected by absorption leading to an exponentially damped intensity I(z): I (, z ) = I (, z = 0 )e ()z (13)
Etalon Oscillator
Prism beam expander
where 4 / is the coefcient of small signal absorption. A characteristic penetration depth is dened by dp 1.
Table 5. Typical Properties of Excimer Laser Radiation
Grating Amplifier Output

Figure 4. Schematic diagram of an injection-locked unstable resonator. The etalon grating sequence allows frequency narrowing and tuning. The electrical discharges of the two modules need to be synchronized to within 1 ns for optimum performance.
Output Power Pulse energy Pulse to pulse stability Pulse duration Beam size Divergence Bandwidth
Figs. 5, 6 Figs. 5, 6 0.51.5%, Fig. 6(b) 20 ns, Fig. 6(c) 5 mm 20 up to 50 mm 50 mm, Fig. 7 0.10.5 mrad (see Table 4) 0.5 nm (planeplane cavity) 1 pm (line narrowing elements)
186
EXCIMER LASERS
1000 100
Pulse energy (J)
240
1 kW 100 W
210 180 150 120 90 60 30
10 1 0.1 0.01
0 5 10 15 Vertical section
1 10 100 1000 Repetition rate (Hz) 10000
20 (a)
25
30
35
40 (mm)
240 210 180 150 120
Figure 5. Energy per pulse as a function of repetition rate.
100 90 80 70 60 50 40 30 20 10 0 0 25
Average power (W)
KrF XeCI ArF XeF F2
90 60 30 0 5 10 0 Horizontal section 15
(b)
20
25 (mm)
50 75 100 125 150 175 200 Repetition rate (Hz)

(a)
Figure 7. Beam intensity prole in (a) vertical and (b) horizontal direction of a KrF excimer laser (ComPex 205, Lambda Physik). From (23).
200
Number of pluses
Std. deviation: 1.2% 150 100 50 0 700 716 732 748 764 780 796 812 828 844 860 876 892 Energy (mJ) (b)
Nonlinear absorption occurs at high input intensities. A corresponding process of nth order is described by an absorption coefcient (n ) = const. I n 1 (14)
Absorption of UV light is generally accompanied by excitation of the electronic system through interband transitions in insulators (wide-gap materials) and semiconductors and interband and intraband transitions in most metals. Optical Components. Optical components with high UV light transmission are needed as outcouplers, beam splitters and focusing optics, for example. Ordinary glass materials are not suitable because of their large absorption in the UV. Table 6 shows some materials used for high-transmission optical components in the UV and their transmission limit determined by the bandgap. At high (pulse) intensities the absorption becomes nonlinear (see Eq. (14)) and, in addition, longer lasting absorption centers like color centers and defects occur. For reection optics, dielectric multilayer mirrors are mostly used with reectivities exceeding 99%. Wide-gap oxide and uoride compounds with high transmission in the UV serve as coating materials. High-reection layers are a sequence of alternating high and low refractive index quarterwave coatings. Typical combinations are HfO2 /SiO2 and LaF3 /MgF2. A common problem with high-power laser radiation is laser damage. Progress in the layer fabrication enabled coatings with damage thresholds above 10 mJ/cm2 when exposed to 20 ns KrF laser pulses. Metal mirrors (e.g., alumi-
10 ns / div
(c)
Figure 6. (a) Output power vs repetition rate for different types of excimer lasers. (b) Pulse to pulse stability of a 100 Hz, KrF excimer laser (LPX 210i, Lambda Physik). (c) Temporal prole of a typical Krf excimer laser pulse. The full width at half maximum is 25 ns. From (23).
EXCIMER LASERS
Table 6. Examples of Materials Used for Optical Components at Excimer Laser Wavelengths Material SiO2 (SQ1) Crystal quartz MgF2 CaF2 BaF2 LiF2 Al2O3 Band Gap (nm) 159 138 105 124 136 107 150 n(308 nm) 1.4856 1.486 1.405 1.46 1.382 n(248 nm) 1.5086 1.508 1.41/1.405 1.457 1.412 n(193 nm) 1.562 1.562 1.466/1.451 1.501 1.445
187
num) are low cost alternatives, but reectivities do not exceed 90% at excimer laser wavelengths. Interaction of UV Light with Strongly Absorbing Materials Absorption Processes. Most industrial UV laser applications remove materials after absorption of laser radiation. At excimer laser wavelengths the absorption is caused by resonances of the electronic system. Many materials exhibit a large absorption coefcient. As a result the excimer radiation is absorbed in a thin surface layer. At sufciently high excitation energies the deposited energy density is so high that a material surface layer is ablated. The physical mechanisms involved are manifold and depend on material, excitation wavelength, and pulse energy and duration. One can roughly divide these processes into photothermal and photochemical mechanisms. In metals, UV photons are absorbed by free electrons occupying states near the Fermi level. With absorption coefcients on the order of 2 106 cm1, the absorption length (skin depth) dp is only a few nanometers. This means that the energy of a UV laser pulse minus the reection loss is deposited in a small surface layer. This quantity (1 R) is shown for some materials and typical excimer wavelengths in Table 7. It is interesting to note that these numbers are about one magnitude larger than achieved with CO2 lasers at 10.6 m in the infrared spectral region. Depending on the metal (work function) and photon energy (laser wavelength), a certain fraction of the excited electrons are ejected out of the material, cooling the metal. The remaining excitative events populate higher states in the metal, and are followed by fast electronelectron collisions and nally electronphonon and electron-impurity scattering, leading to material heating. These processes proceed on a time scale of a few picoseconds and less. The resulting temperature distribution evolves according to the following heat conduction equation: 2 T (r, t ) S(r, t ) 1 T ( r, t ) = t K (15)
where S(r, t) (1 R)I0(t)ez is the source term due to the incident laser intensity I0, K is the thermal conductivity, and is the thermal diffusivity. With laser pulses, the heat generation rate can exceed the dissipation due to diffusion by many orders of magnitude resulting in heating the sample up to the melting temperature. Continuous heat supply can exceed the latent heat to melt a surface layer, and further heating and vaporization occurs. This type of thermal material removal is the essential mechanism of processing metals. In semiconductor and insulators absorption takes place via interband transitions. Intraband transitions in the upper band associated with electronphonon scattering lead to material heating. In addition impurity centers, excitons, and defect states provide additional absorption centers for photon energies below the bandgap energy. Table 8 shows the fraction (1 R) and the absorption coefcient for some semiconductors. A different mechanism dominates the interaction of UV laser light and organic materials such as polymers, for example. The absorption of UV photons leads directly to bond breaking and to photodissociation of one or more components of the material. The resulting gas and radicals are ejected from the material surface. Unlike the thermal processes previously described, photodissociation does not exhibit a threshold. The incident photon energy only has to exceed the bond dissociation energy Eb. Table 9 gives some examples. At high enough input intensities, multiphoton (n-photon) absorption occurs which leads to photodissociation if n Eb. Material Removal (Ablation). In media where photothermal effects play the major role, laser-controlled material removal proceeds via melting and evaporating surface layers. To obtain the surface temperature, one has to solve Eq. (15) where the source term is determined by the absorbed laser intensity, S (1 R)Iin(r, t)ez. Because the lateral dimensions are
Table 7. Fraction of the Incident Fluence (Incident Minus Reected) Absorbed in a Thin Surface Layer Material Ag Au Co Fe Ni Ti
Data from (18).
Table 8. Fraction of Incident Fluence Absorbed in a Thin Surface Layer (1 R) and Absorption Coefcient of Some Semiconductors at Typical Excimer Laser Wavelengths 1R Material Ge Si GaP GaAs PbS InSb
Data from (18).
193 nm 308 nm 1.50 1.54 0.88 0.78 0.92 1.50
(106 cm1) 248 nm 1.62 1.81 1.84 2.07 0.82 1.24
308 nm 0.89 0.63 0.51 0.59 0.58 0.47
248 nm 0.74 0.66 0.60 0.67 0.54 0.60
193 nm 0.74 0.76 0.64 0.73 0.64 0.64
308 nm 0.44 0.41 0.55 0.58 0.53 0.39
248 nm 0.35 0.33 0.42 0.33 0.62 0.46
0.82
188
EXCIMER LASERS
Table 9. Dissociation Energy and Corresponding Photon Wavelengths of Some Bonds Bond H2 O2 CO CUC CuC CIC CUH
Data from (18).
Eb (eV) 4.48 5.12 11.09 3.62 6.4 8.44 4.30
(nm)
277 243 112 343 194 147 289
(1)
Energy
(2)
Interatomic distance
much larger than the absorption lengths 1, 2 (d2 / dz2) in Eq. (15). For a uniform incident intensity Iin(x, y, z 0) I0 and (t)1/2 1, Eq. (15) is then solved analytically and yields the temperature prole. For the temperature increase at the material surface we nd, see for example, (17), that T ( z = 0, t ) = 2 (1 R )I0 K t
1 /2
Figure 9. Simplied picture of the two channels [photochemical (1) and photothermal (2)] that initiate ablation processes after excitation of the electronic system of a polymer.
(16)
For a numerical example let us assume that t 20 ns (typical excimer pulse duration at 248 nm), a gold target with (1 R) 0.66, 1.3 cm2 /s, K 3 W/cm/K, and 106 cm1. With these numbers we obtain a surface temperature increase of about 2 K per mJ/cm2. A 100 mJ excimer laser pulse focused on a spot of 1 cm2 would thus increase the surface temperature by about 200 K. After the melting temperature is reached, continuous energy supply leads to melting, additional heating, and evaporation. From thermodynamical considerations [see, e.g. (18)], maximum ablation rates in metals are about 0.01 m/s at sufciently high excitative energy densities. For typical excimer laser pulses (20 ns) this amounts to about one atomic layer per pulse. In reality energy deposition is more complex and observed ablation rates are considerably larger. Depending on the target material, they reach values of several m per pulse at incident uences of several J/cm2. Several processes occur simultaneously during the incident laser pulse and affect the absorption process and material removal. For example, a mechanical reaction of the expanding laserinduced vapor with the molten material can lead to rapid
removal of the liquid. In this case, the ablation depth corresponds roughly to the material depth in which the melting temperature is reached. The material removal initially increases with increasing laser uence, reaches a maximum, and then decreases at higher uence. This effect is attributed to laser-supported detonation. The laser heats the vapor which in turn expands rapidly, preventing further efcient energy deposition into the target. In organic materials, material removal with ultraviolet laser pulses proceeds witout extensive heating (photoablation). The UV photon energy is large enough to break bonds. The fragments are initially under high pressure and are ejected from the surface thereby carrying away the deposited energy as kinetic energy. This cold material removal takes place within one absorption length and provides a controllable tool for material removal. Because the surrounding material is hardly affected, the ablated area distinguishes itself by sharp boundaries. Figure 8 shows a schematic diagram of the ablation process involving an interplay of photothermal and photochemical processes. As mentioned above, ablation is a very complex process, and many questions still remain to be explored. Fig-
UV laser beam
Figure 8. Schematic diagram of the ablation process. (a) Energy deposition (excitation/heating) in an absorbing surface layer. (b) Bond breaking (direct deposition/thermal vibrational excitation) and explosive expansion. The ejected particles leave the surface nearly perpendicularly with velocities exceeding 2000 m/s. (c) A hot cloud of ablated material with neutral and ionized particles is produced and expands.
;
Mask (a)
Beam homogenizer Aperture Condenser Reticle 1st

+1st Projection lens Wafer (a)
1st
0th
(b)
(c)
(b)
Figure 10. Stepper illumination arrangement for deep UV lithography. (a) conventional illumination; (b) illumination with an annular aperture. For a given projection lens, the application of an annular aperture increases the effective numerical aperture of the system compared with conventional illumination.
EXCIMER LASERS
189
Phase shifter Mask Electric field
Table 11. Data for Commercial Excimer Lasers Used for Deep UV Lithography KrF Laser efciency (broadband) Narrowing efciency Repetition rate Output power Tuning range Bandwidth Wavelength stability Energy uctuations 2.5% 30% 1 kHz 10 W 248.4 0.15 nm 0.60.8 pm 0.1 pm 3% ArF 1.3% 10% 600 Hz 5W 193.35 0.15 nm 0.7 pm 0.1 pm 5%
Intensity
Figure 11. Schematic diagram of the function of a conventional and a phase-shifting mask. Destructive interference between the elds from two adjacent apertures avoids undesired intensity maxima in the shadow region.
ure 9 is an example of how photochemical (dissociation) and photothermal processes occur in one and the same type of species (a polymer in this case) after UV photon absorption. Channel 1 is the direct photochemical decomposition starting from the excited electronic state without thermal effects. Channel 2 proceeds via relaxation through the vibronic ladder (after internal conversion) of the ground state heating the sample to the melting point. Pure photochemical ablation, due to the absence of heating, results in steeper edges and allows for better resolution. It should also be mentioned that incubation effects exist in organic materials with relatively small absorption coefcients, that is, the ablation does not start with the rst illumination pulse but only after a certain number of incubation pulses are absorbed. Application Examples Excimer lasers are broadly applied in various elds ranging from fundamental research to manufacturing. Industrial and Medical Applications. Industrial applications of excimer lasers involve ablation, UV microlithography, micromachining, annealing, metal deposition, and laser cleaning to name a few of the most important areas. Excimer Laser Assisted Chemical Etching. The idea behind laser assisted chemical etching is to efciently remove material layers in the presence of a precursor gas that chemically reacts with the material. Depending on the gas and material, several processes are feasible. They involve the formation of radicals in the gas after UV photon absorption and subsequent chemical reaction with the material and desorption. The latter can also be enhanced by the UV laser. Compared with ablation the necessary uences are considerably smaller. Submicron spatial resolution is possible by controlling the intensity prole of the incident laser beam. Deep UV Lithography with Excimer Lasers. In photolithography for microelectronics, a mask is demagnied and imaged onto a photoresist, typically a polymer, which is then chemi-
cally processed to produce the desired structures. The excimer laser is the radiation source for a high-resolution optical projection aligner, the so-called wafer stepper or scanner. Excimer lasers are an ideal tool for photolithographic purposes for several reasons. The short wavelengths provide potential resolutions of better than 100 nm. The high pulse uences allow for short exposure times and even simultaneous illumination of large areas (wafers) without scanning. For the next few years excimer lasers will remain the main tool in the microelectronic industry for the fabrication of integrated circuits, notably microprocessors and memory chips. Because of the immense importance of this technology we will go into somewhat more detail here. Resolution and Depth of Focus. The performance of a photolithographic system is largely determined by depth of focus and lateral resolution. The resolution or minimum structure dimension that can be realized is d= d NA (17)
where is the illumination wavelength and NA is the numerical aperture of the projection lens or system, and d is a empirical factor depending on the optical system and the resist material. The depth of focus is given by z= z (NA)2 (18)
where z is an empirical factor characteristic for a certain projection optics and resist. Typical imaging systems have numerical apertures of 0.5 to 0.6. The eld sizes of lenses have
Line narrowing and tuning unit
Output Laser head
Power supply
Table 10. Structural Sizes Achieved by Photolithography with Excimer Lasers Wavelength 248 nm 193 nm Regular Structures (memory chips) 0.2 m 0.130.18 m Random Structures (microprocessors) 0.25 m 0.18 m
Gas handling system
Pulse energy monitor
Computer
Wave-meter
Figure 12. Schematic diagram of a commercial excimer laser used for industrial photolithography.
190
EXCIMER LASERS
Figure 13. Scanning electron micrographs of structures cut into a human hair using excimer laser micromachining. (Courtesy Lambda Physik GmbH.)
been increased from 10 to 20 mm2. The -factors (a typical value for 248 nm and 193 nm lithography is 0.5) can be lowered through: (a) the application of a high contrast photoresist, (b) the use of annular illumination (see Fig. 10), and (c) the employment of phase-shifted masks (see Fig. 11). To homogenize the beam, different spatial components are mixed. This, for example, can be achieved by a system of ys eye lenses or a fused silica rod with highly reecting walls. With deep UV lithography employing KrF and ArF excimer lasers structures with dimensions of less than 250 nm are possible. With ArF lithography (193 nm) the production of 4 and even 16 Gbit memory chips seems to be feasible. Table 10 summarizes the achievable resolutions. Projection Lenses. There are two different types of projection systemsall refractive optics, and catadioptic reectiverefractive systems. All-refractive imaging systems consist only of lenses. The most critical type of aberration is chro-
matic aberration. To reduce this effect very narrow-band excimer lasers are necessary. For demagnication factors of 5 to 10 all refractive optics (made from fused silica) require laser bandwidths of 0.8 pm at 248 nm and 0.3 pm at 193 nm. At 193 nm a reduction of chromatic aberration can be achieved by using a combination of fused silica and CaF2 lenses which can accomodate bandwidths of 0.7 pm. The conditions are further relaxed for catadioptic systems which call for 100 pm at 248 nm and 20 pm at 193 nm. Catadioptic systems are made of reecting components only and thus exhibit smaller chromatic aberrations. Table 11 summarizes the current status of narrowband excimer lasers which are suitable for all-refractive projection systems which is the most widely used concept today. The spectral narrowing is achieved through prisms and gratings as detailed in the section on resonators. A schematic diagram of a state-of-the-art excimer laser used for lithography is shown in Fig. 12.
EXCIMER LASERS
191
Laser Mask Cornea

Figure 14. Schematic diagram of photorefractive keratectomy. Depending on the mask structure, the central or outer parts of the cornea are illuminated with excimer laser radiation. The subsequent ablation leads to a reshaping of the cornea, that is, to a change in its radius of curvature. This allows correcting near and farsightedness.
Before
Laser Hardening and Annealing. Hardening steel surfaces requires heating above the martensitic transition point and subsequent quenching by cooling. A short laser pulse is ideal to locally excite (heat) a thin surface layer. Then the cooling is accomplished by rapid heat diffusion into the bulk material. Annealing of semiconductors to heal structural damage from ion implantation is of great concern in fabricating electronic components. Again the material is heated by a laser pulse up to a temperature where the atoms rearrange. Depending on the duration of the illuminating pulse, this process can occur before actual melting takes place. Laser Vapor Deposition. Laser deposition is very attractive for generating high-quality, thin lms. The idea is to focus the excimer laser on a target composed of the material to be deposited on a substrate. The excimer laser pulse causes ablation leading to a plume of gas that propagates to the substrate where it is adsorbed as a thin lm. Micromachining. Material removal with excimer lasers is widely applied in micromachining. There are two major techniques used. One is to place a mask in contact with the object to be processed. Relatively large areas can be processed simultaneously unlike with Nd:YAG or CO2 laser where scanning is usually applied. Pulse uence and number of pulses determine the amount of material removal. An alternative schema involves a high-quality projector that demagnies a mask onto the material surface by an excimer laser beam. Precise drilling and cutting with submicron resolution is possible. Two examples that demonstrate this are shown in Fig. 13. Excimer Laser Application in Medicine. Excimer lasers are increasingly applied in various elds of medicine. One of the
most advanced developments is in opthalmology in corneal surgery. The controllable and localized material removal provided by excimer laser radiation permits a predened reshaping of the cornea to correct vision defects such as farsightedness and nearsightedness (see, for example (19). A schematic diagram is shown in Fig. 14. With ArF excimer lasers, ablation depths per pulse range from 0.01 m at 50 mJ/cm2 to 1 m for ows above 700 mJ/cm2 (19). The absorption of UV light takes place mostly in the peptide bonds of collagen molecules, the most abundant organic molecule in the corneal tissue. This bond occurs in concentrations of about 1021 cm3 and has a molar extinction coefcient of 5.5 103 cm1M1. Bond breaking is followed by a microexplosion on the corneal surface leading to ejection of debris. Heating of the cornea is minor. Cell membranes and other proteinbased organic materials shield the cell nuclei from UV radiation and prevent potential damage (mutation) to the DNA. Applications in Science Pump Source for Other Lasers. Excimer lasers are used uniquely in fundamental and applied science as a source of powerful pulses in the UV. Because many molecules have broad absorption bands in the UV, excimer lasers are used to pump other lasers, notably dye lasers. A schematic diagram is shown in Fig. 15(a). A cell lled with an organic dye solution is pumped transversely by an excimer laser. The resonator is formed by an outcoupling mirror and a grating. The latter serves for frequency tuning and bandwidth narrowing. To increase the resolution, the beam is expanded before impinging on the grating. If higher output powers are required, such an oscillator is supplemented by a second dye cell acting as an amplier. Many different oscillator congurations have been developed depending on the desired output parameters. A few different dyes cover the whole spectral range from the near-UV to the visible to the near-infrared [see Fig. 15(b)]. Such tunable dye laser systems have been broadly applied in high-resolution spectroscopy (20). In recent years though they have been replaced by solid-state laser-pumped optical parametric oscillators (OPO). Typical pulse durations are a few ns with line widths of a few hundred megahertz. Systems based on short-cavity and distributed-feedback dye lasers have been developed to generate pulses as short as a few hundred femtoseconds (21). Frequency-doubled femtosecond pulses at 308 nm and 248 nm have been amplied in XeCl and KrF excimer
Laser beam
DPS Coumarin 120 Coumarin 102 Coumarin 307 Coumarin 153
Styryl 20 Styryl 14 HITCI Rhodamine 800 Pyridine 2
Cylindrical lens Dye cell
Rhodamine B DCM 0.4 Telescope (a) Grating 0.7 Wavelength (b)
1.0
Figure 15. (a) Schematic diagram of an excimer laser pumped dye laser. (b) A few dyes pumped by a XeCl excimer laser at 308 nm cover the spectral range from the near IR to the near UV. Data from (24).
192
EXCITATION CONTROL IN POWER SYSTEMS 17. H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids. Oxford: Oxford University Press, 1986. 18. W. W. Duley, UV Lasers: Effects and Applications in Materials Science, Cambridge: Cambridge University Press, 1996. 19. G. H. Pettit, M. N. Ediger, and R. B. Weiblinger, Excimer laser ablation of the cornea, Opt. Eng., 34: 661667, 1994. 20. W. Demtro der, Laser Spectroscopy, Heidelberg: Springer, 1996. 21. S. Szatmari and F. P. Schaefer, Subpicosecond widely tunable distributed feedback dye laser, Appl. Phys., B46: 305311, 1988. 22. J.-C. Diels and W. Rudolph, Ultrashort Laser Pulse Phenomena: Fundamentals, Techniques, and Applications on a Femtosecond Time Scale. San Diego: Academic Press, 1996. 23. Lambda Physik Excimer Lasers: The UV solution for scientic, medical and industrial applications, 1997, Lambda Physik GmbH, Technical Information. 24. U. Brackmann, Lambdachrome Laser Dyes, Lambda Physik GmbH, Goettingen, 1994.
gain modules leading directly to femtosecond pulses with millijoule energies in the UV with pulse durations as short as 50 fs (see, for example, (22)). Such pulses are used in timeresolved spectroscopy and to generate short pulses in the vacuum UV and X-ray spectral range, to name just a few applications. Spectroscopy with Excimer Lasers. Because of large photon energies, excimer lasers are well suited to excite and ionize electronic transitions in many organic and inorganic molecules. If excitation is followed by uorescence, sensitive detection schemes can be applied that allow detection of small concentrations of species. Such methods are important for environmental remote sensing, for example. If excitation leads to ionization, mass spectroscopy is applied to identify the photofragments with high sensitivity. Application elds are surface analysis (desorption) and cluster physics, to name just a few areas. BIBLIOGRAPHY
1. N. G. Basov et al., Laser operating in the vacuum region of the spectrum by excitation of liquid xenon with an electron beam, JETP Lett., 12: 329, 1970. 2. H. A. Koehler et al., Stimulated VUV emission in high-pressure xenon excited by high-current relativistic electron beams, Appl. Phys. Lett., 21: 198, 1972. 3. P. W. Hoff, J. C. Swingle, and C. K. Rhodes, Observation of stimulated emission from high-pressure krypton and argon/xenon mixtures, Appl. Phys. Lett., 23: 245, 1973. 4. S. K. Searls and G. A. Hart, Stimulated emission at 281.8 nm from XeBr, Appl. Phys. Lett., 27: 243, 1975. 5. Ch. K. Rhodes (ed.), Excimer Lasers. Berlin, Heidelberg, New York, Tokyo: Springer Verlag, 1984. 6. M. H. R. Hutchinson, Excimer lasers, in L. F. Mollenauer and J. C. White, (eds.), Excimer Lasers. Berlin: Springer Verlag, 1987. 7. C. A. Brau, Rare gas halogen excimers, in C. K. Rhodes, (ed.), Excimer Lasers, Berlin: Springer, 1984. 8. U. Rebhan and D. Basting, Excimer lasers: Current status and future developments, Ber. Bunsenges. Phys. Chem., 97: 1504, 1993. 9. M. Obara and F. Kannari, Rare gas-halide lasers, in R. A. Meyers (ed.), Encyclopedia of Lasers and Optical Technology, San Diego, CA: Academic Press, 1991, pp. 568586. 10. A. M. Hunter II, R. O. Hunter, and T. R. Johnson, Scaling of KrF lasers for inertial connement fusion, IEEE J. Quant. Electron., QE-22: 386, 1986. 11. C. P. Christensen et al., High-repetition-rate XeCl waveguide laser without gas ow, Opt. Lett., 12: 169, 1987. 12. T. J. McKee, B. P. Stoicheff, and S. C. Wallace, Diffraction-limited KrF and XeF lasers with a negative branch unstable resonator, Appl. Phys. Lett., 30: 278, 1977. 13. H. Endert, R. Paetzel, and D. Basting, New KrF and ArF excimer lasers for DUV lithography, Microelectron. Eng., 27: 221, 1995. 14. J. Kleinschmidt et al., Extremely narrow-bandwidth, high-repetition rate laser for high NA step and scan tools, SPIE 1996 Int. Symp. Microlithography, Santa Clara, 1996. 15. J. Goldhar, M. W. Taylor, and J. R. Murray, An efcient doublepass Raman amplier with pump intensity averaging in a light guide, IEEE J. Quantum Electron., QE-20: 772, 1984. 16. J. C. White, Stimulated Raman scattering, in L. F. Mollenauer and J. C. White, (ed.), Tunable Lasers. Berlin: Springer-Verlag, 1987.
J. W. NICHOLSON W. RUDOLPH
University of New Mexico
J. KLEINSCHMIDT
Lambda Physik GmbH
}{{}}
HOME
q
ABOUT US
q
CONTACT US
q
HELP
q

Free Electron Lasers Standard Article Dmitri E. Nikonov1, Gershon Kurizki2, Yuri V. Rostovtsev3 1University of California at Santa Barbara, Santa Barbara, California 2Max-Planck-Institut fr Quantenoptik, Garching, Germany 3Texas A&M University, College Station, Texas Copyright 1999 by John Wiley & Sons, Inc. All rights reserved. DOI: 10.1002/047134608X.W6307 Article Online Posting Date: December 27, 1999 Abstract | Full Text: HTML PDF (210K)
q q
Browse this title

q

r
r r
Abstract
The sections in this article are Electron Kinematics: Resonance and Synchronism Conditions Classical Electron Dynamics Dynamics and Gain in Interfering Two-Wiggler Fels Quantum Regime of FEL FEL Experimental Parameters Applications Perspectives About Wiley InterScience | About Wiley | Privacy | Terms & Conditions Copyright 1999-2008John Wiley & Sons, Inc. All Rights Reserved.
716
FREE ELECTRON LASERS

The free-electron laser (FEL) is a device that uses part of the kinetic energy of nearly-free electrons (not bound in atoms or in condensed matter) to generate coherent electromagnetic radiation, see Refs. 13. The electrons are supplied in the form of a beam accelerated to relativistic velocities. The electron beam can either pass through the gain medium only once, or be recycled in a storage ring, which can then allow the electrons to circulate many times through the FEL. Electrons are not entirely free: as will be explained here, free electrons cannot interact efciently with radiation in vacuum. Thus, they will interact with radiation in two types of structures: (1) a device in which electrons are accelerated in an inhomogeneous (periodic) magnetic or electromagnetic eld, called a wiggler or an undulator; and (2) a device in which electrons are unperturbed but the laser wave is subject erenkov transition radiation or Smithto dispersion (as in C Purcell devices). Because the FEL is a laser, it is based on stimulated emission in which radiation is mostly emitted coherently, that is, with the same phase as already existing radiation. For this, the interaction region is enclosed in a laser cavity (see CAVITY RESONATORS) to allow the emitted light to be fed back and to stimulate further emission. If the light passes many times through the gain medium, FEL can operate even in a small-gain regime. Because mirrors with sufcient reectivity are not available for the ultraviolet (UV) and shorter wavelength range, FELs in this range are designed without mirrors on the principle of amplied spontaneous emission, such that radiation from one part of the electron beam or an injected signal stimulates radiation from other parts, passing them only one time. The large-gain regime of a FEL is necessary for this kind of operation (4). Sometimes SPONTANEOUS EMISSION of free electrons is used (termed indulator radiation) (5). Let us consider, as an example, a FEL with a static magnetic wiggler (Fig. 1). The laser has frequency and wavelength L with the corresponding wavevector kL 2 / L. The wiggler has spatial period W and wavevector kW 2 / L. Then the combined wave of the laser and the wiggler eld (called ponderomotive potential) has the phase (kL kW)z
t. In order to interact with a permanent rather than an oscillating force, the electron must maintain constant phase relative to the ponderomotive potential. This is called synchronism condition. Then the electron must have a velocity approaching the resonant velocity
vr = kL + kw (1)
Without a wiggler (kw 0) in vacuum ( kLc) the resonant velocity is equal to the speed of light c, unattainable by electrons. Due to the presence of a wiggler or of a refractive index, the resonant velocity is less than the speed of light and the laser operation is then possible. Perhaps the most salient feature of FELs is their tunability. Velocity v of the electron (with mass m) is related to its energy E mc2 via the Lorentz factor (1 (v / c)2)1/2. Denition of the resonant velocity from Eq. (1) results a rough determination of the wavelength of the laser L = W (1 + a2 w) 2 2 (2)
x kw Aw y Al kl z
e
Figure 1. Schematic of the wiggler and laser elds (top), and the momentum change in the processes of emission and absorption (bottom).
where aw is a parameter depending on the wiggler eld (see below), usually of the order of unity. Electron energy from MeV to GeV corresponds to Lorentz factors ranging from approximately 2 to 2000. For the wiggler period of the order of a centimeter laser radiation can, in principle, have a wavelength that ranges from microwave to hard X-ray. Another important feature of FELs is their ability to yield large peak power, which scales up with the electron peak current, up to hundreds of amperes for electron pulses produced by presentday accelerators. On the other hand, average power output is much lower. The FEL was rst proposed on the basis of quantum electrodynamics (6). It was later understood that, for FELs emitting in the visible and shorter-wavelength range, quantum effects play a negligible role and their operation could be explained within a classical theory (7,8). In the quantum description (9), FELs owe their gain to the fact that an electron recoils in opposite directions, depending on whether it emits or absorbs a photon with a given wavevector kL; hence, the resonant electronic momentum ker for the emission of such a photon differs from the resonant momentum kar for its absorption. Probabilities of emission and absorption of a photon as functions of the initial electron momentum (lineshapes) are centered at ker and kar, respectively [Fig. 2(a)]. Spontaneous emission has the same lineshape as stimulated emission. The quasiclassical limit holds when ker kar is much smaller than the inverse length of the wiggler, and the photon energy ckL is much smaller than the electron energies E(ke(a)r). In this limit, the gain curve is antisymmetric about the mean resonant momentum k (ker kar)/2 [see Fig. 2(b)], which corresponds to resonant velocity vr. In this limit, the quantum expression for gain coincides with its classical counterpart. In the classical description (13), the wiggler eld will periodically deect the electrons perpendicular to their direction of travel (along the wiggler axis). The small-gain regime occurs whenever it is possible to neglect the amplication of the eld when considering the motion of electrons. In this limit, the oscillations of the electrons in the ponderomotive poten-
717
Gain
kar 0 ker
1 (a) 1
1 Momentum (b)
Figure 2. Gain in a FEL for large recoil results from resolved proles of emission and absorption (a). For small recoil, gain is the difference of overlapping emission and absorption and is antisymmetric.
tial are described by the pendulum equation. It turns out that in such a potential electrons having a velocity higher than vr on the average give energy to the laser, thus contributing to gain; electrons having a velocity lower than vr absorb the energy from the laser, thus contributing to loss. This results in a gain curve, as in Fig. 2(b), which we designate Gst. The combined effect of the wiggler eld and the laser eld, i.e., the ponderomotive potential, causes axial bunching of the electrons. The electrons injected at random times are forced into periodically spaced bunches separated by approximately the laser wavelength. This bunching is associated with the gain or loss of energy by electrons, or, equivalently, their axial acceleration or deceleration, depending on the phase between their transverse motion and the laser wave. If bunching is signicant, so is the change of both laser eld amplitude and phase. This change promotes further bunching. This process is a rare example of useful instability. As a result of their bunching, electrons radiate in-phase and their emitted elds add up coherently, so that the total emitted intensity is proportional to the electron current squared, rather than being proportional to the current, as in the case of randomly distributed electrons. When the radiation of electrons is essentially collective, this is called a large-gain regime. In this large-gain regime, the gain lineshape is no longer antisymmetric. The rst FEL (10) was operated in 1977. A variety of FELs (13) now operate successfully over a spectral range of from millimeter- down to ultraviolet wavelengths. On the other hand, FEL operation in the X-ray and extreme ultraviolet (XUV) domains (11) is still facing considerable difculties, primarily because of the stringent requirements FEL poses on the allowed electron beam energy spread and emittance. (The beam emittance is the product of the transverse size of
the beam and the velocity angle spread.) These requirements stem from the antisymmetric dependence on the small-gain standard gain Gst being dependent upon the deviation of the electron velocity v from the resonant velocity vr. [This is related to the Madey theorem, which states that gain lineshape is proportional to the derivative of the spontaneous emission lineshape over the velocity (12)]. Electrons initially below resonance contribute to absorption and electrons above resonance contribute to emission. Therefore this gain lineshape allows for net gain only if the initial momentum distribution is centered above vr, which we call momentum population inversion. It also restricts the momentum spread, at which gain is signicant, to values comparable to the width of the positive (gain) part of Gst. This width decreases signicantly with laser wavelength, thereby limiting severely FEL gain performance at short wavelengths. A variation of a FEL having two wigglers and a drift region between them (called optical klystron) was realized (13). In this device, the rst wiggler serves to bunch the electron phases, which then acquire favorable values in the drift region between the wigglers, and nally yield enhanced gain in the second wiggler. Its gain lineshape has a higher maximum value, which is an advantage for electron beams with small energy spread. However, the width of the gain region is proportionally narrower, which makes the restrictions for the energy spread even more severe. In search of shorter-wavelength sources, researchers con erenkov radiation by quasi-free elecsidered the stimulated C erenkov transition radiatrons in a refracting medium, or C tion (TR) in a periodic dielectric structure, see Refs. 14,15. erenkov TR can exist even above the plasma frequency The C (corresponding to 30 eV) where the refractive index n 1 erenkov effect is impossible, because TR ocand the usual C curs when an electron crosses a boundary between different refractive indices. In a structure with a spatially periodic in erenkov effect results from the condex of refraction, the C structive interference of TR from different layers. The period of these layers (which plays the role of kW) can be shortened much more than in magnetostatic wigglers. The pursuit of shorter wavelengths has also led to the proposal of a FEL in which the magnetostatic wiggler is replaced by an intense electromagnetic wave (the Compton-scattering FEL) (11,16). Its wavelength (w 1 m and thus L 1 nm) would be much shorter than in existing lasers. In all of the preceding schemes, electron momentum spread and beam emittance have been concluded to be major obstacles in the realization of X-ray FEL. In an attempt to overcome the adverse effects of electron spread on short-wavelength gain, the notion of lasing without inversion (LWI) (17) in atomic systems, namely, the cancellation of absorption by interference in the gain medium, has recently been proposed for FELs (18). These proposed schemes involve a two-wiggler FEL, which bears a limited resemblance to an optical klystron (13). Unlike an optical klystron, bunching resulting from the rst wiggler is reversed and the electrons are given a shift of their phase relative to the ponderomotive potential so as to cancel absorption. In the resulting gain curve, the absorption part below resonance is eliminated, whereas the gain part remains intact. Whereas in an ordinary FEL population inversion of electrons in the momentum domain is required to ensure net gain from a momentum distribution, in the proposed schemes the net gain is
Gain
718
obtainable even from a very broad (inhomogeneous) momentum distribution without population inversion. This creates new possibilities for development of X-ray FELs.
approximate the dispersion curve by a straight line with the slope equal to the initial z-component of the electron velocity v
(0 ) a
(0 ) e
kLz kW v
(7)
ELECTRON KINEMATICS: RESONANCE AND SYNCHRONISM CONDITIONS Quantum kinematics gives a more intuitive view of the FEL gain. An electron enters the interaction region in the initial state ki and energy (the curve in Fig. 3) Ei i mc2 =
2 2 4 p2 ic +m c
(3)
where the momentum p k. After absorption (emission) the electron has momentum ka (ke), which is related to the energy Ea (Ee) as in Eq. (3). There is a mismatch of the longitudinal projection of these momenta from the ones obtained from the momentum conservation (see Fig. 3)
e a
In this quasi-classical approximation the detunings for emission and absorption coincide, where emission and absorption cancel each other and there is no gain. The condition of zero detuning correspond to resonance, at which momentum is conserved precisely. It coincides with the synchronism condition, that v vr. For a magnetostatic wiggler it generalizes the expression (1) erenkov wiggler, it yields the by replacing kL to kLz . For a C erenkov radiation. usual condition for C c/v = n cos (8)
= kiz kez kLz kW = kaz kiz kLz kW
(4) (5)
As the interaction is considered to be stationary, i.e., not bounded in time but happening in a nite region of space, the energy is conserved precisely Ea,e Ei , but the momentum admits some uncertainty given by the wiggler length LW
erenkov radiation. This is the usual condition for C To obtain a nonzero contribution to gain, we need to take into account the curvature of the dispersion curve, to second order in the longitudinal momentum variation (kaz kiz). As the laser light propagates at an angle with the axis, there is a corresponding transverse variation of momentum (kax kix) which must be taken into account in the same order in the expansion. Upon combining the two contributions, we obtain unequal values of detunings for emission and absorption (9)
a
}
LW
= =
(6)
R
R,
(9) (10)
} 2
2mv3 3
Momentum and energy transfer kL and kLc from light in vacuum (tilted line in Fig. 3) do not bring the nal state to the dispersion curve of kinematically allowed states and, thus, cannot even approximately satisfy the conservation laws because the speed of light is larger than the electron group velocity v. Therefore, a wiggler with kW or a medium with refractive index n, which modies the light dispersion to n ckL, is needed for either emission or absorption to happen. We decompose the variation of energy with momentum in terms of a Taylor series in (kaz kiz). To this end we rst
n2 sin2 }2 2m vc2
This difference between the emission and absorption detunings is proportional to the recoil of the electrons due to the photons. The rst term in Eq. (10) comes from the longitudinal variation of momentum; the second comes from the transverse variation. In general they have the same order of magnitude. Note that the effective mass for the longitudinal motion M m3 is different from that for the transverse motion M m. The ratio of the shift between the centers of the emission and absorption curves, given by R, to their width L, see Eq. (6) is
R
E +h
hkW
=
=
a
hkL
2 2 c 3 v3
c L 2 3
1+
n2 sin2 2 v2 c2
(11)
(12)
p
Figure 3. Energy and momentum of an electron after emission or absorption: the wiggler momentum kW brings them close to the dispersion curve of a free electron.
Here c /(mc) 4 1013 m is the Compton wavelength of electrons. The regime of operation of the FEL is quantum, if 1, or classical if 1. The parameter reaches unit at wavelength q of the order of several nanometers. Thus, because all FELs currently operate in the classical regime, classical theory is sufcient for their description. The quantum limit will be reached only by X-ray lasers. Examples of gain lineshapes in the quantum and classical regimes are shown in Fig. 2a and b, respectively.
719
CLASSICAL ELECTRON DYNAMICS The classical dynamics of electrons in a FEL (1) is described by the Hamiltonian H mc2 = c
(p eA )2 + m2 c2
(13)
where m is the mass of an electron, e is the charge of an electron, c is the velocity of light, is referred to as the Lorentz factor, p is the canonical momentum, and A AW AL is the vector potential of the combined eld of the wiggler, oriented along the z-axis (designated by subscript W), and the laser eld (designated by subscript L), which propagates at an angle to the axis of the wiggler, i.e., has the wavector kL (kL sin, 0, kL cos), as in Fig. 1. Both elds are y polarized, and is the phase of the laser eld at the instant of the electron entry into the wiggler. AW cos (kW z ) AW = y AL cos (t + kL z cos + kL x sin + )] AL = y (14) (15)
From now on we consider ultrarelativistic electrons ( 1, i.e., v c). There are terms of interaction with the elds having various dependences on the time and the coordinates. Of those, following a standard procedure (19), we drop (in an analog of the rotating wave approximation) the terms that are rapidly oscillating in the frame of reference of an electron moving with the injected velocity vi close to c. The remaining (near-resonant) terms oscillate slowly in this frame of reference. Sometimes lasing occurs at frequencies corresponding to higher harmonics of the electron wiggling motion. This corresponds to the wiggler eld into higher powers, see (1). In this way we obtain the equations of motion for the energy and momenta d = N sin(t + qz z + qx x + ) dt 1 d pz 1 d px mc2 d = = dt qz dt qx dt where N = e2 2AW AL m2 c 2 r (24) (22) (23)
The magnetic eld of the wiggler and the electric eld of the laser are BW = AW , EL = A t L (16)
and y is the unit vector along the y-axis, and is the phase of the laser eld at the instant of the electron entry into the wiggler. Dimensionless potentials (with j L, W) are eA j aj = 2mc (17)
The argument of the sine in Eq. (22) is the phase relative to the ponderomotive potential, = t + qz z + qx x + (25)
Just like any other laser, FEL can operate in multimode regime. Here we consider only a single-mode eld. For details about the mode competition see (19). Hamilton equations determine the derivatives of energy and momenta H d px H d pz H d mc2 = , = , = dt t dt x dt z (18)
Equation (23) expresses the relation between the momentum transfer to the ponderomotive potential (qx kL sin , qz kL cos kW) and the corresponding energy transfer ( kLc). The dynamical equations simplify in the case in which electron energies do not differ much from either the injected energy i or the resonant energy r and the longitudinal coordinate and velocity differ by a small amount from uniform motion with the injected velocity, z vit z and vz vi vz. The equation of motion becomes d = N sin( t + qz z + qx x + ) dt Here = qz vzi qz (vzi vr ) (27) (26)
The Hamiltonian does not depend explicitly on y; therefore, if the initial value of the momentum along y is py(0) 0, it remains zero at all times py(t) 0. Then the wiggling motion in this direction is described by the y component of the velocity H dy eAy = vy = dt py m For the other coordinates dz H H pz dx px = vz , = vx = = dt pz m dt px m (20) (19)
is the detuning of an electron from the resonance with the ponderomotive potential, and vr = qz (28)
In the Hamilton equations substituting these equations back into Eq. (13) we obtain a useful relation 1= v2 1 + 2 c2 (21)
is its velocity corresponding to the resonant energy r. Neglecting the lasing eld compared to the wiggler eld, we obtain that the average square of the transverse velocity is v2 y =
2 2 c aW 2
(29)
720
The equations for the velocity components are

2 d 2 2 qz c d (1 + aW ) 3 dt r dt
To rst order in the lasing amplitude, the coordinates are from Eq. (25) (30) (31) z (1 ) (t ) = x(1 ) (t ) = q z c2 r3 q x c2 r
t 0 t 0
(1 ) (t ) dt (1 ) (t ) dt
(38) (39)
qx c d dvx = dt r dt
2
These equations demonstrate that there is a one-to-one correspondence between the increment of each velocity component and the change of energy. These equations, although describing a two-dimensional motion, result in the one-dimensional pendulum equation for the phase = P sin N (0 ) = (0 ) = P = c2 2 ( 2 q2 + q2 z )(1 + aW ) r3 r x (32) (33) (34) (35)
Expansion of Eq. (26) around the solution of Eq. (37) with the forementioned coordinates gives d (2 ) = PW cos ( t + ) dt
t 0
(1 ) (t ) dt
(40)
When integrated over interaction time T and averaged over the injection phases, the change of energy and, consequently, gain, is nonzero in this order
(2 ) =
PW 2 [ T sin T + 2 cos T 2] 2 3 1 d (sinc2 )| = T /2 Gst ( , T ) T 3 PW 2 8 d
(41)
Small-Signal Small-Gain Regime In general, gain produced by electrons with a well-dened initial energy or longitudinal velocity vz (homogeneous gain) Ghom (vz ) = Jmc2 eIL SL (36)
This gives a gain lineshape similar to that in Fig. 2b. The relative width of the gain curve is 1 = 2NW (42)
2 where J is the current in the electron beam, IL 0c2AL /2 is the intensity of the laser, SL is the effective area of the laser mode, and the average over the uncontrollable injection phase is designated by . . .. In the small-signal approximation, one assumes the laser vector potential being much smaller than the wiggler vector potential. For the effects of a large laser eld amplitude (large-signal gain, or saturation) see (2). In the small-gain approximation one disregards the effect of the change in the laser eld as it propagates through the wiggler on the electron dynamics. In the case of electrons distributed over vz with a normalized distribution function f (vz), the inhomogeneous Ginh(vv) gain is the convolution of f and Ghom. An analytical expression for the gain in the small-gain small-signal regime for a uniform wiggler can be obtained by solving the foregoing equations in a perturbation series in the laser amplitude (in the small parameter N 2T). The travel time in the wiggler T LW / vr. For the effects of a large laser eld amplitude (large-signal gain, or saturation), see Ref. 2. To zeroth order the coordinates z and x vanish, which yields
where NW is the number of wiggler periods, LW NWW. The derivative over the detuning parameter in Eq. (41) is the manifestation of a more general Madey theorem (12) (2 ) = ( ( 1 ) )2 i (43)
where the left-hand side is proportional to gain and the expression under the derivative in the right-hand side is proportional to the energy spread; i is the initial Lorentz factor. The theorem holds for the Hamiltonian motion of a particle in a weak oscillating perturbation (i.e., small-signal, smallgain regime for FELs). It can be shown that the energy spread is proportional to the power of spontaneous emission. The Madey theorem simplies the calculation of gain in nonuniform wigglers. A compact expression for gain per pass is obtainable for 0 by combining Eq. (41) with Eqs. (2) and (28) Gst =
3 2 2 2 NW aW W J d (sinc2 ) r3 SL JA d
(44)
Here JA ec / r0 is the Alfven current and r0 is the classical electron radius. We see that gain rapidly decreases with the increase of r, which adds difculty to achieving short-wavelength lasing. DYNAMICS AND GAIN IN INTERFERING TWO-WIGGLER FELS To focus on the effects of interference of coherent radiation processes, we will consider the interaction in two identical
d (1 ) = N sin( t + ) dt
(37)
When averaged over injection phases, the net change of the energy (and consequently gain) is zero to this order.
721
E-beam
N S N S N kW kL S N S N S Wiggler I
Drift Region
N S N S N kW kL S N S N S Wiggler II
Figure 4. A scheme of realization of a two-wiggler FEL with a drift region between wigglers.
wigglers of length LW with a dispersive drift region of length Ld between them (20) (Fig. 4). Electrons can be guided in the drift region by magnetic eld; light can be deected by mirrors. The effect of the drift region is an addition of phase delay in the second wiggler relative to the rst one, which, on examining Eq. (25), is seen to be = k L sL s e ( v )c xI sin + xII sin v (45)
where se(v) denotes the velocity-dependent electron paths and is the sL denotes the lightwave paths in the drift region, angle of propagation of the laser in the second wiggler and is the angle in the rst wiggler, xI and xII are the transverse coordinates at the exit from the rst wiggler and at the entrance to the second wiggler, and v is the absolute value of the electron velocity, which is not changed in the drift region. The electron oscillates coherently in the ponderomotive potential and, therefore, its oscillations in the two sequential wigglers exhibit interference that depends on the path (or time) difference between the two regions. Then the change of energy in the second wiggler is given by
(1 ) dII
and, with or without the magnetic eld in the drift region, the change of phase grows with velocity, d / dvz 0. Gain of an optical klystron for 5T is shown in Fig. 5. The peak gain can greatly increase compared to that of the usual FEL, but the width of the peak decreases and the whole curve remains an odd function of . The problem with conventional classical interference in an optical klystron, is that it does not distinguish between electrons that emit or absorb energy. The total phase delay, from the entrance to the rst wiggler to the entrance to the second one, remains strongly dispersive (velocity-dependent) and the resulting gain is sensitive to the initial velocity spread. In Ref. (20) we have pursued the following radically new approach to classical interference: (1) the electron beam is separated after the rst region into two components whose velocities correspond to either absorption or emission (on the average) in the rst region; and (2) each of the separated components is given different phase delays that compensate for the velocity spread in . These phase delays ensure both the cancellation of the absorption contributions in the two regions and the doubling of the emission counterparts over a wide range of velocities. To compensate for velocity dispersion, we wish to impose the following phase delay on the beam component, which contributes to emission (with 0, i.e., viz vr), up to the moment it enters the second region ( > 0 ) = 2 N q z ( v z v r )T (51)
where N is an integer. The electrons with 0 that have absorbed (on the average) energy in the rst region must undergo the same function of the phase shift as in Eq. (51), except for an extra phase ( < 0 ) = (2N + 1 ) qz (vz vr )T (52)
dt
(2 ) dII
= N sin( t + + )
)
(1 )
(46) dt (47)
dt
= PW cos ( t + +
Application of Eqs. (40) and (47) yields the phase-averaged energy change in the whole FEL
Note that depends on the velocity at the entrance to the rst wiggler rather than the exit from it. For such a situation, the Madey theorem applies only in a modied form. To implement such a delay function, let us examine more closely the velocity changes in the rst wiggler. From Eqs. (30) and (31) we nd qx r2 dvx = r2 sin dvz qz (53)
(2 ) =
PW 2 [2 T sin T + 4 cos T 4 2 3 + 2 T sin(2 T + ) 2 T sin( T + + 2 cos + 2 cos (2 T +
(48) )]
0.4 0.2
Gain
) 4 cos ( T +
For the case of a usual FEL ( 0), this gives the wellknown expression Gst(, 2T). In the case of the optical klystron, the propagation angle in the rst wiggler 0, and the light goes straight sL Ld. If there is just a transverse magnetic eld Ay in the drift region, the slope of the electron trajectory is dy eAy = dz mvz Then for a small slope L se dse d dvz vz (50) (49)
0 0.2 0.4 20
10
0 Detuning
10
20
Figure 5. Gain in an optical klystron as a function of detuning. Peak gain is higher.
722
vx
vz
1 2
ing to qz(vz vr)T, the smooth part of the delay function [see Eq. (51)]. In addition, the electrons with vx of that in Eq. (55) will be sent to a region of magnetic eld with sharp boundaries, where they travel on additional path corresponding to the phase . This implements the step-like part of the phase delay of Eq. (52). When the selective phase delay, determined by Eqs. (51) and (52), is used in the expression for gain Eq. (48), it results indeed in cancellation (destructive interference) of the absorptive contributions from the two regions and addition (constructive interference) of their emission counterparts GFELWI ( , 2T ) = 4Gst ( , T ) ( ) (56)
Figure 6. Changes of the transverse and the longitudinal velocities are proportional. Open dotsinitial states, closed dotsnal states.
the approximate equality corresponding to a small angle and kL kW. The amount of velocity change is determined by the detuning and the injection phase, but the changes of the vz and vx velocity components are proportional to each other. Integrating Eq. (53), we see that vx = r2 sin (vz vzi ) (54)
where is the Heaviside step function. This gain function, which is positive nearly everywhere, essentially does not require population inversion and yields gain even from broad inhomogeneous distributions (Fig. 7). We therefore designate it as FEL without inversion (FELWI). Field Dynamics in Large-Gain Regime Up to now only the small-gain regime (neglecting the change of laser eld) of FEL was considered. We describe the variation of the laser eld by Maxwells equations for the transverse part of the eld 2 1 2 c2 t 2 Ay = 0 Jy (57)
Hence the transverse velocity after the rst wiggler is correlated to the change in the longitudinal velocity. It is thus possible to distinguish by their vx value those electrons that experienced net emission from those that experienced net absorption. As seen from Fig. 6, electrons with initial velocity below resonance (vzi vr) end up in the half-plane above the line vx = r2 sin (vz vr ) (55)
where the transverse current density is Jy = e

j
vy (x x j (t ))
(58)
Electrons with initial velocity higher than the resonant one (vzi vr) are now below this line. Then the step-like change in the phase delay from 0 to needs to be arranged along this line, as it is shown in Fig. 7. The electrons will enter the drift region at different angles depending on their transverse velocity. By their deection in a magnetic eld, they will receive a phase delay correspond-
Here the sum runs over all electrons and vy is obtained from (19). We write the equations only for phasors of the vector potentials AWy = AW (z ) exp(ikW z ) ALy = AL (z , t ) exp(it + ikL z + i) (59) (60)
0.035 0.03 0.025 0.02

Gain
vx
vz
0.015 0.01 0.005 0
We make the slowly-varying envelope approximation, i.e., assume that AL(z, t) varies little over a wavelength or the time period of optical oscillations in the direction of propagation z z cos x sin. In this approximation one neglects the second derivatives of the envelope. Besides, the right-hand side should also be averaged (designated by . . .) over several wavelengths, which reduces to the above averaging over for a constant envelope. Thus (57) becomes 1 + z c t AL = 2 p 2ic AW e i 1 + AL (61)
0.005 10
where the plasma frequency

5 0 5 10 15 Detuning 20 25 30
2 p =
Figure 7. The implementation of FELWI using both longitudinal and transverse components of the velocity (upper right-hand corner); numerical result for the gain; N 2T 0.03.
e2 n 0m
(62)
and n is the density of electrons.
723
0.5
0.5
0.5
0.5
Figure 8. Motion of electrons in the coor of the pendulum. dinates ,
If the density is high, we need to correct the equation of motion for the electrons for the electric eld created by a nonuniform charge distribution, following (4) 2 2 d p = ( cos sin sin cos ) dt The dimensionless parameter (4) = 1 aW p 4ckW
2 /3
(63)
(64)
is the indicator whether the plasma effects due to Coulomb interaction between electrons are important ( 1, called Raman regime) or whether one can consider electrons interacting directly with the eld only ( 1, called Compton regime). One can see from the pendulum Eqs. (3235) that if the change of the laser eld is small, the gain will saturate when the electrons will make about one cycle in the pendulum coordinates, i.e., P N T 2 2. To offset this effect, tapered wigglers (with variable wiggler wavelength and/or eld) are used (2). Even in the Compton regime, light emission can be collective. If the gain is large enough, the electrons bunch in the pendulum coordinates and correspondingly in space over the length of order of L, see Fig. 8. This is expressed as a nonzero average phase factor ei. The eld adjusts its phase so as to cause the bunched electrons to give even more energy to the eld. Thus the gain is much larger than expected from a small-gain analysis in Section III. This collective effect corresponds to superradiant emission. Description in terms of collective variables, such as bunching, is sometimes possible (4). FELWI Versus Ordinary FEL in the Large-Gain Regime We can solve numerically the set of Eqs. (3235) together with Eqs. (61) and (63) to investigate the electron beam behavior and the gain in a large-gain FEL. They will be compared to their counterparts in a large-gain FELWI, using the scheme of Sec. IV. To this end, we substitute d vr + dt z t (65)
and calculate the space dependence of the eld and beam parameters. We will vary the dimensionless size of the wiggler W 2kWLW (which corresponds to either the change of the L current of electron beam or the length of the wigglers). Numerical results for the monoenergetic beam can be checked against the available analytical ones in the smallsignal small-gain regime. Figure 9 shows the comparison of the results obtained by a computer simulation for the ordinary FEL and for the FELWI. W 0.5) the results practically For a small a current (L coincide with analytical calculations (20), namely, the integral over detunings equals to zero for the ordinary FEL, but W is nonzero for the FELWI. At a slightly higher current (L 1) the nonzero integral over detuning appears for an ordinary FEL as well due to a nonlinear synchronization of the electron bunching and the phase of the ponderomotive potential. We see that the peak gain for an ordinary FEL is higher than for FELWI, but this situation is reversed in the large-signal regime.
2.6
Gain max
IFEL 0.1 FEL a 0 5 Laser intensity 10
2.6
Gain max
IFEL 0.1 FEL 0 5 10
Figure 9. The dependence of maximum gain on the laser eld intensity (a) and on the width of electron spread (b) for the ordinary FEL W 1. (solid line) and for the FELWI (dashed line). Wiggler length, L
724
The dependence of gain on the laser eld intensity is very important because it determines maximum laser intensity. As the laser intensity grows, the electrons perform more than one revolution in their pendulum-like phase space, and start contributing negatively to gain. This decrease of the net gain is referred to as saturation. In Fig. 9(a), we present the dependence of maximum gain on the laser intensity for the IFEL and the FEL. It is clearly seen that only for small intensities of the laser eld does the FEL gain exceed that of the IFEL. As the laser intensity grows, the IFEL gain exceeds that of an ordinary FEL. That means that an IFEL promises a higher saturated laser intensity. Let us assume that the electron momentum distribution in Gaussian with mean value c and variance el. In Fig. 9(b), we present the dependence of the maximum gain on the width el of the electron momentum spread. The maximum gain for an ordinary FEL dramatically drops with the inW crease of el, and we can remark that even for a large L small-gain regime conditions are valid. The maximum gain drops much faster for the ordinary FEL, namely, (Gmax )FEL than for a FELWI (Gmax )FELWI 1
el
and after emission or absorption are as in Section I. The second term in the brackets is applicable in a magnetostatic wig herenkov wiggler. gler, and the rst term in a C Gain is proportional to the difference between the squares of the foregoing two expressions. The emission amplitude in an interaction region of length LW is given by the integral (which we take to extend to innity in the transverse directions)
Te =
LW eA L ki exp(iki r ) d 3 r exp (ike r ) exp (iq j r )}e m V 0 eA L j }ki sin (kix kex qx )(kiy key qy ) = m L (70) LW
exp[i(kix kez q jz )z] dz

e L) eL
exp(i
and, analogously, the amplitude for absorption is given by Tu C exp(i

a L) aL
1
2 el
(66)
(71)
Here we introduced the detunings and the coupling constant (67)

e
= kiz kez q jz ,
= kaz kiz q jz
(72)
Thus we can conclude from the preceding considerations that, due to absorption cancellation at negative detunings, the FELWI has a higher gain for the electron beam with a spread of momentum; this gives us a powerful way to extend the FEL to the short-wavelength region up to VUV and X-ray. QUANTUM REGIME OF FEL In order to calculate the emission and absorption amplitudes quantum mechanically, the phasor of the vector potential in the magnetostatic wiggler (59,60), is used in its time-independent form. Likewise, in the Cerenkov wiggler, where the index of refraction periodically changes along the axis, we use the time-independent electromagnetic vector potential
The coupling constant C= eAL1 }ki sin m (Cerenkov ) (73)
For the magnetostatic wiggler, only the coupling constant changes C= e2 AL AW m (magnetostatic) (74)
Hence, the probabilities of emission and absorption are Me ( Ma ( where sinc(x ) = sin(x ) x (77)
e)
AL = e
j=0
= |C|2 sinc2 = |C|2 sinc2
eL 2 aL
(75) (76)
AL j exp[i(kL + j kW )r]
(68)
a)
(cos , 0, sin ), and the harmonics are deterwhere e mined by the inverse period of index variation kW, with amplitudes ALj. We will consider only the rst harmonic for simplicity. The amplitudes for emission and absorption can then be written in the general form (9)
eA L (p eAW Te = ke | )|ki m eAL (p eAW )|ki Ta = ka | m
(69)
The standard homogeneous quantum gain Gqst is proportional to the difference between the emission and absorption rates, Gqst sinc2 (
e L/2 )
sinc2 (
a L/2 )
(78)
Here ki is a quantum state of an electron normalized in the volume V, and the momenta and energies of electrons before
where e(a) is determined by Eq. (4) and Fig. (3). In the limit of small recoil (small difference beween a and e) it has the

Table 1. Parameters of Some of the First FELs Location Stanford TRW/Stanford Novosibirsk Orsay Santa Barbara Livermore Frascati Los Alamos Boeing Type Superconducting, RF linac Permanent, RF linac Klystron, storage ring Klystron, storage ring Permanent, electrostatic Permanent, induction linac Electro-pulsed, microtron Permanent, RF linac Permanent, RF linac
725
W, cm
3.3 3.6 6.9 7.8 3.6 9.8 2.4 2.73 2.2
N 160 153 22 17 160 30 50 37 229
aW 0.71 0.97 2.7 2 0.11 2.5 1 0.56 1.3
85 130 686 432 6.8 6.9 42 43 223
I, W 2.6 2.5 7 1.3 1.25 850 2.4 50 100
L , m
3.4 1.6 0.62 0.4630.655 400 8700 10.6 10 0.5
same lineshape as in Fig. 2b. The gain can be approximated by Gqst M( ) (79)
which is a restatement of the Madey theorem. The gain prole in Eq. (78) is almost antisymmetric about 0, resulting in a very weak gain for a broad, nearly symmetric electron distribution f (). FEL EXPERIMENTAL PARAMETERS Parameters for a set of early experiments on FELs are collected in Table 1. Here IW is the average output power of the laser. Other parameters are dened in the text. The second column in Table 1 shows the type of the wiggler and the source of electrons. Usual sources of the electron beam are a linear accelerator (RF linac), Fig. 10, or a storage ring. Sometimes RF recovery is used in a storage ring to restore the energy of electrons. APPLICATIONS Free electron lasers have numerous applications. Due to their wide tunability and at the same time high peak power, they
are used in precision and nonlinear spectroscopy (especially in infrared (IR) and UV regions) for the purposes of condensed matter physics and chemistry. Other uses include medical and surgical applications, microcircuit fabrication, material processing, and directed energy weapons. For a review see Refs. 13. PERSPECTIVES The main direction of further development of FELs will probably be in the achievement of an X-ray wavelength via the large-gain regime (4). As a result, the problems of electron energy spread and emittance will have to be solved. Theoretical constructs of a collective atomic recoil laser (21) are expected to bridge the gap between the usual lasers and FELs. Studies on optimization of a conventional FEL design (see, for example, Ref. 22), including the effects of saturation and FEL geometry, are also underway at this time. BIBLIOGRAPHY
1. C. A. Brau, Free Electron Lasers, New York: Academic Press, 1990. 2. P. Luchini and H. Motz, Undulators and Free-Electron Lasers, Oxford: Clarendon, 1990. 3. G. Dattoli, A. Renieri, and A. Torre, Lectures on the Free Electron Laser Theory and Related Topics, London: World Scientic, 1993. 4. R. Bonifacio, C. Pellegrini, and L. Narducci, Opt. Commun., 50: 373, 1984; R. Bonifacio et al, Revista del Nuovo Cimento, 13: 1, 1990. erenkov, Sov. Phys. 5. D. F. Alferov, Yu. A. Bashmakov, and P. A. C Usp., 32: 200, 1989; E. G. Bessonov, A. V. Vinogradov, Sov. Phys. Usp., 32: 806, 1989. 6. J. M. Madey, Jour. Appl. Phys., 42: 1906, 1971. 7. F. A. Hopf et al., Opt. Commun., 18: 413, 1976; Phys. Rev. Lett., 37: 1342, 1976. 8. W. B. Colson, Phys. Lett., 59A: 187, 1976. 9. A. Friedman et al., Rev. Mod. Phys., 60: 471, 1988. 10. D. A. G. Deacon et al., Phys. Rev. Lett., 38: 892, 1977. 11. P. Dobiash, P. Meystre, and M. O. Scully, IEEE J. Quant. Electr., QE-19: 1812, 1983; J. Geo-Banacloche et al., IEEE J. Quant. Electr., QE-23: 1558, 1987. 12. J. M. J. Madey, Nuovo Cimento, 50B (64): 1978; N. M. Kroll, P. L. Morton, M. N. Rosenbluth, IEEE J. Quan. Electr., QE-17: 1436, 1981. 13. N. A. Vinokurov and A. N. Skrinsky, Institute of Nuclear Physics, Novosibirsk Report No. INP77-59, 1977 unpublished; N. A. Vinokurov, Proc. 10th Inter. Conf. on High Energy Particle Accelerators, Serpukhov, 1977.
N S N S N LINAC N S N S N (a) E-beam
LINAC
E-beam Wiggler N S N S N
N S N S N (b)
Figure 10. A scheme of realization of a linac (a) and a storage ring (b).
726
FREQUENCY AND TIME INTERVAL METERS
14. A. Gover and P. Sprangle, IEEE J. Quant. Electr., QE-17: 1196, 1981. 15. S. Datta and A. E. Kaplan, Phys. Rev. A 31: 790, 1985. 16. L. R. Elias, Phys. Rev. Lett., 42: 977, 1979. 17. O. Kocharovskaya and Y. Khanin, JETP Lett., 48: 630, 1988; S. E. Harris, Phys. Rev. Lett., 62: 1033, 1989; M. O. Scully, S. Y. Zhu, and A. Gavrielides, Phys. Rev. Lett., 62: 2813, 1989. 18. G. Kurizki, M. O. Scully, and C. Keitel, Phys. Rev. Lett., 70: 1433, 1993; D. E. Nikonov, B. Sherman, G. Kurizki, M. O. Scully, Opt. Commun., 123: 363, 1996. 19. A. Yariv, Quantum Electronics, New York: Wiley, 1989, chap. 13. 20. B. Sherman et al., Phys. Rev. Lett., 75: 4602, 1995; D. E. Nikonov, M. O. Scully, and G. Kurizki, Phys. Rev. E, 54: 6780, 1996. 21. R. Bonifacio et al., Phys. Rev. A, 50: 1716, 1994. 22. E. L. Saldin, E. A. Schneidmiller, and M. V. Yurkov, Phys. Rep., 260: 187, 1995.
DMITRI E. NIKONOV
University of California at Santa Barbara
GERSHON KURIZKI
Max-Planck-Institut fu r Quantenoptik
YURI V. ROSTOVTSEV
Texas A&M University
FREE ELECTRON LASERS. See SUBMILLIMETER WAVE

LASERS.
FREE FORM SURFACE RECONSTRUCTION. See

FUNCTION APPROXIMATION.
FREE-SPACE PROPAGATION. See FRIIS FREE-SPACE

TRANSMISSION FORMULA.
FREE-SPACE TRANSMISSION. See FRIIS FREE-SPACE

TRANSMISSION FORMULA.
FREE-SPACE TRANSMISSION FORMULA. See FRIIS

FREE-SPACE TRANSMISSION FORMULA.

(John G. Webster (Editor) ) 50.quantum Electronics

Uploaded by

Copyright:

Available Formats

You might also like

(John G. Webster (Editor) ) 50.quantum Electronics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(John G. Webster (Editor) ) 50.quantum Electronics

Uploaded by

Copyright:

Available Formats

50 Quantum Electronics

Home / Engineering / Electrical and Electronics Engineering

Wiley Encyclopedia of Electrical and Electronics Engineering

Browse this title

Search this title Enter words or phrases

Advanced Product Search Search All Content Acronym Finder

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

The Bragg Period

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS is equal to an integer multiple of half wavelengths

Edge-Emitting Single Frequency Lasers

DISTRIBUTED BRAGG REFLECTOR LASERS

When n(z ) is squared, the term including (

n)2 can be neglected. Therefore

DISTRIBUTED BRAGG REFLECTOR LASERS

The second term of Eq. (7) becomes

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

DISTRIBUTED BRAGG REFLECTOR LASERS

S. D. ROH R. B. SWINT J. J. COLEMAN University of Illinois

Home / Engineering / Electrical and Electronics Engineering

Wiley Encyclopedia of Electrical and Electronics Engineering

Browse this title

Search this title Enter words or phrases

Advanced Product Search Search All Content Acronym Finder

DISTRIBUTED FEEDBACK LASERS

DISTRIBUTED FEEDBACK LASERS

Buried heterostructure Dielectric layer p contact electrode

n-InP p-InP n-InP InGaAsP cap layer

DFB grating Active region

n contact electrode Longitudinal direction

InP InGaAsP InGaAsP or InGaAs

Single quantum well

Multiple quantum well

100 1.3 m window

1.2 1.3 1.4 Wavelength ( m)

DISTRIBUTED FEEDBACK LASERS

The expression for k2 in Eq. (2) can then be written as

k2 2 + 2 j 0 + 4 cos (20 z ) = 2 + 2 j 0 + 2(e 2 j 0 + e2 j 0 ) n +j r + j i = 2 (9)

DISTRIBUTED FEEDBACK LASERS

Substituting Eq. (9) and Eq. (10) into Eq. (1),

With these simplications, the coupled wave equations reduce to

DISTRIBUTED FEEDBACK LASERS

i 1 2 n 0 ( v v o ) = tan 1 L 2 r c i 1 1 v vo = q + + tan 1 (c/2n0 L ) 2 r

where 2v and q is an integer such that q .

DISTRIBUTED FEEDBACK LASERS

Back facet monitor Ball lens Isolator

2.5 2.0 1.5 1.0

10 Laser threshold 0 0 100 200 300 Current (mA) 400