wave mechanics applied to semiconductor heterostructures GERALD BASTARD les éditions @2... Avenue du Hoggar, Zone Industrielle de Courtabeeuf, B.P. 112, 91944 Les Ulis Cedex, FrancePreface Following the pioneering work of Esaki and Tsu in 1970, the study of two dimensional semiconductor heterostructures, namely modulation-doped hetero- junctions, quantum wells and superlattices, has developed rapidly, both from the point of view of basic physics and of applications. For example, modulation-doped heterojunctions are nowadays currently used to investigate the integer and fractional quantum Hall effects as well as to make very fast transistors. This book, which began as a set of lecture notes for a graduate course given at the Ecole Normale Supérieure, is specifically concerned with the basic electronic and optical properties of two dimensional semiconductor heterostructures based on III-V and (to a lesser extent) I-VI compounds, but applications are not considered. It is intended as an introductory textbook for undergraduate and graduate students and engineers working in this field. It is essentially an attempt of exploring various consequences of one-dimensional size-quantization, a genuine quantum-mechanical effect, on the most basic physical properties of heterolayers. Because one of their degrees of freedom is frozen by this size-quantization, the carriers in heterolayers effectively behave as if their motion were bi-dimensional, a feature which alters a number of physical properties. The book starts with a chapter recalling a few basic quantum mechanical properties of idealized quantum wells and superlattices (those found in quantum mechanics textbook). Chapter II is a summary of the k.p analysis of the electronic dispersion relations in direct gap bulk III-V and II-VI compounds. In chapter III we show how it is possible, in a simplified approach, to intertwine the results of chapters I and II to obtain the electronic dispersion relations in flat band super- lattices. In chapter IV we discuss the occurence of bound states when the hetero- structure is imperfect (hydrogenic impurities or interface defects) or when it is shone with near bandgap light (excitons). Then we switch to charge transfer mechanism and discuss in chapter V the carrier bound states in the quasi-triangular well formed nearby the interface of a modulation-doped heterostructure. Chapter VI is devoted to the in-plane ohmic electrical properties of these modulation-doped heterostructures, focussing the attention on elastic scattering processes. Chapter VII deals with some of the basic optical properties (absorption, photoluminescence) of quantum wells and superlattices. Then, in chapter VIII, we discuss the alterations of heterostructure energy levels by static electric or magnetic fields. 1 have attempted to make each chapter reasonably self contained, which has led to the occurence of a few Tepetitions. Several difficulties are encountered in the writing of a book on such a rapidly evolving subject. One problem is the large number of papers which are continuously published, providing additional results and leading sometimes to the modification of current ideas. I can only hope that the principles presented in this book will beIv Wave mechanics applied to semiconductor heterostructures helpful and fruitful to understand new results and to generate new ideas. Another difficulty is the large number and the variety of papers already published on semi- conductor heterostructures. Instead of trying to include all these papers, a representative bibliography is given at the end of each chapter, and I apologize in advance to all my colleagues whose impostant contribution has not been quoted. I wish to thank those who granted me permission to reproduce their figures. Finally, the author would like to acknowledge all the physicists with whom he had the privilege to work : Drs. J. A. Brum, C, Delafande, Y. Guldner, M. H. Meynadier,. J. Orgonasi, J. P. Vieren, P. Voisin and U. O. Ziemelis at the Ecole Normale Supérieure (Paris) and Drs. C. A. Chang, L. L. Chang, L. Esaki, E. B. Mendez and F. Stern at IBM (Yorktown Heights, U.S.A.). Above all, Michel Voos deserves hearty thanks for having taken the time and displayed enough patience to discuss the physics and to carefully check the manuscript.Contents CHAPTER I : IDEALIZED QUANTUM WELLS AND SUPER- LATTICES. A SUMMARY II. Density of states 200.000... 000 cceveceee eee eeeeeteeeee eee eeees Ill. Tunnel coupling between wells: the symmetric double square well IV. Superlattices ... IV.1. Superlattice dispersion relations . IV.2, Symmetry properties of the eigenfunctions . . IV.3. Superlattice density of states ........---060.esceceee ewes [ 1. Single quantum wells .. cocteeeteeeeees APPENDIX A : Symmetric double square well. Time-dependent aspect . APPENDIX B : Dispersion relations of superlattices with arbitrary potential profiles . CHAPTER II : BAND STRUCTURE OF BULK III-V COMPOUNDS 1. Crystalline properties . Il. Electronic properties .. III, Electronic dispersion relations in the vicinity of the zone centre : k.p analysis and Kane model IIL.1. k.p analysis and effective masses 111.2. Effective mass tensor. Electrons and holes . II1.3. Beyond the quadratic dispersion relations : the Kane model .. @) Dispersions relations of the Kane model b) Band edge effective masses ©) Band non parabolicity . @) Inclusion of higher bands . e) Accuracy of the Kane model . IV. Conclusions APPENDIX A : Inclusion of remote bands effects APPENDIX B : Motion of Kane electrons in slowly varying perturbing Gils ieee eect cee eeet ee eeenetee teen page pageVI Wave mechanics applied to semiconductor heterostructures CHAPTER III : ENVELOPE FUNCTION DESCRIPTION OF HETE- ROSTRUCTURE ELECTRONIC STATES ......... Te Introduction oo... cece cece tee e eee eee e eens en etn n eens Il. The envelope function model . IL.1. IL.2. IL.3. IL.4. ILS. IL.6. IL.7. IL8. Ill. In-plane dispersions in semiconductor heterostructures . Preliminaries ... The envelope function framework . The Ben Daniel-Duke model 11.3.1. The Ben Daniel-Duke quantum wells (ramp > 0) 11.3.2. Interface states of Ben Daniel-Duke quantum wells (mymg <0; k, = 0) Quantum wells and superlattices with hosts which display Kane-like bands Simplified calculations of superlattices and quantum wells statesk, =0.. : Miscelleanous limiting cases . 11.6.1. Evanescent propagation in one kind of layer. Quantum well bound states . 11.6.2. Tight-binding expansion of the superlattice states . 11.6.3. Propagating states in both kinds of layers . 11.6.4, Heavy hole superlattice states Specific examples . Labelling and counting superlattice states . APPENDIX A : Boundary conditions and stationary states .......... APPENDIX B : Coupling between light and heavy particle states due to inversion asymmetry splitting in bulk materials. Qualitative aspects ......0..cecceeeeevveeeeeeteees CHAPTER IV : COULOMBIC BOUND STATES AND INTERFACE 1. Qualitative aspects . I.1. Bulk hydrogenic impurities .. 1.2. Impurities in heterostructures . Il. Approximate solutions of the hydrogenic impurity problem . IL. 11.2. I1.3. IL.4. ILS. DEFECTS IN HETEROSTRUCTURES ...........- Coulombic bound states Formulation of the problem .. Results for the ground impurity state attached to the ground SUBBANG eee eee cece eveeeeeeeeeeueeceeu esse neeees Excited subbands. Continuum Excited impurity levels attached to the ground subband Acceptor levels in a quantum well page Page 63 BIURKLLS 79 83 88 89 89 90 o1 92 93 100 101 112 113 119 119 119 120 121 121 123 125 127 128Contents Ill. Excitons Excitons in idealized bulk materials . Excitons in idealized quantum well structures . Excitons in actual quantum well structures . Interface defects 1. Graded interfaces . II. Quantum welt interface defects . Ill. Superlattice defect states APPENDIX : Energy levels in a quantum well with narrow graded interfaces . CHAPTER V : ENERGY LEVELS IN MODULATION-DOPED HETEROSTRUCTURES ......... 0.000. :ee ese eee I. The modulation doping of heterojunctions. Qualitative aspects . 1.1. The unperturbed heterojunction (flat band case) 1.2. Single heterojunction containing diluted Coulom| 1,3. Actual heterostructures Il. Self consistent calculations of energy levels and charge transfer in single heterojunctions IL.1. Energy level calculations IL.2. Charge transfer in single heterojunctions .. IIL. Energy levels in modulation-doped quantum wells . APPENDIX A : The algebra of the modified Fang Howard wave- function . APPENDIX B : ‘Heterojunction energy levels : qualitative aspects ..... CHAPTER VI : ELECTRICAL CONDUCTIVITY OF QUASI BI- DIMENSIONAL ELECTRON GASES ............-- I. Static conductivity of a quasi bi-dimensional electron gas . 1.1, Electrical conductivity in the electric quantum limit 1.2. Intersubband scattering . 1.3. Screening in a quasi bi-dimensional electron gas. Application to Coulombic impurities 1.4. Discussion of the dielectric function Il. Evaluation of some scattering mechanisms IL1. Mobility limited by Coulombic scattering . 11.2. Alloy scattering 11.3. Interface roughness scattering Ill. Vertical transport . IV. Resonant tunnelling APPENDIX : Dimensional dependence of the screening effects . page page VII 128 130 132 139 142 143 149 151 155 155 156 157 160 161 165 172 180 186 188 193 194 199 204 207 212 212 219 223 225 228 233Vill Wave mechanics applied to semiconductor heterostructures CHAPTER VII : OPTICAL PROPERTIES OF QUASI BI-DIMEN- SIONAL SYSTEMS page 237 1, Absorption (one-electron approximation) . 237 1.1. Intraband transitions ........... 243 1.2. Interband transitions .......... 246 1.2.1. Polarization selection rules 247 1.2.2. Selection rules on the envelope function quantum num- bers : evaluation of < f/f; > 247 1.2.3. Order of magnitude of the absorption coefficient. Com- parison between type I and type II systems . : 251 - 1.2.4, Interband optical transitions in superlattices . cee 255 Il. Absorption : a simplified description of excitonic effects 258 Il.1. Absorption in the absence of Coulombic intera valence with the one-electron model . 260 \ ma. Absorption in the presence of electron-electron interactions 262 (Ill. Phoroluminescence of quasi bi-dimensional systems . 272 Il,1, Introduction 272 111.2. Quantum well luminescence (steady state) 281 11.2.1. Band to band emission .. 282 111.2.2. Excitonic recombination . : 289 11.2.3. Extrinsic photoluminescence .............0020000 295 CHAPTER VIII : EFFECT OF STATIC EXTERNAL ELECTRIC AND MAGNETIC FIELDS ON THE ENERGY LEVELS OF QUASI BI-DIMENSIONAL ELEC- TRON GASES .......00. 066-00. ccee cece ee eeee page 303 I. The Stark effects 303 1.1, Transverse Stark effect in a quantum well (F/x) 304 1.2, Longitudinal Stark effect in a quantum well (F/z) ... 308 Il, Landau quantization of a quasi bi-dimensional electron gas . 317 11.1. Energy levels 317 11.2. Magnetic field dependent density of states 325 11.3. Magnetoconductivity of a quasi bi-dimensional electron gas . 329 11.3.1. Macroscopic derivation . 329 11.3.2. Microscopic discussion . 332, 11.4, Cyclotron resonance 339 APPENDIX A : Motion of three-dimensional electrons in crossed elec- tric and magnetic fields ........ 666... ceeee eevee ees 345 AI. The Drude model 346 AII. Quantum motion of bulk electron: 348 a magnetic field . APPENDIX B : Microscopic evaluation of the magnetoconductivity LONSOP oes eevee vec ee ee eee te eeeeeeeeeennetenes 351CHAPTER I Idealized quantum wells and superlattices. A summary In this introductory chapter we shall briefly recall some of the properties of the solutions of the Schrédinger equation for one dimensional square quantum wells, double square quantum wells and superlattices. By idealized we mean that the carrier mass in such structures is both position and energy independent. In principle, this mass should be taken as equal to the bare electron mass mp. However, for the sake of numerical examples, we shall often use an effective mass m*. This differs from rp and will be taken as equal to either the conduction or to the heavy hole effective mass of GaAs. (0.067 mp and ~ 0.4 mo respectively) L. Single quantum wells. Let us consider the one-dimensional motion of a particle of mass m* which is subjected to a potential energy V,(z) (see Fig. 1) which is such that @ Vy where L is the thickness of the potential well (hereafter termed quantum well). 1.1 BOUND MoTIONs (- V,< ¢ <0). 1) Classical analysis : For energies ¢ such that — V, < © <0, the classical solutions consist of bound particle motions which take place between + 5 . The particle never + ye 0 Es Ee “Vyb es Lois, “Oak 2 2 2 Fig. 1.— Potential energy profile of a square quantum well.2 Wave mechanics applied to semiconductor heterostructures A a L . . penetrates the barrier since © <0 and |z|> would correspond to an imaginary velocity, which is meaningless in classical mechanics. Inside the well the carrier velocity is constant and equal to zy +Vy). (2) At the well boundaries the carrier is perfectly reflected and its velocity changes instantaneously from + |v,| to + |v,|.The classical motion is thus periodic in time with a period of 2L 3) Any energy ¢ > — V, is allowed, i.e. it can be associated with a possible motion. 2) Quantum-mechanical analysis ; The quantum-mechanical motion is described by a wavefunction y/(z,#) solution of the time-dependent Schrédinger equation ay a ( ha ) ne! = Piz) yy 4 i aP=Fay)Y (4) where 9 is the Hamiltonian of the classical problem : 2 am* we + Volz), co) since does not depend explicitly on time 1, y(z, #) factorizes into vet ¥G, 1) = x@exp (-1), © provided that x(z) satisfies the eigenvalue problem 3 (2, .) x(2) = &x@)s o or more explicitly Pe [-teS+ Vs@)] x) = xC. ® (2) must fulfill the following boundary conditions : i) x(z) is continuous everywhere ; ii) by integrating equation (8) around any zp, and for the specific potential of the square quantum well, S is continuous everywhere ; iii) lim |x(2)| is finite. r4a20Idealized quantum wells and superlattices 3 Since V,(z) is piecewise constant, exact solutions of equation (8) can be obtained. For both |2| .) while it is in the ground state (e = E,): 2 Py= F(z) dz. 17) J.e8 Iz. a7) In figure 2 Py is plotted versus L for a quantum well problem corresponding to V, = 224 meV, m* = 0.067 mo and V, = 150 meV, m* = 0.4 mg respectively. It can be seen that the penetration of the barrier by the carrier is very small when L> 100A, For these L’s one may consider to a good approximation that the carrier is completly confined in the well. Equations (15, 16) can only be solved numerically. However, one may easily show algebraically that the solutions E, ,, of these equations (n = 0, 1...) are such that dE, ,;/dL <0. We also notice that the solution E, ,; = 0, ie. «, (E,41) = 0 which happens just as the (n + 1)" bound state enters the well, corresponds to ky (Ena. =0) Lanz, =0, 1... (as) These solutions are thus evenly spaced in L with a period /k,(E = 0). In other words a quantum well of thickness L admits n(Z) bound states where : am* Vy L? me) =1 +14] jar | a9) Int (x) denotes the integer part of x (see Fig. 3). A one dimensional quantum well supports at least one bound state, irrespective of the height of the confining barrier. It supports an infinite number of bound levels if V, is infinite. Under this condition the solutions of equations (15, 16) are kyL=pm, p=1,2. (20) and thus, with the energy zero coinciding with the bottom of the well, pal (21)Idealized quantum wells and superlattices 5 100 Py (%) io" i l ! l | 0 100 200 300 L(A) Fig. 2. — Quantum well thickness (L) dependence of the integrated probability P, of finding the particle in the barrier of a square quantum well. The carrier state is assumed to be the lowest ground bound state of the well. Curve (a) : V, = 224 meV, m* = 0.067 mp, Curve (b) V, = 150 meV, m* = 0.4 my #2? 2m* L? conduction band mass). The normalized even and odd wavefunctions of the infinitely deep well are : The quantity is equal to ~ 56 meV if L = 100 A and m* = 0.067 my (GaAs 2 Wz L Xop i) = P2cos [er+n 2 5 lelsZs peo (22) rapt) = fin [@r+2 3]; le]<55 pao (2) * Classical limit (p + 00 ) of the infinitely deep well.6 Wave mechanics applied to semiconductor heterostructures 5 I T T 3 wn Number of bound states 0 100 200 300 L (A) Fig. 3. — The number of quantum well bound states of a square well are plotted versus the quantum weil thickness ZL. Curve (a): V,=224meV, m* =0.067 mg. Curve (b): V, = 150 meV, m* = 0.4 m In the limit of large p we study the probability P,(zo) of finding the carrier between zy — F and Zy+ ¥ while having the energy E,. It is equal to Pp (zo) dz =F + (1p ¥eos [22% ]} as pol (24) i.e. the carrier becomes uniformly distributed over the well (the cosine term averages to zero when p+ co for any zo # 0). This is just the classical result : during a period T of the classical motion, one may define the probability of finding the particle between z- Gand n+F as the ratio between the duration Af spent by the particle on the segment dz which is centred at zo, to the period of the motion. This yields : Preasiat (20) dz = © (25) The coincidence between the quantum description at high quantum numbers and the classical description is a general property which holds, irrespective of the exact shape of the Hamiltonian #.Idealized quantum wells and superlattices 7 + Confinement energy. Links with the localization. We have seen that the lowest lying allowed energy in the quantum description is not — Vp as in the classical description but differs from ~ V¥, by some positive energy (4 2 Sm ® in the specific case of a quantum well of infinite barrier height). This phenomenon is quite general and may be qualitatively derived using the Heisenberg relations. Consider a quantum system with eigenstates |n). The root-mean square deviation of a quantity a when the system is in the state |n) can be defined as : 4,a = / (0°), ~ (ay? (26) (at), = Qnlatiny = [ xt aP xa de. @n where ‘Then, the Heisenberg relations, when applied to the position of a particle with exhibits a one-dimensional motion, state that : A,z Anp, = W2, (28) where h = 27h is the Planck constant. The inequality in equation (28) means that it is impossible to know simultaneously the position and the momentum of the particle with an arbitrary accuracy. Rather, if the position of the particle is known exactly, its momentum is undefined and vice versa. Let us apply the inequality equation (28) to the square quantum well problem by taking n = 1 (ground state). First we notice that (2), = (p,), = 0 since x,(2) is even in z. As x,(z) is essentially localized over the quantum well width L, the quantity (2*), is of the order of L’. By using equation (28), we get ms 2 »~ 5. @) The average energy of the particle is (3), : B= (6),<-v,a—P,)+ (2 V,0-Py)+—~, 0) = (3), = Vall Py) + (Pe) ~- Vol Po) +A, where P, has been defined in equation (17). Thus E;>-V,. The difference between the quantum and classical results does not originate from the term Vy Py which can be made arbitrarily small if V, (or L) is large enough. Rather, the difference lies in the confinement energy term f?/8m* L?, which is of kinetic origin and is a genuine quantum feature in as much as it results from the Heisenberg relationships which involve #. Whenever the carrier position is known with an accuracy of L, (which is the meaning of the statement that the particle is localized in a region of size L), the carrier momentum is larger than or of the order of8 Wave mechanics applied to semiconductor heterostructures h/L, and the kinetic energy is of the order of #?/2m* L?, The reader wil! check by direct computations that for a quantum well with infinite barrier height one finds 7 2 p10 Az Ap; $[F-2] ~116h, G1) 2 whereas for the ground state of the one-dimensional harmonic oscillator (Vole) = 5m w?2?) whose wavefunction has a Gaussian shape, one finds biz Dip, = 1.2 UNBOUND MOTIONS (¢= 0). — For positive energies, the classical motion is unbound. A carrier moving from the left to the right of figure 4 has a constant velocity fik,/m* in the left-hand side barrier. At z = — L/2 it accelerates instan- taneously ; it moves across the well with the constant velocity hk,,/m* until it reaches the interface z = L/2 where it decelerates instantaneously and finally escapes to z= +4 at the constant velocity fik,/m*, A second classical motion occurs at the same energy : the carrier coming from the right hand side barrier impinges on the quantum well and finally escapes to z = — oo. In this motion the velocities are the opposite of those found in the previous description. The quantum-mechanical motion is characterized by a continuous spectrum of allowed energy states, Each eigenvalue ¢ is twice degenerate because the carrier motion occurs either from the left to the right or from the right to the left of the quantum well. The carrier wavevectors in the well and in the barrier are real, corresponding in both cases to propagating states : ky = Pe ey ky= Pe (e4+Vp)- (32) Fig. 4. — Schematic representation of the reflection-transmission phenomena experienced by an electron whose energy falls in the continuous part of the energy spectrum of a square quantum well.Idealized quantum wells and superlattices 9 An electron which impinges from z = — co on the quantum well is partially reflected, partially transmitted at the interface z= —L/2. Inside the well, the eigenstate is a combination of plane waves characterized by wavevectors + ky. The occurrence of a wave with wavevector — k,, accounts for the partial reflection at the second interface z = + L/2. For > L/2 the carrier is also partly transmitted and escapes towards z = oo with a wavevector + ky, Thus exp[ity(2+ 5) | +rexp[-its(2+4 )] zak x(2) = Ja exp(iky z) + B exp (~ iky z) lal== G3) 1 exp|iky (2-5 )| 745 Writing the continuities of y(z) and & (z) at both interfaces we obtain, after some manipulations : i “1 (es) feos ky 2-5 (¢ +4) sink, L} (34) 5 (€-f) sinky Zoosk, £5 (€+4) sink, 2" (35) where E=ky/ky (36) Let T(e) and R(e) denote the transmission and reflection coefficients of the well : T(e) = |t(e)|? 3 R(e) = | rCe) |? (7) Then R(e) + T(e) =1 (38) _“tral(e1\Peneg po Te) = (t+3 (¢ #) sin kL] (39) Equations (38, 39) look familiar. Actually they are nothing but the reflection and transmission coefficients of a Fabry-Pérot dielectric slab. This is not accidental but arises from the close analogy between the one-dimensional Schrédinger equation and the equation governing the propagation of an electromagnetic wave in a medium characterized by a position-dependent refractive index n(z). The transmission coefficient has two remarkable features : i) If Lk,(e = 0) # pz, the transmission coefficient vanishes at the onset of the continuum. We might have expected that a carrier with a vanishingly small velocity in the barrier would easily fall into the potential well. However, just the reverse is true in that the electron does not penetrate the well (a = 8 = 0). This behaviour reflects the wave-like nature of the electron, which is most strikingly revealed when the potential has sharp discontinuities. ii) The transmission coefficient is not a smooth function of the energy. Instead10 Wave mechanics applied to semiconductor heterostructures T(e) reaches unity whenever k,L= pm. (40) oot ?>7 TRANSMISSION ( TRANSMISSION 1. oly 0 1 2 30 40 30 60 70 & (meV) Fig. 5. — Energy dependence of the transmission coefficient T(e) in a square quantum well of thickness L = 250 A. Curve (a): V, = 224 meV, m* = 0.067 my. Curve (b): Vy = 150 meV, m* = 0.48 my This corresponds to constructive interferences inside the quantum well slab, whose effective thickness 2L should fit an integer number of carrier wavelengths. The discrete energies which fulfill equation (40) are called transmission resonances. They correspond to an enhanced probability of finding the carrier inside the quantum well (for the same reason that lasing action, as seen from outside the laser, corresponds to a very large amplitude of the electromagnetic wave inside the cavity). The transmission resonances may equally be viewed as virtual bound states of the quantum well. Like regular bound states, the virtual ones occur for discrete energies. True bound states however, are stationary solutions of the Schrédinger equation, which decay far from the quantum well, whereas virtual bound states are associated with extended wavefunctions. Another way to describe these states is to consider them as bound but non-stationary (decaying) states, i.e. to add to their energy a finite, negative, imaginary part. This view is supported by the time-dependent analysis of the transmission of a wave-packet by the quantum well (see e.g. [2]). When equation (40) is not fulfilled, the delay experienced by the wave-packet when it travels across the well is negative, as expected classically (the particle has a larger velocity in the well). On the other hand, when equation (40) is fulfilled, one finds that the carrier spends a long time inside the quantum well slab, swinging back and forth between z = + L/2, as in a regular bound level. However the particle finally escapes from the well. The delay time is longer for the narrower transmission resonances. This means that long-lived virtual bound states can be seen only if Ky/ky <1, ie. ife Vy.Idealized quantum wells and superlattices su Finally at © = 0 the resonance peaks (k, L = p7 ) becomes infinitely narrow. This in turn implies that the virtual bound state has become a true bound state. Indeed, the conditions « = 0 and Lk,(0) = pm are those which express the binding of a new bound state inside the quantum well (see Eq. (18)). Thus, true and virtual bound states match at the onset of the quantum well continuum. The latter are the continuation of the former when their confinement energies exceed the height of the confining barrier. Il. Density of states. It is important to know how many quantum states |») per unit energy are available around a given energy ¢. This quantity, the density of states, is equal to : p(s) =) 8 (e~«,), (41) where e, is the energy associated with the state |»). It often proves convenient to rewrite equation (41) in slightly different forms : ple) = S(e-2,) =F Cv 8(e — #)|v) =Trace 5(e- 9), (42) where Jf is the Hamiltonian operator whose eigenvalues are ¢,. The advantage of the Trace is that it can be calculated on any basis |Z) , even if |/) are not eigenstates of #. In some instances one can also rewrite &¢ — Jf) as: 8(e-#) =- 2 tim Im (e~ 96 +im)-%, (43) aa where Im z stands for "imaginary part of z”. Using equations (42, 43), p(«) is found to be equal to: p(e) — lim lm YE (e 96 ony"). (44) 10 7 For the square quantum well problem, the discrete spectrum is labelled by the index n of the bound state and, since electrons have a spin, by a spin quantum number g, +}. Thus |») = |n, o,) and p()=2E8(e~e,), e<0 (45) where the factor 2 has accounted for the spin degeneracy of each level n. We know in fact that a carrier motion is three dimensional. Thus a more realistic description of one-dimensional quantum well structures is provided by the solutions12 Wave mechanics applied to semiconductor heterostructures of: rele 2 -—.| —=+5+5 Vv = . (5 H+ 5) 90 +V.0 40) = ev 46) As the Hamiltomian is the sum of x, y and z dependent contributions, we know that we can look for eigenfunctions which are separable in x, y, z. Moreover, the carrier motion is free along the x and y directions. Thus 1 . . Y= Se oxplik x + ik, ¥1 x(@) «7 where S = L, Ly is the sample area. With equation (47) we find Pe c= Ext aes (48) where E,, is one of the eigenvalues of the one-dimensional Schrédinger equation of the quantum well problem (Eq. (8)) and k2 = k?+ kj. Thus, one may associate a two-dimensional subband, which represents the kinetic energy arising from the in- plane motion of the carrier with each of the quantum well bound states E,. The density of states associated with the motion described by equation (46) is : Rie Qm* o(e)=2 see, ~ ba ] > E,<0. (49) The sample area S is assumed to be of macroscopic size. Applying cyclic boundary conditions to the x and y motions, we find that Qn L a (50) Since L,, Ly are very large, any summation Y a(k,, ky) can be converted into an ky integration ; L,L, a(ky ky) EX | ona Ky ky)s 51) BD os Gay) ene by) a and finally m* § . ple) = ae L¥le~ Eli E, <0, (52) where ¥(x) is the step function : 1 if x>0 ¥@) = {o if x<0. 63)Idealized quantum wells and superlattices 13 p (ey = 3, 2 2 205 5 Ss Po ° o & Eo Es € Fig. 6. — Density of states associated with the quasi bi-dimensional motion of a carrier whose z motion corresponds to bound states of a square quantum well ; py = m*S/ah?. The energy zero is taken at the bottom of the well. The density of states is thus staircase-shaped. (see Fig. 6) Two facts are worth noting : i) p(e) =0 for © < Ey. A classical description would have led to allowed motions for any e. Here again we find the effect of the confinement energy associated with the localization of the particle along the z direction ii) p(e) exhibits jumps of finite amplitude whenever the energy passes through the edge of a two-dimensional subband. Such behaviour contrasts with the smoothly varying p(e) which is found for the free particles (p(e) = A Ve). The reduction of the dimensionality (i.e. two versus three) is witnessed by the increasingly singular behaviour of the density of states. In practice each step is smoothed out by defects. An empirical description of the broadening effects is obtained by replacing the step function Y(e— E,) by: e-E, Y(e-E,) = 4 { $+ arctan [‘S=2 | (54) broadening ™ 2 where I’, is the broadening parameter. ‘As will be shown in other chapters, points i) and ii) have important practical consequences : i) If a bulk material has a bandgap ¢, it absorbs light with an energy fiw = €,. A slab of the same material, clad between confining barriers to make a quantum well structure, has an absorption edge which is blue-shifted (with respect to the bulk value) by the sum of the confinement energies of the electrons and the holes. By adjusting the quantum well thickness L, this blue shift can be varied from zero to a few hundredths of a milli-electron volt. ii) If at low temperatures, the carriers only populate the lowest lying subband E, of a quantum well, their scattering by static defects only occurs through intra- subband (E,) mechanisms. If, by some means, some carriers become energetic14 Wave mechanics applied to semiconductor heterostructures enough to have energies which exceed F}, elastic inter-subband mechanisms become allowed. The onset of these new scattering channels is rather sharp in high quality materials and is witnessed by a sudden drop in the carrier mobility. * Classical limit of the density of states When the quantum well thickness L becomes very large, many states are bound in the well [Exo » ] and their energy separation becomes small (decreasing like L~2). Let us suppose that we are interested in evaluating p(e) for energy e such that -Vy<©<0. (85) Since L —-» 00 the energy levels such that E, <0 nearly coincide with those of an infinitely deep well and p(e) can be approximated by the expression : ole) = 223 Y Vie Vy—n Eh (66) mi nel where E, is the confinement energy of the ground state (see Eq. (21)). Since L is assumed to be large E, is small. Thus, by converting the summation over n into an integration we obtain for e + V,> Ey m*S e+Vy m*S jet+V» ams -1}.ms 87 ple) “| Is | nS (57) m* SL = mse 4V, 0) =e ae Ne Equation (58) coincides with the result obtained by neglecting the effect of the quantum well potential on the energy levels of the particle, i.e. equation (58) is the density of states of a particle of mass m* moving freely in a bulk sample having a volume LS. In practice, the classical limit equation (58) is reached if L = 10° A in GaAs-Ga,_,Al,As quantum wells, In table I a comparison between the expressions of some physical quantities calculated for bulk materials (volume @) and quasi bidimensional structures is summarized. or (8) III. Tunnel coupling between wells: the symmetric double square well. Let us consider two equivalent one-dimensional quantum wells of height V,, width L and separated by a distance A (Fig. 7). Each of the wells possesses na, bound states when they are isolated (my, =1). A localised wavefunction, which may be associated with each of these bound states, exponentially decays far away from the well. In the limit of infinite h, the bound states of the discrete spectrumIdealized quantum wells and superlattices 15 ° Wa) | | | -bi ch 0 2 bbe z 2 2 2 Fig. 7.— Potential energy profile of a double, symmetric, quantum well structure. (0 = © =~ V4) are twofold degenerate : the particle can be found either in one well, or in the other. At finite A, the previous eigenstates are no longer eigenstates of the coupled wells Hamiltonian : 2 P: HE = 5 + Vole ~21) + Vole ~ 22) 69) where 0 |z-a,| =k Volz -—z,) = 1,2 (60) L -V, - = » lF-al. The solutions of the one-dimensional Schrédinger equation is known exactly for each layer : well-acting (A) layer or barrier-acting (B) layer. For positive energies we have : x(2) = « exp [ik,(z — nd)] + B exp[—iky(z—nd)]; lz—ndl < 4 (74) x)= 7 exp[ik, (z—nd-4)] +6 exp|—ik, (z-na-4)] (75)Idealized quantum wells and superlattices 19 V(z) nd (n+ hyd (nel) d z Fig. 9. — Potential energy profile of a segment of a superlattice which consists of an infinite sequence of square quantum wells a: ee ke and : east Vota (76) We shall now exploit the periodicity of V,(z). Consider the translation operator Gz which is such that for any functionf(z), G,f(z) = f(z +4). The operator %z commutes with the Hamiltonian % of the particle. We can thus find the eigenfunctions of € which are also the eigenfunctions of G,. The ©, eigenvalues can be written exp(igd) where q is an arbitrary complex number. Moreover E,q = [Gz]"- Thus the functions y,(z), which are eigenfunctions of both and G,, must fulfill not only the Schrédinger equation but also the equation : Byg XqlZ) = X Q(z + nd) = exp lignd] x,(2z). (77) Equivalently to equation (77), the x4’s can be written in the form : Xq(2) = u,(z) expliqz] ; uz + 4) = u4(z), (78) ie. the y,’s are the products of a plane wave term by a periodic function in z. Equations (77, 78) constitute the Bloch-Floquet theorem ; x, is the Bloch function and Uu, is its associated periodic part. Let us impose cyclic boundary conditions on the eigenstates of IC. We imagine that the length of the crystal Nd is very large and we assume that the crystal maps onto itself end to end. Thus x4(z) must be such that : XZ + NA) = x4(2), (79) which means that : qNd = 2p 7, pinteger. (80)20 Wave mechanics applied to semiconductor heterostructures Consequently q must be real. N independent values of g are allowed, the spacing between any two of these consecutive values being 2%/Nd. Without loss of 7 , 3445 [: The q space is called the reciprocal space and the segment [- Ti4e [ which is a particular unit cell of the generality we may restrict q to the segment [- reciprocal space, is the first Brillouin zone. In the limit of large N, the discreteness of the allowed q values becomes unimportant and we may convert any summation of the form — g(q) into an integral : ae trebz Na [(7T aq) + 5 |" 9(a) dq, (81) Now = a fis BZ. r where B.Z. stands for Brillouin zone. For a more complete treatment of reciprocal space, Brillouin zone, etc... the reader is for example referred to Ashcroft and Mermin’s textbook [11]. dxg Let us express the continuity of x, and 4 at the interface I of figure 9 : « exp (iky 5 ) +Bexp(-ik, 5) = = vexp (~ ik, 4) +8 exp (ity ) ikl exp (ik, 5 ) - Bex (-ik, =) | = ikl y exp (ik ) ~ bexp (ik, 3 ) J. At the interface II of figure 9 we have : (82) xalz= @+1)4-F40] = vexp (iky4 ) + bexp (—ik 4). (83) To express xl? =(n+1d— £ + o| we take advantage of the Bloch theorem and write xgle= (41) d-5 +0] =e xa[eand—5 +0]. (4) Using the same method for the derivative, we finally end up with a 4x4 homogeneous system in a, B, y, 6. Non trivial solutions exist only if the following equation is satisfied : cos (qd) = 005 (ky L) cos (ky h) -3 iG +3 ) sin (ky L)sin (ky h) (85)Idealized quantum wells and superlattices 21 where : : & = he/ky (86) Equations (85, 86) are implicit equations linking the allowed energy ¢ and the wave vector q, i.e. of the dispersion relations e(q) of the Bloch waves. For a given value of q, these equations admit an infinite number of solutions. Thus, in order to label the various solutions, a subband index n is affixed to « and x. A one- dimensional Bloch state depends on two orbital quantum numbers 7, q and will thus be written as |nq). The allowed subbands ¢,, are separated by forbidden gaps. For energies corresponding to these energy gaps, there are no allowed Bloch states, i.e. there are no allowed eigenstates of J€ which fulfill the cyclic boundary conditions of equation (79). Evanescent states, corresponding to imaginary q's, do however exist and are only allowed if the superlattice has a finite length along the z axis. The magnitude of the forbidden gaps which separate two consecutive superlattice subbands |n,q), |n+1,q) decreases when n increases. In fact at large e(€ > Vy), ky~ ky and equation (85) reduces to ky = 4 + 2j 5 The superlattice potential is hardly felt by highly energetic electrons, or equivalently the electron wavelength (27/k,) becomes much smaller than the characteristic length of the obstacles (d) and thus the effects of diffraction become negligible. It can be noticed that the energies e, which are such that k, L = pz, always correspond to an allowed superlattice state (see Eq. (85). The superlattice states which correspond to the continuum states of the isolated wells can thus be viewed as the result ot the hybridization of the virtual bound states of isolated wells. We have up to now only considered the case of Bloch states associated with propagating states in both kinds of layers (wells and barriers) : ¢,, > 0. Let us now look at the case corresponding to — Vy < €,,<0. The wave is evanescent in the barrier and thus : Ky aikys kya (87) &~ ik; =z: (88) Allowed Bloch states must then fulfill : cos (qd) = cos (ky L) cosh (oh) -} [-2 +3] sin (k, L) sinh (yh). (89) z In the limit of infinitely thick barriers the right-hand side of equation (89) will diverge like exp(«, h), unless the multiplicative coefficient in front of it vanishes. This occurs if : cos (ky L)~ 5 [8 +5 | sn (b, £) =0, (90)22 Wave mechanics applied to semiconductor heterostructures which coincides with the transcendental equation whose solutions are the bound states of isolated quantum wells of thickness L. Actually equation (90) is the product of equation (15) by equation (16), i.e. we obtain all the levels, odd or even, at the same time. This is because we directly exploited the symmetry property (evenness in z) of V,(z) in the single well analysis. Thus, the superlattice subbands of negative energies appear as the hybridization of isolated quantum well bound states due to the tunnel coupling between the wells across the finite barriers, If these barriers are thick (and/or high) but finite, the product yh is large and the subbands are narrow (the subband width which is proportional to the transfer integral, decreases roughly exponentially with h at fixed 4). The function F(«) which appears on the right-hand side of equation (89) shows large variations with © and the energy segments where |F(e)|<1 are narrow (see Fig. 10). In order to obtain approximate, but convenient, subband dispersion relations, let us denote by E, E2,... E,, the isolated quantum well bound states solutions of equation (90). In addition, let us expand F(e) in the vicinity of the £’s, say Ej. To the first order in e~E) we obtain the dispersion relation of the j® subband in the form: ¢) (4) = Ej +5; + 24; c0s (qd), 1) ai l l 0 100 200 300 400 € (meV) Fig. 10.— The right-hand side of equation (89) is plotted versus the energy ¢. Three barrier heights are considered: curve (1): Vy, =0.1eV; curve (2): V,=0.2eV; curve (3): Vy, = 0.4 eV. m* = 0.067 mo, L = 100 A, h = 50 A, In all three curves the energy zero is taken at the bottom of the well. Each curve is interrupted at the top of the well. The shaded area corresponds to the allowed superlattice states : |F(e)| <1.Idealized quantum wells and superlattices 23 where : rE) _ Laan oF kes,” 2 (92) 1 '" Flas, and where the prime denotes the derivative with respect to «. Equation (91) is simply a tight-binding result (see e.g. [11]). Explicit expressions for t; and s, can be obtained i) by expanding xj(z) in terms of the "atomic" functions o{(z — nd), where y((z ~ nd) is the j" bound state wavefunction of the quantum well centred at z, = nd when it is considered as isolated : ¥ explignd) 9 Qe - nd) (93) 1 Xn) and ii) by retaining only nearest-neighbour interactions : i= | Rove ee-o o% 5=> j QZ) Volz — nd) QE). (95) ano IV.2 SYMMETRY PROPERTIES OF THE EIGENFUNCTIONS. — For q=0 or q= > the Bloch wave is stationary [Xmg—o(Z +4) = Xng=0(2)3 Xn,q-2@ +4) = = Xn,q22(2)]}- The wavefunction repeats itself identically (q = 0) or changes sign (4 -4 ) when going from one elementary cell to the next. Let us consider the ratio r= a/B (see Eq. (74). We find that : _ =) ,, lexp(iky h) ~ exp i (ky L + 44)] (1+ €) © [exp i(k, L + ky h) — exp(iqd)] (96) One can show that for g =0,q = a the ratio ris real and such that |r|? = 1. Thus r + 1. This shows that for g = 0 and g = 5 the Bloch function is either odd or even. with respect to the centres of the well acting layers. The same conclusion holds for the parity with respect to the centre of the barrier-acting layers. These analytical results arise from the symmetry properties of the superlattice Hamiltonian : let Ag, Ap be two parallel axes separated by 5. The product of two symmetries with respect to these two axes is equal to a translation Gy, the vector d(|d| = d) being perpendicular to A,,Ay. Let us apply this geometrical property to the axis Aa, Ag which bisects two consecutive A and B segments of a superlattice (Fig. 11). We have RgRa=bai Ra Rg= Ea (97)4 Wave mechanics applied to semiconductor heterostructures where Gy is the translation characterized by d = dz with d = L +h. We know that the Bloch states an eigenstates of G, with an eigenvalue e'%*, In addition however (Vi@), Ral = (Vo), Re] = 9 (98) [eh-a)- [ska]: ‘Thus R4, Rp commute with the superlattice Hamiltonian X. However Rq, Rp do not commute with each other since IRa, Ry] = -— Gy + Ba (100) and also (99) It is thus impossible to diagonalize simultaneously JC, Gy, Ra and Ry, in the general case. The points q = 0 and q = 4 of the Brillouin zone are noticeable exceptions since the commutator [Rq, Rp], evaluated over the Bloch states, vanishes for these two values of g. For gq = 0, RyRy Xn.q-0(2) = +Xn.q-0(2)- If we know that Xn,q=0(2) is even (odd) with respect to the centre of the A layers, then it should also be even (odd) with respect to the centres of the B layers. For a=5, Ra Raltng=51= ~Xma=F(2)- Thus if xp,¢-4() is even (odd) with respect to the centre of the A layers, it should be odd (even) with respect to the centres of the B layers. IV.3 SUPERLATTICE DENSITY OF STATES. — Let us reintroduce the free motion in the layer plane (x, y directions) and label a superlattice state with n, q, k,, o, where ‘zis the subband index, q the superlattice wavevector (ia1 <3 ) eke = (ky ky) the M'sRy (M), M’=Rg (M!') =, (M) Fig. 11.Idealized quantum wells and superlattices 25 carrier wavevector in the layer plane and o, the electron spin orientation °, } ) . The eigenenergies # kt £ (1, qs hy 2) = ae + Onl) (101) are o,~ independent and are the sum of the energies corresponding to the motions Wk parallel (e,(q)) and perpendicular ( om = ) to the z axis. Nd denotes the length of the superlattice along the z direction and S the area of the layers. Both Nd and S=L,L, are macroscopic quantities. The density of states p(e) is then equal to wD p(e) = we 8 [« oe | (102) or (103) where p,(e) is the density of states associated with the n" subband and ¥(x) is the ‘step function. In general the subband e,, has a finite energy width when q describes the first Brillouin zone’: Enin < &ny S Emax: FOF © < Emin Pn(€) =O whereas fOr € > Emaxs p,(e) is a constant : m*§ Pale) =N Se em (104) We observe that the plateau value N — is equal to N times the density of states 7 associated with a single bound state of a given quantum well. From equation (104) we deduce that when e falls into the bandgaps of the superlattice dispersion relations, the density of states is quantized in units of N m* $/t?. To study what happens when © corresponds to the allowed superlattice states, let us take the simple case : €nq = En + Sn ~2 |tn| cos (qd), (105) which, as shown previously, corresponds to a tight-binding description of the superlattice subbands. In the vicinity of q = 0 we find : ng = En + $n —2 In| + [tn] a, (106) whereas in the vicinity of q 3 we find : 2 ng = En + Sn +2 Ital ~ [tal [7-3] a (07)26 Wave mechanics applied to semiconductor heterostructures Notice that in the vicinity of both g = 0 and q = 5 the carrier dispersion relation is quadratic upon q. In both cases the effective mass along the 2 axis is proportional to the modulus of the transfer integral |¢,|. The smaller |¢,| (thick and/or high barrier), the heavier the effective mass. In the limit of vanishing |1,| (isolated quantum wells) the subband dispersion is flat (discrete level). For uncoupled discrete levels the effective mass for the propagation along 2 is infinite, although the carrier is characterized in each layer by a finite mass 7 *. The reason for this is that the carrier oscillates back and forth in the well and on average does not move along the @ axis. Thus its effective mass is infinite. Using equation (105) the density of states is readily evaluated : 0 if © 0, aq © it Po Os. -2ltnl =O tal €-E,-Sp Fig. 12. — Energy dependence of the density of states of the superlattice subband correspond- ing to equation (105). py = m*NS/mh?Idealized quantum wells and superlattices 2 Appendix A Symmetric double square well. Time-dependent aspect. Let us consider a symmetric double square well and assume that at time 7 =0 the particle is in the state |y,) : the ground state of the well "1". We want to evaluate the probability P(7) of finding the particle at time T in the state |x2) : the ground state of the well "2". If | ¥(r)) denotes the state of the particle at time 7, the rules of the quantum mechanics assert that : AT) = | x2] ¥(T)) |? (AL) To evaluate | ¥(r)) we have to solve the time-dependent Schrédinger equation in |W (ry) = (T+Vi +V2) | ¥(7)), (A2) with the boundary condition : |¥(r =0)) = |x). (A3) In equation (A2) we have : V, = Vee —21) V2 =Vi(z—2), (a4) and similarly : (lx1) = xi@— 21) (1X2) = x1(@ -22)- (AS) E, denotes the eigenenergy of the ground state of an isolated well. To solve equation (A2) we make the same assumptions as when we derived equations (61-68), ie, we write : IYO) = {a()|x1) + a(7) [x2)} (A6) If we insert equation (A6) into equation (A2) and project them onto the 90 ~ ° Lv0- fiowsroor] 9z0°0 _| a07 hd al , ee ee + 10" “zoo uo} [sm FeO fer0'0-s0H0°0} My | evo} eo} wool [2m 7 None cor a ee ey orca 080°0 ~ 86'0 1e0] 1r0 HOR~L)sL0~ Soro} — sz. 9sz'0 zsu-0}] Teel eo tot cot coro} [aae | size) | ewe sez) ox‘oyfsze't + ocr oaue|]_oar0) 1820) re) bays (ALL) 7% ete) [szrsOeTT8"0) TOIS'T) | S88°7 > 698°Z, 7S€7'0. stro. er T i sser'g 099°S £6560°9 SzEs9's LUISY'S. ce6ee'9 €3s0"9 1398'S asiv sviv qseD sveo avo sui svul aul | [2] s2qqm wraisusgg-4opunT] ayt uz avaddo Asmuvnb uaai3 v fo sain3yf posadas 1Yyi saworpul —- (y)2 joqucs YJ, ‘anyon asmosaduias W001 D st v darsuind aout] ay “soma {aun8Yf 42M01) aunsd1adwas woos ay) pun (aunByf s0ddr) aunioiaduia Moy ay aatB siayo0iq ayy “[z] aoussafas woul paptdiuos spunodwos ,-]]] poraas Jo sumauning — "| 1QeLBand structure of bulk II-V compounds 35 band structures, e.g. the solid solutions Ga;_,Al,As where the conduction band edge occurs at the I” point for GaAs and near the X point for AlAs, the same non linear variation of the fundamental bandgap is observed. IL. Electronic dispersion relations in the vicinity of the zone centre : k.p analysis and Kane model. In a bulk crystal the one-electron Schrédinger equation which has to be solved is : h 4g c? where mp is the free electron mass and V (r) is the crystalline potential. The latter includes some average of the electron-electron interaction and is periodic with the periodicity of the underlying Bravais lattice. The third and fourth terms in equation (1) are relativistic corrections. The third one is the spin-orbit coupling (o is the electron spin) and the fourth one, which is only important for very heavy atoms, includes the so-called mass-velocity and Darwin terms. If 4,i = 1,2,3 are three basis vectors of the Bravais lattice, we have : v (r+ Ema) =V(n), Q) ; [R+vo- 0 (ox vv p+ 02,] ¥e)= «40 (1) for any rand any relative integers n;. Thus the total potential energy which appears in equation (1) is periodic in r, and we may look for eigenfunctions of the Hamiltonian which are also eigenfunctions of the translation operators G4 where d= Sn 4: a Cav (t) = W(r+ d) = expi kd) y(n). @) Equivalently, the solutions of equations (1, 3) can be written in the Bloch form : Ynu(T) = Nung(r) exp (i kr), (4) where N is a normalization coefficient and u,4(r) a periodic function of r with the periodicity of the lattice : ta (2 na) = Uy (t). 6) If we normalize ¥,, over the whole crystal (volume 2), which consists of WP elementary cells (volume 2g), and assume k to be real we get N = 2-1”, provided the u,,’s are normalized to @, in a unit cell. A Bloch state |k) is thus labelled by a discrete band index n and a crystal wave vector k, which can be restricted to the first Brillouin zone of the reciprocal lattice (the spin index is omitted for brievity). Applying periodic boundary conditions to a macroscopic crystal Yale + Ni) = bul, Ni +00 ©36 Wave mechanics applied to semiconductor heterostructures we find that k must be real and such that KN, a =2mp;, i =1,2,3 ) where p; are integers. The allowed values of k thus form a quasicontinuum. Any summation of the form S= 5 fe) kel Be can be converted into an integration : so 2 f &k f(k). @) kel BZ. axe (Qn) ‘The Bloch functions are seldom known explicity. However, at the high symmetry points of the first Brillouin zone (noticeably the I"point), the way in which the Bloch functions change under the action of the symmetry operations belonging to the crystal structure point group can be analysed using group theory arguments. Four I” point Bloch functions and their linear combinations will repeatedly appear in the following discussions. They describe the crystal states for energies which correspond cither to the top of the occupied valence bands or to the bottom of the lowest-lying, empty, conduction band. The states are labelled |S), |X), |¥), |Z) and their associated wavefunctions transform in the same way as the atomic s, x, y, z functions under the symmetry operations which map the local tetrahedron onto itself. IIL.1 k.p ANALYSIS AND EFFECTIVE MASSES. — Global descriptions of the dispersion relations of bulk materials are available (pseudo-potentials, tight-binding) and quite accurate. However, they rarely provide explicit expressions for the quantities of common use in semiconductor physics : effective masses, wavefunctions etc... In fact, for many aspects of semiconductor electronic properties, such global descrip- tions of the dispersion relations over the whole Brillouin zone are unnecessary. What is needed is a knowledge of the ¢,, relationship over a small k range around the band extrema. Thus, in the following, we describe the results of a local description of the band structure : the k.p method, The periodic parts of the Bloch functions u,, are the solutions of [Eevee hg x WV). p+ 0 4mj c? WK? hk A } + + oxvw) gk = Ee Mo 9 2m m (P ding j i ® where the relativistic corrections have been dropped for simplicity. Equation (9) can be formally rewritten as : [H(k = 0) + W(k)) Un = nk Uno (10)Band structure of bulk III-V compounds 37 where H(k=0) is the crystal Hamiltonian whose eigenfunctions are jo (or té,q equivalently ) : (k= 0 J ey9 = 80 Mra (a) The k-dependent operator W(k) vanishes at k=0 and commutes with the translation operator Gy. Thus we can write : Han = Yen(K) tn (12) By inserting equation (12) into equation (10), multiplying by x) and integrating over a unit cell we obtain : zr (em ent EE) 0 42K nop + (o x WV) [my } cn (k) = 0 & me Om, |" my Amy c? ™ . (13) where : (n0|A|m0) = Ann = | Kip Aug dr. (14) anit cet Note in equation (13) the absence of diagonal (in n) terms proportional to k. One can easily show that such terms vanish identically. Equation (13), which is equivalent to equation (10), is well suited for a perturbative approach. Let us Suppose that the n"" band edge (energy e,9) is non degenerate (apart from spin). We can thus assume that for small k: 6,(K)~13 Cm(k) = Kk (15) since Cp (@) = 5 yy. In fact we have : 1 hk Cm) = 9 FO" Sag = eg (16) which, when inserted into equation (13), gives the second order correction to E03 KE | ram? nk = €9 + = +5 : 17. 10 Sng * me Sea “ In equations (16, 17) the vector a is defined as : a =p+—— (ox). (18) Amc Thus, as long as & is small (i.e. that €,,— €,9 remains much smaller than all the38 Wave mechanics applied to semiconductor heterostructures band edge gaps ey — &yo) the dispersion relations of the non degenerate bands are parabolic in k in the vicinity of the I” point. c= 0+ De sokes a,B=% yz as) Be where Tian Tam Vat 5,42 (20) Mar Mo ING man © ~ Fmd IIl.2 EFFECTIVE MASS TENSOR. ELECTRONS AND HOLES. — 1,7 is the effective mass tensor which describes the carrier kinematics in the vinicity of the zone centre and for energy close to the n" band edge. If the carrier energy is such that equation (20) is valid, the overall effects of the band structure (i.e. the fact that the carrier experiences the periodic potential V(r) instead of moving in the vacuum) are embodied in the use of an effective mass instead of the free electron mass. The notion of the effective mass is at the heart of the semiclassical description of the carrier motion in semiconductor. It can be shown [5] that, if an external force F is applied to the carrier, the crystal wavevector changes according to the law atk. =F. 2) The carrier velocity in a Bloch state |k) is exactly equal to 1 8€, Y= aK: (22) This relation together with the quadratic dispersion law equation (19), leads to : Ya 3 (4 nae a) kp (23) My which is reminiscent of the familiar relation v = - for free particles in vacuum. The semiclassical assumption is that the external force is weak enough to preclude any interband transition. In this way, despite the increment of its k vector through equation (21), the carrier velocity remains equal to equation (22). This is because the band index is constant. A more complete treatment of the semiclassical motion in solids can be found in [5]. For the conduction band edge associated with the antibonding s orbitals (I"; symmetry ) 4 2 is the simplest since it is a scalar : 5 & Bax, yz (24) (25)Band structure of bulk III-V compounds 39 Equations (20, 25) explicitly show that the sign and magnitude of an effective mass are governed by two factors : i) the squared strength of the matrix element between the Bloch functions of the edge of interest and those of all the other k = 0 edges. ii) the sign and magnitude of the Fbandgaps separating the edge of interest and all the other edges. In GaAs and InP for instance, the I" edges nearest to the I", conduction band edge are associated with the topmost valence bands y, (bonding p orbitals), and the dominant term in equation (25) is |7},,,|"/er,—€, which is positive. The Trg effective mass is therefore positive. It should also be noted that if the v, edges alone contributed to ~ and if my, ,was material-independent, mj) — mg would be proportional to the | inverse of the bandgap which separates the bonding states and the antibonding states, i.e. to the fundamental bandgap ¢q of direct gap TII-V or II-VI compounds. In fact, it is true that the smaller 9, the smaller mr, (see Tab. I). However the strict proportionality between my, and 9 is only observed in very narrow gap materials (ey <0.1 eV as in Hg,_,Cd,Te alloys with 0.16 = x = 0.21 ot low temperature). In wider gap materials, the other edges give a significant contribution to mr,. Finally it is important to notice that the ms ! term in the expression of »z4 ensures that the effective mass coincides with sy as it should for an empty lattice ‘characterized by V(r) = If the dominant contributions to the Stective mass of a given edge arise from k.p coupling between this edge and higher band edges, this effective mass is negative. This is what happens to the effective mass of the topmost occupied valence band v,. The analysis of the semiclassical motion of a particle with negative mass is not easy, as we are more accustomed to working with particles of positive mass. Let us assume for simplicity that the effective mass tensor is diagonal: HS" = —m,8qg,m,>0. Equation (23) then shows that the carrier velocity v is opposite to the carrier wavevector k. Newton ’s law, equation (21), reads : (26) Under many circumstances F has an electromagnetic origin : pe-e [estas]. (a Equations (26, 27) can be rewritten +m, Wave [Esta B], (28) which’ shows that the motions of valence electrons (negative charge, negative effective mass) when subjected to electromagnetic forces can be analysed just as well40 Wave mechanics applied to semiconductor heterostructures as the motions of fictitious particles which are characterized by a positive effective mass and a positive charge. Having considered the carrier motion, let us turn to the electrical current. If the carrier velocity in the Bloch state |nk) is vj, the electrical current density is sen. 029) d= The macroscopic current is the sum of J, over all the occupied states. Let f,(k) denote the distribution function of these occupied electron states (at thermal equilibrium, f,(k) would be the Fermi-Dirac distribution function). Then : I= FES fel 60) where we have assumed that a single band was populated. If this band is a partially filled valence band v,, the electrical current carried by electrons in such a band can be rewritten : J=S;+5; G1) where : (2) I=+ BY wi FWOI=+ ST fi ve (3) The first term, J,, is the contribution due to a filled band (f,(k) = 1 for all k). This contribution vanishes identically owing to equation (22). Thus the current carried by electrons which occupy a partially filled valence band is equivalent to that caused by fictitious particles of positive charge whose distribution function is that of unoccupied states. The fictitious particles are called holes. They are similar to free electrons in that they have a positive effective mass but, unlike the free electrons, their electrical charge is positive. In addition, the distribution function of the holes f,(k) is the same as that of the unoccupied electron states : 1 — f,(k). Finally let us compare the equilibrium distribution function (Fermi-Dirac) of electrons and holes in isotropic parabolic bands. We have : ke cb) = 6 BES ee, (4) 2 for the electron energy spectra, The Fermi Dirac distribution function of an electron is 1 vam = [rverr| ap (EE eee) |} oo where kg is the Boltzmann constant, T the temperature and « the chemical potential.Band structure of bulk III-V compounds 41 The hole distribution function of a partially filled valence band is therefore filoomou = U = fe Irom ont = [1 tex [= hee, EE — al] (6) which can be rewritten yy ful®hvom st = {1 +er0[ ge (SA) || on with 68) Thus the hole distribution function of a partially filled valence band coincides with the electron distribution function of a partially filled conduction band. Both f’sdecay exponentially at large wave vectors. Note however that the hole chemical potential measured from the valence edge is minus the electron chemical potential measured from the same edge. III.3 BEYOND THE QUADRATIC DISPERSION RELATIONS ! THE KANE MODEL. — For a given k, the lighter the effective mass, the larger the kinetic energy term Equ — Eno SOMetimes this term ceases to be negligible with respect to the various bandgaps and consequently, a more detailed analysis of the e,, relations is required. The first way to improve our calculations would be to extend the perturbative treatment of W(k) beyond the second order. This procedure, seldom used, is very cumbersome. In a celebrated paper [6], Kane took a different approach from that of the perturbative expansion of €,, in ascending powers of k. He noticed that for InSb (e9 ~ 0.23 eV), the topmost valence states v, and the lowest-lying conduction band I, were very close and well separated from all the other bands. Thus Kane diagonalized exactly W(k) within a limited set of band edges (I',, v,), and afterwards introduced the W(k) coupling between (I°,, v,) and the other F edges within the framework of a second order perturbative treatment. For a detailed account of the Kane model the reader is referred to his review papers [7, 8]. Here we restrict our considerations to the first step of Kane’s analysis. a) Dispersion relations of the Kane model : Since the spin-orbit coupling is non- zero in IH-V compounds it is desirable to have a k = 0 basis in which this term is already diagonal. Thus, instead of using the 8 band edge Bloch functions Ist), XP), [yf 1zt>, S$), xb), [¥ 4), 1Z$), we form linear combi- nations of these functions. These are such that the total angular momentum J= L + @ and its projection J, along the z axis are diagonal in the new basis. For the Sedge, the addition of L = 0 and o = 1/2 only gives J = 1/2 (I’g symmetry ). For the Pedges, adding L = 1 to « = 1/2 gives either J = 3/2 orJ = 1/2. In III-V compounds the quadruplet J = 3/2 is always higher in energy than the doublet J = 1/2. The2 Wave mechanics applied to semiconductor heterostructures Fig. 3.— Band structure of a disect gap II-V compound in the vicinity of the zone centre. Only the states (I',, 1';, I") near the Fermi energy at T = 0 K for a perfect crystal are presented. quadruplet corresponds to I, symmetry and the doublet to I’; symmetry. Their energy separation is noted A (see Fig. 3).The basis wavefunctions of the Ig, I'7, Ig edges which we shall use are labelled in table II. Note that each edge is twice degenerate and corresponds to the tWo opposite values of m, which are the eigenvalues of J,. Taking the matrix elements of W(k) between these 8 Bloch functions and dropping the relativistic corrections together with the k-dependent spin-orbit term (a — p ), we find that the energies ¢(k) are the eigenvalues of the following 8 x 8 matrix :43 Band structure of bulk II-V compounds44 Wave mechanics applied to semiconductor heterostructures Table I.—T4,1>,I's periodic parts of the Bloch functions. The notations J,m, label the eigenvalues of the operators J and J,, respectively. Ym, «(k= 0) ilsty 0 - Pe izty+Lyawsirysy ~« 5 Ye 0 quiero -e Fyla rity +S izty —e-4 ilSsy 0 { - riety ~ Bizhy —€ yyle-mh ~ 1 -i == |Z+ ey - lx MN +s > e-A AL (k,#ik,) and the energy zero is taken at the v2 T, edge. The eigenenergies depend on three parameters : the fundamental bandgap 9, the F’ spin-orbit coupling of the topmost valence band A In equation (39) k. = f= ent 5 a= ener, wo) and the interband matrix element of the velocity operator : P= Fe (SlpalX) = (SIPy|¥) = 5 4S1Pe1Z)- (4) All the other p matrix elements vanish by symmetry.Band structure of bulk IH-V compounds 45 If A (k) denotes the energy ad) = 2 FE, 42) We find from equation (39) that A (k) is the solution of A(k) = ~ & (43) or: A(K)LA Ch) + eo ][A (Kk) + 9 + A] = 2 wpa (H+ e+ 28 | (44) Each of the solutions of equations (43, 44) is twice degenerate. We notice that the dispersion relations are isotropic in k, i.e. e(k) depends only on k = ||. This occurs in spite of the fact that a preferential axis seems to be present: the J quantization axis, Actually, band warping [i.e. anisotropic ¢,,] does exist in HI-V materials, noticeably in the valence bands. However it resuits from k.p interactions between the Ty, I°;, 0, edges and the remote bands. In the present analysis, restricted to Tg, Fy, Ug bands, we could have chosen to quantize J along the k vector instead of along a fixed reference axis. Such a procedure renders the matrix (Eq. (39) block- diagonal (k, =0,k =k,) and, in addition, the solutions of equations (43, 44) correspond for any k, to a given value of my: m,= 5 for equation (43) and for equation (44). When J is quantized along a fixed direction, the ; and my +3 components. Clearly in a bulk material the quantization of J along k is the simplest solution, due to the isotropicity of the dispersion relations. Later on, we shall see that in the case of heterostructures it is convenient to quantize J along the growth (z) axis of the heterostructure. The dispersion relations will become anisotropic and will be simply obtained only if k//J//2. This is because the m, = + 1 eigenstates are in general a mixture of my = 51m = 25 decoupling will remain valid. On the other hand if k, is different from zero, the states of different my's will become admixed, leading to complicated algebra. b) Band edge effective masses : If we take W//J//2 we notice that the my states (Eq. (43)) correspond to the heavy particles. In fact the lack of k, p, coupling between I's (mm =+5 ) and Ig [(S|p,|X +i¥) =0] leads to an effective mass which coincides with the bare electron mass. The m, 5 states, however, are associated with light particles (I's electrons, " T light holes, I’; holes): their effective mass is already lighter than m) even if k.p coupling is allowed only within the I, 1"; 1°, subspace. In the vicinity of the band46 Wave mechanics applied to semiconductor heterostructures edges (€ = 0, © = — €9, © = — &) ~ A) we may expand the solutions of equation (44) to the second order in k, thus obtaining explicit expressions for the band edge (k +0) effective masses of the light particles. We find : 1 4P? 2p? — + 45 mr, Bey” 3(e9+ A) “ 1 _ 1 4p? tL a ing,” my 36 (46) 11 2p? t.1-_? _ 47) Mr, My 3(e9 + 4) “ ie. mp,>0 and mr, my, <0. Equations (45, 47) illustrate the previous general considerations on band edge effective masses. For instance mj (I stands for light) does not depend on A, which is quite natural since there is no k.p interaction between the I, and I’, edges. For the same reason mi, depends only on #) + A, which is the magnitude of the bandgap separating I’, from the I’, band with which it is k.p coupled. ‘The mirror effects for k.p couplings in equations (45, 47) should also be noted. The T, and F's k.p contributions to mr, are equal in magnitude and opposite in sign to the Ts &p contributions to my, and mr, respectively. ¢) Band non-parabolicity : Equation (44) in an implicit equation for A (k) versus k. put is explicit for k versus A(k). In figure 4 we show a plot of A(k) versus for both GaAs and InSb. We notice that the parabolic dispersion laws A(k) ome e ~ k* only hold when close to the band edges. By increasing k, the conduction band effective mass increases, This phenomenon, called the band non-parabolicity, is for a given k larger with a smaller ¢9. Indeed if ey — 0, A(k) tends towards a linear instead of a quadratic dependence upon the wavevector. Another way to depict band non- parabolicity is to rewrite the I’, dispersion relation in the form RK fk 48) mp) 8) where from equations (44, 45) : mr (A) (A + €9) (A + £9 + A) (69 + 24/3) (9) mr, 0) & (ey) +A) (A + &q + 2013) In this way it is clearly revealed that the mass increases as energy increases, d) Inclusion of higher bands: One of the main drawbacks of restricting the kp interaction to the I's, I°7, I, edges is that the heavy particle states are dispersionless, apart from the free electron term. For K//J it is easy to cure thisBand structure of bulk III-V compounds 47 oe T1175 ‘> Gabs P InSb 4 20 OF J z i" TR ab tT 4 L 1-5 1 J 1 on i on 7 0 1-4-2 0 2 4 x x Fig. 4,—The light particle energy A(K) scaled to the bandgap ¢% is plotted against xe for GaAs (left pannel) and InSb (righter pannel). In both figures the matrix Imre element P is fitted (Eq. (45)) to reproduce the actual effective mass m,,. Notice that negative x's correspond to evanescent states. weakness. Since both the light and heavy states decouple, we can take care of the k.p coupling between I's, ';, I's and the remote bands only for the heavy particle states. In doing so, the m, = + 3/2 states acquire a finite effective mass. Since this arises from an interaction with the remote bands, a parabolic treatment suffices. Thus : aie ek) = — 6-H, im 50 (k), °— Figg (50) with (3 1 1,2 2 t1.1,2 (51) myn Mo ME ma Mets. Fy As the energy denominators in equation (51) are large, the heavy hole effective mass is indeed heavy : usually my,/my~0.4—0.6 in all the INI-V compounds. Another way exists for including the remote bands effects inside the I's, ';, I's basis. It is based on the construction of an effective kp operator inside the I's, I, I's basis. For details see Appendix A. e) Accuracy of the Kane model : Because the Kane model goes beyond the simple parabolic dispersion laws, it represents a significant improvement on the latter ones, especially for narrow gap materials like InSb or InAs. However the Kane model has48 Wave mechanics applied to semiconductor heterostructures obvious limitations, For instance, the I", band dispersion relation asymptotically approaches the expression Re er, ()—~ = + AKP, (52) when A(k) > £9, A. This is incorrect when ep, (k) is no longer negligible compared with the bandgap which separates I", and the nearest remote band. To ascertain the accuracy of the Kane model it is useful to compare its predictions with those derived from a global description of the band structure, for example, the empirical tight binding method. The latter model builds the crystal band structure from a knowledge of the atomic properties of the constituting atoms. It is, in its construction, a global band structure calculation, valid over the whole Brillouin zone. The adjective “empirical” comes from the fact that some of the atomic properties, which are the inputs of the calculation, are fitted to the experiments to reproduce exactly some of the important features of the Brillouin zone, e.g. F bandgaps, effective masses etc.. Figure 5 presents a comparison between the dispersion relations obtained from the Kane model and the results of empirical tight binding calculations for GaAs and AlAs and 4//[100] [9]. The right hand part of each panel corresponds to propagating states (real wave vector) whereas the left hand part corresponds to evanescent states (imaginary wave vectors: k = ix in equations (44, 50)). The evanescent states do not play any part in the bulk materials (their wavefunctions cannot be normalized). >) -0h -04 Imky Imk, GaAs [100] Fig. 5.—Comparison between the electronic dispersion relations of GaAs and AIAs calculated either by the empirical tight-binding method (solid line), or by the Kane model (dashed line). After reference [9].Band structure of bulk HI-V compounds 49 They are of paramount importance in heterostructures made of bulk layers of finite thicknesses. It is remarkable that the Kane model and global band structure descriptions coincide so well for energies which correspond to the entire GaAs and AlAs bandgaps and up to ~ 0.3 eV in the GaAs conduction band. It is quite reassuring, with a future use of the Kane model in mind, that the heterostructure energy levels, which are the most important for pratical realizations (e.g. quantum well lasers or two-dimensional field effect transistors), are precisely those which are close enough from the I point where the Kane description is at its best. IV. Conclusions, Since we are going to rely on the Kane model, it is instructive to know the value of P or E, = 2m P? in III-V materials. This value can be extracted from reliable experiments and the use of equation (45). The bandgap values ¢9 and A are accurately known from optical experiments. The conduction band effective mass is also well known from cyclotron resonance experiments. Thus with these values and equation (45) we can extract an “empirical” Kane matrix element ,. It is empirical in the sense that higher band effects are present in the experimental m, but not in the theoretical expression. With this reservation in mind, table III, which shows the E, values for several III-V compounds, demonstrates that £, is mostly constant over the whole II-V family: £, ~ 22.3 eV if InP is excluded, E, ~21.2eV if InP is included. This means that an atomic like property f U,(r) py ux(r) Pr My is merely material independent. Thus the Bloch functions of two different III-V materials, which can be written (see Eq. (4)) as the products of a slowly varying envelope function exp ik.r. (k < Brillouin zone boundary) by a rapidly varying (atomic-like) function ,(r), are such that the atomic-like functions of the host T, Tz, I's band edges are quite similar. This result will lead us in the following chapters to analyse the heterostructure eigenstates in terms of the products of the Table III. — The values of eo, A, Eps mr, are listed for several I-V compounds. The quantity E, is equal 10 2m, P? and is fitted to reproduce the experimental mp, via equation (45). 0.811 0.341 0.752 mri 0.0665 | 0.0405 E,(eV) 22.49 22.71 | 22.8850 Wave mechanics applied to semiconductor heterostructures points atomic-like functions which are taken as identical in both host materials by functions which slowly vary at the scale of the hosts elementary cells and which will alone experience the differences between the band structures of these host materials. Appendix A. inclusion of remote bands effects. In this appendix we attempt to present a more complete discussion of the remote bands contributions to the 8 x 8 I", I; I's effective Hamiltonian. This topic has been the subject of many papers and complete reviews. Thus we will only describe the method of handling the remote band effects and then give the result of the simplest approach. We are interested in constructing an effective k.p operator which acts inside the I’, I, I°, basis by allowing at the lowest order for the presence of k = 0 edges other than I; IP’; I's. Our problem for the w,, functions can be formally written as : (Ho + W)I¥) = © |¥) (Al) where Ho is the crystal Hamiltonian and W the operator ke W=2kp+r (A?) mo *?* Ing (A2) ‘The energy spectrum of Hp consists of a set of closely spaced levels |/) (1 2, all display a p-like symmetry). The adjunction of the correcting term W — W to the effective k.p matrix equation (A7) cures this shortcoming. Thus, the I’; heavy particle states, which were dispersionless (apart from the free electron term), now acquire a dispersion. The price we pay for this improvement is that the diagonalization of equation (A7) can only be achieved numerically. The group theory analysis of WY - W helps to reduce the number of independent constants. [10-12]. However since the zinc-blende lattice lacks for inversion symmetry (the two atoms of the basis being different), this number is still too large. Inversion asymmetry-induced terms are usually very small. Thus, in a simplified approach we may discard them. Four independent parameters survive to this “quasi-Ge” assumption [10]. The 8 x 8 matrix ¥ —W has been derived by several authors. Here we follow Bir and Pikus [13] to obtain :Wave mechanics applied to semiconductor heterostructures 52 y _» _| oy zz WI (es NT 0 0 Balm) Hep I~ 0 £) z ow z - at. «H? alts * an aad H 0 ry | 0 I 0 x ou z z Zz -ot yo | So. g - L.z (4-9) TS He aw? 0 gq) +H? al 0 0 T= zy 0 0 ° 0 | 0 ou x 0 Ar ae I- 0 Ft-wo+nt a —|__ . yy ow Z 0 I 0 a sat z 7 0 0 w-ok HE 0 ° 0 0 0 2.t z_z zz ze fete aeBand structure of bulk III-V compounds where the polynomials F, G, H, I are given by : F(k)= Ak? + 3 (2 ~ 342) B 2 H(k) = — Dk, (k, + iky) G(k) = Ak? — 2 (k? — 3k?) ie) - 2.8 &-W)- 1D k,k, with : N , C?=D?-38?, p= v3 2 2 Xx 2 roy By |All) | 2m me 2 2 xX 2 mah By Weyl») | 2g © mi *y wa By Aled) Crleyl¥) + ele») Cv iesl¥) ms ee, 12s Msi) mmf €-&, 53 (A10) (All) (A12) (A13) (Al4) (A15) (Al6) (A17) (A18) In contrast to equation (43), the eigenvalues of the full Hy + W matrix are anisotropic in k (warped energy surfaces). This effect is most significant for the T, bands. When ¢) and A are large enough a parabolic description of the I's dispersion relations is sufficient. To obtain it one must include the I, levels back in the |») states in equations (A15-A17). Then, one ends up with a 4x4 matrix Hamiltonian # ;,. Rearranging the lines and columns we get finally : 2a) HED ed) ea) (33 F-«, | -in -1 0 oF 7, (k) = (3.5 | in* | Ge, 0 I Gif (3-31 | 9 (A19)54 Wave mechanics applied to semiconductor heterostructures The dispersion relations of the I's bands are exactly obtained in the parabolic limit : F-G en) = 6045 +6) (SE) esta (A20) ep, (k) = — 9 + Ak? + \/BOKS + C? [kp ky + 5 2 + k2 Kz] (A21) The relation (A21) explicitly reveals the cubic (but not spherical) symmetry of the iso-energy surfaces. The band warping disappears when the parameter C is set equal to zero. This spherical approximation is often done for computational conveniences, but, as seen from table I, is not very well justified. Note that if k /2 equation (A21) simplifies into er,(k,) = ~ 9+ (A+ B) k? (A22) which for the |3/2, = 3/2) states coincides with equation (50). Appendix B. Motion of Kane electrons in slowly varying perturbing fields. Up to now we have focused our attention on the energy spectra of perfect III-V compounds. In this appendix we attempt to describe the quantum motion of an electron moving in an imperfect III-V crystal. We look for the solutions of [90+ $(7)] ¥(r) = ©) (B1) where 3 is the Hamiltonian of the perfect crystal and (r) is a non periodic potential which we assume to be slowly varying at the scale of the primitive cell [14]. Let us first establish a property linked to the slowly varying nature of a function f(r). We consider the integral = I f(r) a(r) Pr (B2) where a (r)is a periodic function with the periodicity of the direct lattice : a(r+ Sma) = ae) (3) where a;, i = 1, 2, 3 are the basis vectors of the direct lattice, n; are relative integers,Band structure of bulk II-V compounds 55 and where f(r) decays fast enough at large r to ensure that J converges. The periodicity of a (r) allows the Fourier expansion. a(r) = @(K)expGK.r) (B4) where the K’s are vectors belonging to the reciprocal lattice : K=S mb, (B5) i the b’s being the basis vector of the reciprocal lattice. The Fourier coefficients a(k) are given by : a(k)= 4 [, a (r) exp[-iK.r] dr = x I, a(r)exp[-iK.r]@’r (B6) where @ is the volume of the unit cell of the direct lattice. Thus, the integral J is equal to 1=a(K) J, @r f(r) exp [i K.r]. (B7) x If we decompose the integral over 2 into a sum of integrals over the unit cells 2; centred at the lattice points Rj, we get : f a f(r)expiK.r= > f Pp f (Ri +p) expfikKe }. (B8) a RM Here, we have made use of the property K.R; = 2n7 where nis a relative integer. We now exploit the slowly varying nature of f(r) in order to write {(R +p) ~ f(R;) (p is restricted to a unit cell ). Thus : [ erreentnr~ 57m) J, expliKp lap. (39) The last integral is zero (unless K = 0). Thus : I~ a(0) [, @r f(r) (B10) or: j, fr)a(r) ara 4 f, a(n ar [, @r f(r). (B11) The Bloch states |nk) are normalized (ak| n! KY = Bey ep56 Wave mechanics applied to semiconductor heterostructures This means that 1 3 x I, ult) ty (0. 12) J | ited me aleve To solve equation (B1) we assume that $(r) only couples the I, I; I’, states between themselves. Thus we look for a solution of (B1) in the form of : HO= Eo ile) ple), (B13) mt where the f;(rys are envelope functions which slowly vary on the scale of My and where the up's are the I point Bloch functions for the 8 band edges (see Table II). For instance, if ¢ (r) vanishes, the f; are the plane waves 2~ "exp k.r. As long as the Kane model is justified (k ~ 75 of the Brillouin zone, see Fig. 5) they vary slowly. The wave function ¥/(r) is normalized : l= [, |¥@P Pr = Le om J, FM Fy Who yo Br. (B14) Making use of equations (B11, B13) we obtain 1= y Jer]? I, [filr)|? ar. (B15) Thus, the envelope functions have to be normalized to 1 in the volume 2, the ¢ being such that |e? = 1. T By inserting equation (B13) into equation (B1), multiplying by uj‘ ff and integrating over 2 we obtain : ‘ : 0 Safer 2) ffi frumutodtr + | utisa ine fier + dal a na 2m Z| uo mn fa phere [1860 frata uo | 16) ™M Joa a where the k-dependent spin-orbit coupling has been dropped, Making use of equation (B11) we finally obtain O= | Oar Fa [on [oer Fe een] +m of £00 corn where : Uo Pltyy Pr. (B18)Band structure of bulk I1I-V compounds 37 It is worth noting that the uo's no longer explicitly appear in the coupled system (Eq. (B17). Their existence is embodied into the matrix elements py. This is a direct consequence of the slow spatial variation of the perturbing potential (nr). It implies that ¢ (r) has only intra-edge matrix elements, whereas all the mixed terms | fi £% bux wu, | # m, vanish. The system (B17) can be formally rewritten as (f| D+ ((r)- €)1 | f) =9, (B19) where | f) is a (8 x 1) column vector, D an 8 x 8 matrix and 1 the 8 x 8 identity matrix. Non-vanishing solutions of the system exist if f is the eigensolution of the problem (D+ $(r)) f=e1F. (B20) The differential operator D is simply the (8 x 8) matrix of equation (39) where hk has been replaced by the differential operator p = 7 V. Indeed equation (39) is recovered if ¢ (r) is zero. In this case, f = 2-'expik.r C, where C is a 8 x.1c- column vector. Instead of handling the 8 x 8 D, it may prove useful to eleminate six of the f's to obtain a 2x 2 effective D which acts on the two remaining envelopes [15]. For instance, if ¢ (r) is the potential created by a Coulombic donor, which we expect to give rise to bound states below the I, edge, we will get rid of fs, fy... fg to the benefit of f,, f; Which are the envelopes associated with the I", edge. If we introduce £iPy) (B21) and drop the free electron term, we obtain : f= leo+e-o@P [ - rr. hia ge- fr fil) = [eo +e - OP! [-4. fi- (pre. 1 foe) = leo +e- FOF! Pe. fal fot) = [oot @- SOO) (PP fl (B22) fie) = [et at e-o (yy! [ ae fisP Joe. | fied=tovave-ocor'|- Rep. f + Fen Notice that in equations (B22) the two brackets do not commute. The effective58 Wave mechanics applied to semiconductor heterostructures 2x 2 system, which is a projection of the 8 x 8 system onto the I’, edges, is finally rewritten as ; [es ¥l [J- . A (B23) where : Pp? 2 4 = ag fe + 60 ~ OC pe + PPP, Le + &0- (TP. + 2x 2 2 ples ey- OO) py, +p [et ay tasty, + 2p. x [e+ e+ 4- OO) 12. + 6) (B24) Hoa -2rrp, {le + eo- o (et = [e+ 9+ 4- dP} pe —P: {le + e0- be)" = [e+ + 4-6 M1} pl] (B25) Let us recall that f,(f) is associated with the S f (S 4) band edge Bloch states. Equations (B23-B25) show that a scalar (i.e. spin independent) potential ¢ (r) mixes the |f) and |$) states. This is not surprising since the host crystalline matrix displays, even in the absence of a perturbing potential, a non zero spin-orbit coupling. For perfect crystals we have shown that it is always possible to obtain eigenstates of D which are also, at finite k, eigenstates of J, and thus do not display $ f and S$ mixing. To obtain such states we only had to rotate J in order to line it up along the k vector. The envelope functions of the perturbed crystal are linear conbinations of plane waves. Thus in the presence of a perturbing potential @(r) it is in general impossible to achieve a simultaneous diagonalization of D + (r)1 and of J,. The eigenstates of a perturbed semiconductor are, in general, a mixture of different m,’s. It should be noted that 3,4 Vanishes identically if A = 0, i.e. if the host matrix has no spin-orbit effect. We also note that 3,4 involves the energy denominators [e+ €9— (r)}! and [e + e9 + A— $(r)}-'. Thus the smaller ¢9 favours the larger spin mixing effects. In fact, striking evidences of the influence of 3,4 have been found in narrow gap materials with large spin orbit effects, e.g. InSb or Hg, _,Cd, Te alloys (x ~ 0.2). For instance, strong electric dipole spin-flip magneto- optical transitions have been observed [4], as well as the anomalous Hall effect, i.e. a Hall conductivity which is induced by the spin magnetization of the carriers instead of by the external magnetic field [17, 18]. Finally it is interesting to notice that the eigenvalue equation (B23) is non-linear in eas far as #, and #4 explicitly depend on e. This situation is not unique as it is also found in the quantum treatment of relativistic particles [16, 19].Band structure of bulk IlI-V compounds 59 The hydrogenic donor problem. It may be illustrative to discuss more thoroughly a specific case. We take o(r) (B26) where « is the static dielectric constant of the semiconductor. We are interested in deriving the first order corrections to the usual hydrogenic Hamiltonian which are induced by the band non-parabolicity. We notice that for almost all r | (r)| < eo. When this inequality is not fulfilled, r is so small that #(r) can no longer be considered as slowly varying. Under most circumstances this range of the r values plays a negligible part in the binding energy. Since |#(r)| < €9, the energy ¢ which is of interest to us is also much smaller than ey. Thus we can expand #,and #,, to the lowest order in (- $(r) + ©) / eq, (— o(r) + €)/(e0 + 4). We obtain : tn OSG ea + Forse. + Ep. op. +P O(n) Pp. + Tae Sep eay? O(r)p, + oP o(r)p, (B27) es b-S()p,-p.O()p-L (B28) The zero order term HP = co +o(r) (B29) is the “parabolic” hydrogenic donor Hamiltonian whose solutions are the same as those of the hydrogenic atom problem, the only exception being that the Bohr radius and Rydberg energy are scaled according to aft = 0.53 Ax x 20 my mr, Re =13.6eV x K (B30) The next two terms to #{” in #, are kinetic energy corrections induced by the band non-parabolicity : they are of the order of £ - . To construct the hydrogenic £o9mr, wavefunctions we have to build a localized wave packet by picking up states of finite kin the conduction band. As soon as kis non zero, the effective mass attached to this60 Wave mechanics applied to semiconductor heterostructures k state becomes heavier than mf, (see Eq. (49). The parabolic term p?/2m,,thus overestimates the kinetic energy of the electron. Consequently, the non-parabolicity- induced kinetic energy corrections can only lower the energy. This explains the presence of a minus sign in front of the two terms. The other correcting terms of 39) in 2, involve ¢ (r). Nevertheless, they are also of kinetic origin. Locally, at the rpoint, the kinetic energy of the carrier is not ¢ (as would be the case if 6 (r) were zero) but rather ¢ — ¢ (r). Thus in establishing the previous kinetic energy corrections we have gone too far and must allow for the spatial non-uniformity of the non-parabolicity effects. This explains why the ¢-dependent terms are all affected by a sign which is opposite to that of the first set of corrections. As for #4, it has no non zero-order terms but results from the combined actions of the external potential (r), the band non-parabolcity ( yg ~% ) and the host £0 spin-orbit coupling (3,4 = 0 if A = 0). The eigenstates of 3) = 91 are the two sets of column vectors (aH o 0 Pntm(P), where n, J, m are the usual hydrogenic quantum numbers [19]. To the first order in 3 — #9) only the diagonal corrections, i.e. 9, — 9) appear. Higher order corrections involve 2,4 Which admixes, say, the |1s? ) level with |ndm $) levels. For the p states, which are degenerate at the parabolic approxi- mation, one should diagonalize 9 — 9 exactly. Clearly 3,4 acts as a spin-orbit effect and partially removes the sixfold degeneracy of the |n, Pim, 5) levels. References [1] W.A. Harrison, J. Vac. Sci. Technol 14 (1977) 1016 and B 3 (1985) 1231. [2] LaNDoLT-BoRNsTEIN, Numerical Data and Functional Relationships in Science and Technology, group Ill, Volume 17 (Springer Verlag, Berlin) 1982. [3] J.C. Priturs, Bonds and Bands in Semiconductors (Academic Press, London) 1973. [4] see, e.g., M.H. WEILER in Semiconductors and Semimetals Volume 16, edited by R.K. Willardson and A.C. Beer (Academic Press New, York) 1981. [5] N.W. Astcrorr and N.D. MERMIN, Solid State Physics (HoltRinehart and Winston, New York) 1976. [6] E.0. Kane, J. Phys. Chem. Sol. 1 (1957) 249. [7] see, e.g., E.O. Kane in Physics of I1I-V Compounds, Volume 1, edited by R.K. Willardson and A.C. Beer (Academic Press, New York) 1966. [8] E.O. Kane in Narrow Gap Semiconductors. Physics and Applications edited by W. Zawadzki, Lect. Notes Phys., Volume 133 (Springer Verlag, Berlin) 1980. [9] J.N. SCHULMAN and T.C. Me GILL, in Synthetic Modulated Structure Materials, edited by L.L. Chang and B.C. Giessen (Academic Press, New York) 1984. [10] C.R. Pipceon and R.N. Brown, Phys. Rev. 146 (1966) 575. [11] J.M. Lurrincer, Phys. Rev. 102 (1956) 1030.Band structure of bulk III-V compounds 61 [12] G. Dressetnaus, F. Kip and C. KirteL, Phys. Rev. 98 (1955) 368. [13] G.L. Bir and G.E. Pikus, Symmetry and Strain-Induced Effects in Semiconductors (Wiley, New York) 1974, [14] JM Lurrincer and W. KonN, Phys. Rev. 97 (1955) 869. [15] D.M. Larsen, J. Phys. Chem. Solids 29 (1968) 271. [16] W. Zawapzki, Adv. Phys. 23 (1974) 45. [17] Y. Yarer, Solid State Phys. 14 (1965) 1. [18] P. Nozieres and C. Lewiner, J. Phys, France 34 (1973) 901. [19] A. Messian, Mécanique Quantique (Dunod, Paris) 1959.CHAPTER III Envelope function description of heterostructure electronic states 1. Introduction. From Chapter II we know which are the eigenstates of bulk I1U-V compounds in the vicinity of the zone centre. In this chapter we shall deal with the determination of the eigenstates in heterostructures. The emphasis will be placed on a simple description of these eigenstates, i.e. a description based on the Kane analysis of the dispersion relations of the host materials. In this envelope function description [1-5], our problem will be to find the boundary conditions which the slowly varying parts of the heterostructure wavefunctions must fullfil at the hetero-interfaces. It should be stressed that other approaches to the heterostructure energy levels have been proposed which are more microscopic in essence than the envelope function scheme. In the empirical tight-binding calculations for instance (see e.g. [6]), one begins with a series of energies which are characteristic of the sp’ bonds linking one atom to its neighbours. The heterostructure wavefunction is then built atom after atom. In other words a heterostructure is nothing but a bulk material with a very large unit cell (e.g. the superlattice unit cell in the case of a periodic stacking of the A and B layers). It was previously thought that the size of the ogmputer calculations necessary to handle such large cells would become rapidly prohibitive, restricting the tight-binding analysis to short period superlattices where the unit cell is not too large. Chang and Schulman have however shown how to cure this drawback [7] and the tight-binding model is nowadays successfully used for heterostructures of any size, although it may have difficulties in handling self- consistent calculations which arise when charges are present in the heterostructure. Another microscopic approach is the pseudo-potential formalism, which is very sucessful in bulk materials. Recently, it has been applied to a variety of heterostruc- tures by Jaros et al. [8, 9]. Here, the core of the model is to consider, say, a periodic stacking of GaAs and AIAs slabs as a perturbation over the zero-order situation, which in this case would be the bulk GaAs. In other words a calculation is made agalogous to the deep level ones in bulk semiconductors. The advantage of these microscopic approaches is their capacity to handle any heterostructure energy levels, ie. those close to or far from the I edge. This occurs because these models reproduce the whole bulk dispersion relations. The model that we shall develop has no such generality. Basically, it is restricted to the vicinity of the high-symmetry points in the host’s Brillouin zone (I’, X, L). We feel however that it is invaluable due to its simplicity and versatility. If often leads to analytical results and leaves the user with the feeling that he can trace back, in a relatively transparent way, the physical origin of the numerical results. Besides, most of the heterostructures’ energy levels64 Wave mechanics applied to semiconductor heterostructures relevant to actual devices are relatively close to a high symmetry point in the hosts’ Brillouin zone. We shall first present the assumptions used to derive the envelope function model. As the reader may have perceived in chapter II, the algebra is relatively easy when the in-plane wavevector k, of the carrier is zero. The heterostructure states can then be classified according to my, the z projection of the total angular momentum J, and two kinds of energy levels result. The first kind corresponds to light particle states which are hybrids of the I’, {and I’; host states, whereas the second kind corresponds to the heavy hole levels. These k, = 0 states in quantum wells and superlattices will be extensively discussed and our considerations will be illustrated by specific examples. The third part of this chapter will be devoted to a discussion of the in-plane dispersion relations in the heterostructures. This is a topic which is currently being actively researched. Because of the mixing between the k, = 0 light and heavy particle states it has proved, up to now, impossible to obtain analytical solutions of the k, # 0 heterostructure problem. Therefore, we shall only present the assumptions used for the calculations and some examples of the in-plane dispersion relations in several heterostructures. Unless otherwise specified, the heterostructures are assumed to be under flat band conditions, i.e. contain no charges. In the presence of charges, self consistent calculations of energy levels are required. They are the subject of chapter V. Il, The envelope function model. 1.1 PRELIMINARIES. — Advanced epitaxial techniques, such as molecular beam expitaxy or metal-organic chemical vapour deposition, have made it possible to grow interfaces between two semiconductors which are flat up to one atomic monolayer (2.83 A in GaAs), which is the ultimate resolution which can be achieved. It is common to represent such an ideal interface in terms of a continuously varying position-dependent band edge (Fig. 1). Basically speaking the sketch shown in figure 1 means that for z > 0 the electron experiences a one-electron potential which is identical to that of a perfect bulk A material, whereas for z < 0 it experiences a one- electron potential which is the same as is found in a perfect bulk B layer. Of course, this scheme that we shall follow later on is only approximately true. Its accuracy is limited by several factors, both fundamental and technological in origin. On the one hand the sketch shown in figure 1 by-passes any ambiguity concerning the description of the electron states which originates from the border atoms, i.e. that hybrid interface bonds exist which are present in neither the buik A’layé nor in the bulk B layer. On the other hand, figure 1 tacitly assumes a perfectly bi-dimensional growth, i.e. that all over the area S the structure grows monolayer after monolayer. This bi- dimensional growth is, in practice never completely achieved and for that reason the interface location along the z axis varies with the in-plane coordinates. In practice, we feel it is better to envisage the interface as having a finite thickness to account for the effects previously mentioned. Fortunately, most of the electronic states which we shall discuss in the rest of the book hardly experience the interfaces, as they have small probability amplitudes of being found in these regions. Thus, in a firstHetorostructure electronic states 65 Elz) D0 O-eD2D-0-D-0-D-0-D-0-D— ABABABABCBCBCBCB es 0 z Fig. 1. — One-dimensional sketch of a heterojunction formed between two perfectly lattice matched A and B semiconductors with chemical formulae CB and AB respectively. Upper part: representation in terms of a position-dependent conduction band edge. Lower part : actual bonds, Notice the formation of the hybrid bonds A-B-C at the interface. approximation and for technologically abrupt interfaces, we shall retain the notion of mathematically abrupt interfaces. Consequences of the deviations in the actual interfaces with respect to this idealized model will be examined in chapter IV. For certain heterostructures, the notion of abrupt interfaces is irrelevant from the growth point of view due, for example, to an excessive and uncontrolled interdiffu- sion between the A and B materials. For those heterostructures with severely graded interfaces, one should resort to modelization of the grading effect. To our knowledge, little effort has been devoted to this technologically important problem. Since the notion of interfaces is somewhat fuzzy, the definition of layer thickness is also rather imprecise. However in the case of periodically arranged heterostructures (superlattices) the X-ray determination of the superlattice period is precise (see e.g. [10]). On the other hand, the thickness of an individual layer is rarely known to an accuracy of more than one atomic monolayer. In-situ measurements such as Reflection High Energy Electron Diffraction oscillations [11] have proved to be very valuable in ascertaining the layer thickness. Simpler techniques such as measuring the growth duration and using calibration obtained on thick (~ 1m) layers are less precise for the narrow layers (~ 100 A) which interest us. The abrupt interface model sketched in figure 1 makes a clear distinction between barrier-acting (B) and well-acting (A) materials. One may however raise questions on the significance of barrier and well denominations since we have seen that the interface notion is not very well defined. The two notions call for different properties. The interface is ill-defined because it is difficult to know at the atomic scale what kind of environement an electron sees when it is in the transition region between the A and B layers. The notion of wells and barriers calls, on the contrary, for the asymptotic behaviour of the carrier wavefunction occurring far from the66 Wave mechanics applied to semiconductor heterostructures interface. The exact heterostructure wavefunction corresponding to e, where 0 , Fg couplings vie a single excursion outside the I's, I>, Is multiplet (seeHetorostructure electronic states 1 Appendix A of the previous chapter) : m Mae gh Pei”) = pel») a) 1 @ —V,@) where v labels the remote I edges of the host layers, whose energies are position- dependent in the heterostructure (as expressed by the piecewise constant functions V,(z)) and where in an average energy of the I’,, I", I’ set in the heterostructure. To summarize, the y,(z) envelope functions are the solutions of a 8 x 8 second order differential system : é We we ih (a) tL 2 x {[-9 +Vnl2) + a 3 3] Bin ~ pa sl mat nk Pail a iv 1a,a1 a ho 18 ik 242 * he m,>0. It is also interesting to notice that the effective mass mismatch leads to a discontinuity in the derivative of the envelope function at the interfaces. In the extreme case where m, and mp are of opposite signs, this discontinuity causes a cusp at the interfaces of the envelope function. The latter situation occurs in HgTe-CdTe heterostructures (but only at k, = 0) (see section II 3.2). U.3.1 The Ben Daniel-Duke quantum wells (mam >0). — The k,-dependent potential energy V,(z) +=—~ is even with respect to the middle of the A layer. 2u@) Thus, as in chapter I, one can look for bound states solutions in the following forms : L Xeven(Z) = Acos (Kaz) Izle< > L, L, Xee(2) =B exp| - va(e- FF )] a) Xeven( — 2) = Xeven(2)Hetorostructure electronic states 75 Fig. 4. — Dispersion relations versus the real and imaginary wavevectors in the A and B layers stacked to form a BAB quantum wells, The three dashed lines, drawn for three energies in the heterostructures and for k, =, show which wavevectors in each kind of layers participate in the heterostructure state. The upper line corresponds to a delocalized quantum well state and the middle line to a quantum well bound state. No heterostructure state can be associated with the lower line as the carrier effective masses are assumed to have the same sign in this particular Ben Daniel-Duke quantum well. or: . La xeu(2) = Asin (kz) <8 Xead(2) = Bexp|- ip (:-3 )] z 28 (33) Xosal — 2) = — Xots(Z) with : We? PAE =V. af AR (34) 2m, ” 2my Equations (32-34) hold if «3 >0, i.e. if the heterostructure state is built from the evanescent states of the B layers (see Fig. 4). L By matching x(z) and ani) at the interface z= —* (or equivalently 4 L z=- = ), one obtains the implicit equations whose roots are the bound solutions of the Ben Daniel-Duke quantum well problem. These are : ms ka cos -—— sin = Pa ma Kp Pa for even states (35)6 Wave mechanics applied to semiconductor heterostructures Maks. cos pats ~2sing, =0 — for odd states (36) mp F zs ea =p kaln en A comparison with the results obtained in chapter I shows that equations (35-37) are the same as those of text book quantum wells except that the wavevectors Ka, py have been replaced by kg/m, and kg/m. This is a direct consequence of the matching conditions of ~1(z) at the interfaces. In all other instances however, many of the results obtained in chapter I can be applied to equations (35-37). In particular, the number of states bound by the well (at k, = 0) is equal to n wai sin | (2 v.23) | (38) In figure 5 are shown the variations of the confinement energies E,, Ey, E; at k,=0 of a Ben Daniel-Duke quantum well (La= 100A, V,=0.3eV, 300, ES S200) | = = V, = O3eV =007 3 mag « L=100A & 100 Ee a Ey 0 io 4 10 = 10? = 10808 mg/m m Fig. 5. — Evolution of the confinement energies E,, Ey, Es with the mass ratio om ina Ben a Daniel-Duke quantum well. L = 100 A; V, = 0.3 eV; mg = 0.07 mg.Hetorostructure electronic states 7 m my = 0.07 mg) with the mass ratio = . It can be seen that all the E;'s decrease with A ‘ Pa 5 increasing = and tend to values E;°,, which are such that : A ky (ES1)La=pt p =0,1,2.. 9) This equation resembles the bound states equation in a quantum well with an infinite barrier height. It expresses however quite a different physical situation. In a quantum well with infinite V,, the envelope function vanishes at the interface. Besides, according to equation (38) Wis infinite. If V, diverges, xs also diverges. This leads either to cos ¢ 4 = 0 (Eq. (35) or sin @, = 0 (Eq. (36). The ground state solution, which is nodeless, fulfils Laka(V,= 00 ) = 7, the associated envelope = . On the other hand, when V, is kept fixed but mg/m, increases to infinity, W” remains unchanged. Moreover, if the ground states envelope function is to be nodeless, it has to be a cosine in the well. It should barely penetrate the barrier (since xy increases) and, in addition, should have a function having a finite slope at z = derivative at z = whose modulus becomes smaller and smaller to comply with 2 the continuity of 4-2) & at the interface. Thus at infinite mg/m, ratio the only possible wave function is constant in the well and zero in the barrier, so that both the envelope function and its derivative are also zero in the barrier. The only cos (ka z) function which is constant corresponds to k, = 0 and thus to E, = 0... This is why when mg/m, diverges and V, is fixed the even states of the well fulfil sin g,=0 (and not cos ¢ 4 = 0), ie. admit E, = 0 as an acceptable solution. Symmetrically, the odd states fulfil cos g, = 0 and finally the series of levels (Eq. (39) is recovered. The ground state envelope function of a quantum well with cither infinite V, or finite V, but infinite ms is shown in figure 6 in order to depict the differences between the two physical situations. When the effective masses m, and my are not widely different, as for instance in GaAs-Ga(Al)As heterostructures, the in-plane dispersions of the subbands attached to the k, = 0 bound states of a quantum well are nearly parabolic in k, : we By (ls) 65 + By(0) + 5 (40) The in-plane mass m, should, in principle, be obtained by numerically solving equations (32-37). However when k, is smal] enough, an approximate scheme can be designed in the following way. The term f°k2/2y(z) in equation (27) is formally rewritten ; Wk RE KP L aL 2] 14 | (a) ue) 2m, 2 La@ m,8 Wave mechanics applied to semiconductor heterostructures VLX,(Z) Z/L Fig. 6. — Ground state envelope functions for a quantum well with infinite V, (dashed line) or m for a.quantum well with finite V, but infinite = (solid line). n and the second term on the right-hand side of equation (41) is considered as a perturbation to #, where Wai a, Wk 2 32 we) 02 * Im,” (42) whose eigenstates are in the form given by equation (40). The first order corrections to these eigenstates are given by: WEE 1 1 1 ae, = [2 1- Pel + tre) -2 | (43) where : ., RP PE) =2 fp, xed = & (4) z is the integrated probability of finding the electron in the barriers while in the E, state. The first order energy shift will vanish if : 1 1 = ig Po (End + 5 Po (En) (45)Hetorostructure electronic states 79 Equation (45) defines the in-plane effective mass of the n" subband in the vicinity of k, = 0. It may be remarked that if mg > ma, as is the case in GaAs-Ga(Al)As or Gap,71ngs;As-InP, this in-plane mass m, wil! increase with increasing subband index n. Using the approximately parabolic in-plane dispersion laws (equation (40) it is very easy to calculate the density of states p(e) associated with the bound states E,, Proceeding exactly as in chapter I we obtain: ole) =F pale) (46a) exe) = ms ¥(©-E,) (460) where Y(x) is the step function. We recover the familiar staircase density of states. The properties of a Ben Daniel-Duke quantum well are summarized in figure 7. ae & & | & MI Ey E ey 0 -Lot 0 (e) 2 7 a Fig. 7. — A recollection of the main properties of the quantum well bound states, solutions of a Ben Daniel-Duke Hamiltonian. From left to right : conduction band edge profile, energy levels E, and E, and their associated envelope functions ; in-plane dispersion relations of the E, and E, subbands ; energy dependence of the heterostructure density of states p(e). 11.3.2 Interface states of Ben Daniel-Duke quantum wells (mam, <0; k, = 0). — The case mamy <0 is practically realized in HgTe-CdTe heterostructures [26] (see Fig. 8). CdTe is a conventional open gap semiconductor whose level ordering is the same as is found in GaAs. HgTe is a symmetry-induced zero gap semiconductor. The I, band, which is a conduction band in most III-V and II-VI semiconductors, is a light hole band in HgTe. The I, edge lies ~ 0.3 eV below the I’ edges. As the Ts light band and I, band are nearly mirror-like, the I’, light band is a conduction band in HgTe, degenerate at the zone centre with the Is heavy hole band (inversion asymmetry splitting having been neglected). Ignoring the absence of centro-symmetry of the zinc-blende lattice, we shall see in section (II.4) that the light particle and heavy hole states decouple at k, = 0. We can80 Wave mechanics applied to semiconductor heterostructures fe & 3 - te & Me ‘Ay A B -L- f G5 oN HgTe CéTe Fig. 8. — Band structures of bulk HgTe (left panel) and CdTe (right panel) in the vicinity of the point (schematic). thus treat the problem of the light particle states associated with a I's edge as if we were considering a single band. The interesting feature of the HgTe-CdTe sheterostructure is that the light particle changes the sign of its effective mass across the interfaces, being electron-like in the HgTe layer and light hole-like in the CdTe layers. To be specific, let us consider a CdTe-HgTe-CdTe double heterostructure. According to [27] the bottom of the HgTe I’; conduction band lies at an energy A ~ 40 meV above the top of the CdTe Ig valence band. Thus, bound states of the heterostructure only exist if © = — A (the energy zero being taken at the Ig edge in HgTe). If — A < e <0, the states are evanescent in both kinds of layers while if « > 0, the carrier wavevector is real (imaginary) in the HgTe (CdTe) layers. Clearly, bound states of positive energies will exist (an infinite number in the one- band description of each host layer). Proceeding as in section IT 3.1 their energies willHetorostructure electronic states 81 fulfit for even states (47) cos eo, -—* “Bsing,=0 — for odd states (48) | ka (49) (50) -20 L | O 100 200 300 400 ° L (A) Fig. 9. — Evolution of the ground and first excited bound states (labelled 1 and 2 respectively) versus the HgTe slab thickness in a CdTe-HgTe-CdTe double heterostructure. The bound state wavefunctions are all characterized by cusps at the interfaces due to the change in the carrier effective mass at the hetero-interfaces. This sign reversal also implies that equation (48) can be fulfilled at ¢ = 0 for a certain La while equation (47) can not. This means that at least one state (even in z) should lie below the bottom of the HgTe conduction band edge. This state is an interface level, built from evanescent states in each of the host layers, whose wavefunction peaks at the82 Wave mechanics applied to semiconductor heterostructures interface. More precisely, we can write : xl) = A cosh (x42) lz} <3 La (51) xi@) = Bexp| a(2-5 La) | zehly (52) xi(~2) = x1 @) (53) 2m, with : Ka= [FAC-os kg= 2m Pll (e +A). (54) By matching x,(z) and 4~(z) oM atz=5 thas we find that ¢ should be the root of the implicit equation tanh (Genta) ~ tele (55) al Ka It is very easy to check that equation (55) always admits one solution E, (and only one) which extrapolates to — A when L, -» 0. A second state may actually exist in the energy segment [— A, 0] if the HgTe layer is thick enough. It corresponds to an odd envelope function : xo) = Asinh (42) 5 delshLa (56) x2) = Bexp[- kp (2-344) |: zehLa (57) X= 2) = — x22) (58) The E> energy is the solution of the implicit equation : my « cotanh (fata) = A ZB (59) Imig] a which admits a solution if 2 2 Ae [msl tt (60) my 2|mp|A Again, the solution of equation (59), if it exists, is unique. When L, becomes very large the energies E, and E) converge to the value : Ey =- Teal (61) m 14 ma which is the energy position of the interface state in a single HgTe-CdTe heterojunction (28, 29]. Clearly, at large La (i.e. «,L,>1) the two states E, and E are very well approximated by the symmetric and antisymmetric combinations of the two interface states centred at + xh respectively. TheHetorostructure electronic states 83 behaviour of E, and E, versus L, is presented in figure 9 to illustrate the previous discussion. In figure 10 we show the calculated y,(z) envelope functions in Hg, _,Cd,Te-HgTe-Hg, _,Cd,Te quantum wells to illustrate the interface nature of the E, state. Although the existence of the interface state relies only on the relative position of the I"; edges of HgTe and CdTe, their actual energy position, as well as their behaviour at k, #0 (where they strongly couple to the heavy hole states), remains a subject of active research. ‘ Eee) ' ; L=150A ' “1 0 1 Z/L oe Hg, 0d Tes AgTe Hg, ,04, Te Fig. 10. — Dimensionless envelope functions of the ground states in Hg, _,Cd,Te-HgTe- Hg,_,Cd,Te double heterostructures (x = 1 and x = 0.2) for two different HgTe slab thicknesses. 1.4 QUANTUM WELLS AND SUPERLATTICES WITH HOSTS WHICH DISPLAY KANE- LIKE BANDS. — This situation is found in most of the III-V and II-VI heterostruc- tures, e.g, GaAs-Ga(AIAs, GapalngsAs-InP, HgTe-CdTe, Gags7lmy s;As-Alpag Inps)As etc... At k, =0 the 8 x 8 differential system defined in equation (20) is block-diagonal if the mp) basis diagonalizes both the total angular momentum84 Wave mechanics applied to semiconductor heterostructures J and its z projection J,. Specifically we shall take mee fobs) =ush ts = |P re) = Fleemty uy = |P 345) = - fizn+% (x+y) uy = |? 14,3) = ier) + izty tee fbf) atist ug = |P Bo. -3) = Fle ilty we [Pbk] = gia - fpieb ty = |r 5 -3)- Fy lee Mt> +124) The differential system Dy = ex factorizes into : [oo |e] el] (62) (63) where D,, D_, x,,x_ ae 4x 4 matrices and 4 x 1 column vectors respectively. D, and D_ are identical and therefore each eigenenergy will be twice degenerate. BD, (and P_) are equal to: D, (27,=-im2) = v, 1p. F, 0 - FP, 1) + 5 Pe Fes wa ~ en + VG) 0 1p 4-29 0 0 ~ Sng? 1 ~ 2922 = en +V,(2) P 1 Pane Tat Pe 0 ~ Bagh (11 + 20D, mg PP oe P v2 =. TP VPs + Ve(2) ~ Ps 11 Ps AP 0 Wg Pt 72 e x 1 (64)Hetorostructure electronic states 85 where V,(z) and V,(z) are the step functions which describe the algebraic shifts of the I's and I’; edges when going from the A to the B materials and : i i i =— xX) =— Y) =— Z (64a) wig (51P#14) =F (Sles¥) =F SlpslZ) (64a) In equation (64) we have included the free electron term in the definition of F, Ys 2? 2 1 FoR Sil) aay 2 Aled) <¥ IrelX) +2(X|py|¥) Cv |Py|X? (¥|pe|S) +1 (64b) 155 ze) -V,@) “te (X1pelY) Cea XD — Appl) Ce ppl 7 Sng Poa—V.G) oo The explicit z dependences of F, 73, 72 (via the V ,(z) functions) has led us to use combinations like p, Fp., P; 71 Pz» P: ¥2 P; in equation (64) to ensure the hermiticity of D,. It is important to notice that the |P.3 #3 3) lines are uncoupled to the others in Zt equation (64). The associated envelope functions correspond to the heavy hole states which can therefore be treated separately from the other light particle states. The latter are hybrids of I", I’, light and I’; host states which are admixed by the non- parabolicity of the host materials and the V,(z), V,(z),V(z) functions (see Appendix B of the previous chapter). Notice however that the decoupling between heavy holes and fight particles is only approximate as, even at k, = 0, the effect of the non centro-symmetry of the host zinc blende lattices in equation (64) has been neglected. This effect is usually very small (smaller in III-V than in IJ-Vi compounds) and may eventually be treated in perturbation (see Appendix B). Its main effect is to prevent any crossing between light and heavy particle states. The latter would occur, for instance, in a quantum well when the confinement energy of a heavy hole state HH,, equals that of a light hole state LH, for certain thicknesses of the well-acting material. Notice that the lack of centro-symmetry appears automatically in tight- binding calculations [6] where the difference between the anion and the cation in the host’s unit cell is naturally taken into account. Across a A-B interface, the boundary conditions are such that x, (x_) and A, x, (A. x_)are continuous where A, equals A_ and re 0 0 0 az a 0 -(-2%) 5 0 0 As (23) = * a a | © Q 0 -(n +2 e 2V2ne 3 a 0 0 Wine -NE where we have made use of the x, continuity to simplify A,86 Wave mechanics applied to semiconductor heterostructures ‘As expected, A, only involves small terms, which arise either from the free electron contribution or from the k.p interaction between I"g, 7, 'g and the remote edges. This interaction is very important for heavy hole states since the heavy hole bands would be dispersionless in the hosts if this interaction was neglected. On the other hand, for light particle states, the F, Y,, ¥2 parameters contribute very little to the host effective mass. Their presence allows a much better fit of these effective masses than was obtained by retaining only a single parameter (P). In fact, we have to account for four band edge effective masses. With P, F, y, and y, we obtain four adjustable parameters, which is the required number. The reasons why we wish to discard the remote band parameters for light particle states are twofold. Firstly, without F, y; and y», the differential system for the three light particle components of x, becomes of the first order in 2 and therefore easier to handle, to the extent that all the results can be obtained in closed forms. Secondly, by retaining F, y, and 2, We face a difficulty termed by White and Sham [1] the “wing bands”, which are extensively discussed by Schuurman and tHooft [3]. Let us try to pinpoint the origins of this difficulty by using a simplified model : we shall neglect the I) band, i.e. we shall assume that the spin-orbit energy is very large compared with the light hole confinement energy. In each type of layer, the heterostructure states of a given energy ¢ are linear combinations of bulk solutions. These solutions are either propagating (ka, kgreal) or — evanescent (kg teal, kg imaginary) in the A and/or B layers. Thus in order to know our heterostructure states, we simply have to calculate the k,, ky corresponding to the energy ©. Therefore to calculate k,, for example, we have to find the roots of : Det [D. (p, = tka) - el} = 0 (66) In our three band (I'g,, Py, I's) model, equation (66) factorizes into WkK FRA Wk 2 a2 (eee [( Ime ) (2+ son +2) org] (67) The first root is obvious and describes the heavy, hole branch. At the moment however, it is of no concern to us. The two other ay roots, which are always real, deseribe the dispersions of either the coupled (I's + I'4) bands or the wing band (see Fig. 11). The former, which has a clear physical significance is i) small compared with the size of the A Brillouin zone along the Z axis and ii) almost insensitive to any change in the small parameters F and y, + 27>. ka is real for both e <— e, and e>0 and imaginary for —e,<©<0: the propagating light hole state (e<—e,) : becomes evanescent as © enters into the bandgap (-=*=-$) and e progressively transforms into an evanescent I’, electron (- zt 0.They prove troublesome however in the numerical computations. Therefore as the introduction of higher bands adds little physical information to the Tg, P;, 'g bands (except of course for the heavy hole band) and produces artefacts like the wing bands, we feel that it is reasonable to get rid of them wherever possible.88 Wave mechanics applied to semiconductor heterostructures 11.5 SIMPLIFIED CALCULATIONS OF SUPERLATTICE AND QUANTUM WELL STATES (k, = 0). — Atk, = 0, setting F, y,, 72 equal to zero and discarding the heavy hole line, we obtain a 3 x 3 first order differential system which can be transformed into a scalar, non-linear in e, second order differential equation. The latter is obtained by eliminating y5 and x; to the benefit of x, in equation (64) (and similarly in the D_ matrix). Proceeding as in Appendix B of the previous chapter, we readily obtain : 1 bee tA, Vee ]ee+ vate} xi) = exe 8) together with a similar expression for y,(z). The boundary conditions that X; fulfils at the A-B interfaces are such that 1 4x X1(z) and Wen are both continuous (69) where : 1 2p? 2 1 Stepp a mea 3 [2 Wi * oF ent ay 75 | (70) #(¢, z) is nothing other than the energy-dependent effective mass which appears in the Kane three-band model when the dispersion relations in each kind of layer are written in the implicit form mcs in the A I ea tA inthe A layer TAO) om V. fk in the B I: e-V,=*® inthe B layer ual) ys the energy being set at the I, edge of the A material. We are now in position to calculate the dispersion relations of an A-B superlattice. Let us consider a superlattice unit cell (thickness d = L, +L) containing two interfaces. At each interface we apply the continuity conditions given by equation (69). In addition, the band edges V,(z), V,(z), V(z) and the effective mass are periodic functions of z with periodicity d. Thus x\(z) may be written as a Bloch wave : X12 + d) = exp(igd) x 1(2) (72) with (73)Hetorostructure electronic states 89 Inside the A and B layers x, is a linear combination of incoming and outcoming plane waves xa(2) = @ exp(ikaz) + Bexp(—ikaz); EA (74) Xxu(2) = 7 exp(ikgz) + 8 exp(~ikgz); 7 EB Four linear equations are obtained (two boundary conditions per interface) for four unknowns («, B, y, 8). These equations can be satisfied only if the determinant of the associated matrix vanishes, which in turn leads to the superlattice dispersion relations. cos (4d) = cos (ka La) cos (Ky Ly) — 5 (¢ +t ) sin (ky La) sin (ky Lp) (75) . kx Ba (©) ith : a 76) “i HA) Ro “) 242 p2 24g e(e + eq) (e+e, + Aq) = PARP («+204 538) (7?) 272 p2 2 Ag (©-V.)(e V4 ep e—V, + ep + dp) = KAP (e-Yt S (78) Compared with the results obtained in chapter I, we see that the same kind of superlattice dispersion relations are obtained in idealized situations (a single band, a single effective mass etc...) and in the envelope function description of semiconductor superlattices. This is, after all, not very surprising. All our efforts have been put into neglecting the atomic-like details specific to solids. Some of the important features associated with them have, however, survived : i) the wavevectors k, and kg in equations (75-78) are related to the energy via expressions which are more complicated than those found in vacuum. This accounts for the multiband nature of solids. ii) €is no longer given by ka/kg but should be corrected by the effective mass ratio #p/Ma to account for the effective mass mismatch at the interface. IL.6 MISCELLANEOUS LIMITING CASES. 1.6.1 Evanescent propagation in one kind of layer. Quantum well bound states. — Suppose that V,>0, V, <0, Vs <0 as found in type I heterostructures. Thus for energies e such that OV,,