PHYSICAL REVIEW B 'VOLUME 29, NUMBER 4 15 FEBRUARY 4

Collective excitations of semi-infinite superlattice structures:

Surface plasmons, bulk plasmons, and the electron-energy-loss spectrum

R. E. Camley
Department of Physics and Energy Science, University of Colorado, Colorado Springs, Colorado 80907

D. L. Mills
Department of Physics, University of California, Irvine, California 92717
(Received 8 August 1983)

We consider a semi-infinite superlattice structure, with constituent 4 characterized by dielectric

constant €4() and constituent B by ‎‫(مأى‬, and examine collective excitations of the system in the
absence of retardation effects. Also, we explore the energy-loss spectrum of electrons backscattered
from such a structure. Detailed application is ‎‫ عقو عط ما‬where one constituent, a semiconductor or
metal, contains free carriers, while the second is described by a frequency-independent dielectric
function. Surface excitations (surface plasmons) ‎‫ ده‬adjacent interfaces couple through macroscopic
electric fields ‎‫ ما‬form a propagating band of collective excitations of ‎‫ عط‬whole structure capable ‎‫له‬
transporting energy normal ‎‫ ما‬the interfaces. We then find surface excitations of the entire struc-
ture. These are linear superpositions of modes localized at successive interfaces, combined with an
envelope function which decays exponentially as one moves down the stack. By explicit calculations
of the energy-loss spectrum, we show how electron-energy-loss spectroscopy may be used to study
these and other collective modes of the array.

1. INTRODUCTION ple the elementary excitations of the various layers.

Recently there has been considerable interest in the
properties of superlattices, which are structures composed
of alternating layers of different materials. Typically the
thickness of an individual layer ‎‫ كنا‬in the range
100—5000 A. If one constituent, material 4, always has
thickness ‎‫ مرك‬and the second, material B, always has
thickness d, one has built a periodic structure known as a
superlattice.
Superlattices composed of semiconducting materials
have been studied extensively. Here electrons may be
trapped on the near vicinity of the interfaces by band-
bending effects, to form a thin, two-dimensional gas.
More recently, metallic superlattices have been fabricat-
ed,! and their elastic properties have been studied. One
may also fabricate structures which consist of alternating
layers of magnetic and nonmagnetic metals.
Most studies of ‎‫ عط‬semiconducting superlattices sup-
pose ‎‫ عط‬material consists ‎‫ له‬parallel sheets of free car-
riers, each independent of the other. The superlattice
structure than allows access to the two-dimensional elec-
tron gas, under conditions where the average volume den-
sity is rather high, which is convenient from the experi-
mental point of view.
One may inquire if there are collective excitations of the
whole superlattice structure, with properties distinct from
those characteristic of either constituent. The answer is
clearly in the affirmative. Some years ago, Fetter’ con-
sidered the elementary excitations of an infinitely extend-
ed stack of two-dimensional plasmas. Excitation of a
plasmon within one constituent produces electric fields
which extend outside its boundaries, and these fields cou-
Through use of Bloch’s theorem, one then sees that a
consequence of this coupling is a set of collective plasma
excitations of the whole superlattice structure, character-
ized by a wave vector normal to the interfaces; these
10006 may thus transmit energy normal ‎‫ عط م‬layers of
the superlattice structure.
Recently, the collective spin-wave excitations of semi-
infinite magnetic superlattices have been studied theoreti-
cally,>* and a striking prediction emerges from these anal-
yses. First of all, on each interface between a ferromagnet
and its nonmagnetic partner, one has surface spin waves,
and surface spin waves ‎‫ ده‬adjacent interfaces couple
through dipolar fields generated by the spin motion, very
much as in Fetter’s study of plasmons. When the super-
lattice is terminated, of course each normal mode of the
structure (this is a linear combination of surface spin
waves, combined via the Bloch theorem to form a propa-
gating mode of the whole structure) may propagate to the
surface, to reflect off of it. In addition, one finds a sur-
face mode ‎‫ له‬the semi-infinite superlattice.>* This is a
linear combination of surface spin waves, combined with
an envelope function which decays exponentially as one
moves into the stack. This mode has the remarkable
property that its frequency is exactly that of the surface
10006 appropriate ‎‫ ما‬a semi-infinite ferromagnet, though
its wave function is clearly very different. If ‎‫ رك‬is the
thickness of ‎‫ عط‬magnetic constituent, and 42 that of the
nonmagnetic layers, the surface mode exists only when
dy>d,. As d,—d, from above, the surface mode merges
with the “bulk” excitations of the system, and is absent
when 4: <d,. Experimental studies ‎‫ له‬spin waves in these
matersials by Brillouin scattering confirm these predic-
tions.

29 1695 ©1984 The American Physical Society

1696 R. 2. CAMLEY
The present paper presents analysis of a semi-infinite
superlattice, which consists of alternating layers of materi-
al, where one or both constituent contains free carriers, or
possibly electric-dipole-active collective excitations such
as optical phonons or excitons. We inquire if a semi-
infinite structure of such materials may support a surface
excitation similar to the spin wave described in the preced-
ing paragraph, and find the answer to be yes, under the
conditions outlined below. We argue that one should be
able to observe such modes by means of electron-energy-
loss spectroscopy, and we present an analysis of the
electron-energy-loss spectrum in a backscattering geo-
metry to illustrate this point.
The paper is organized as follows. Section II presents
the theory for both the “surface” and “bulk” collective
modes of the structure just described, and we derive the
dispersion relations of the various modes. Numerical re-
sults are then presented for particular structures. In Sec.
111] we present the theory of the electron-energy-loss cross
section, and Sec. IV is devoted ‎‫ ما‬a summary of our prin-
cipal conclusions, including the numerical studies of the
electron-energy-loss cross section.
II. DISPERSION RELATIONS AND GENERAL
PROPERTIES OF ELECTRIC-DIPOLE-ACTIVE
COLLECTIVE EXCITATIONS
OF SUPERLATTICE STRUCTURES
1t is useful ‎‫ م‬break the discussion of the present section
into several parts. We consider first a review of the col-
lective modes of an isolated dielectric slab, then turn to a
superlattice structure of infinite extent, and next the ter-
minated superlattice. Then we proceed to some numerical
examples.
A. Isolated slab
For what follows, it will be useful ‎‫ م‬consider those col-
lective excitations of an isolated dielectric slab (possibly a
metal) which in the long-wavelength limit generate a mac-
roscopic electric field. Then if we confine our attention to
only this limit of long wavelengths, the mode structure is
described fully by macroscopic theory. Suppose the slab
has thickness d, vacuum above and below, and has a
frequency-dependent dielectric constant €(w), assumed
real in this section. The macroscopic electric field associ-
ated with the elementary excitations of interest may be de-
rived from an electrostatic potential ¢(X,t); throughout
this paper, we ignore retardation,® so @(X,r) obeys
Laplace’s equations everywhere outside the slab,
V2(%,1)=0 . (2.1a)
Then inside the slab we have
€(w)V?p(X,1)=0 . (2.1b)
In general, an excitation in the slab will set up a fluc-
tuating electric field in ‎‫ عط‬vacuum above and below it.
We denote the electrostatic potentials in these two regions
by ¢> and ¢ <, respectively, and that within the slab by ¢;.
Translational invariance parallel ‎‫ م‬the two surfaces en-
sures that all elementary excitations are characterized by a
AND D. L. MILLS 29
two-dimensional wave vector K parallel to the surfaces.
‘Without loss of generality, we may chose K directed along
the X axis. Then the electrostatic potential everywhere
has the form
B(X,1)=D(z)e’ k") | 22»
We then have ‎‫ عط‬following two distinct sets ‎‫ له‬elementary
excitations which generate a macroscopic field.
(i) Surface excitations. An isolated dielectric-vacuum
interface supports a surface excitation with a frequency
independent of wave vector k, at any frequency w, for
which e(w,)=—1. If the interface coincides with the
plane z =0, the electrostatic potential falls to zero ex-
ponentially, as one moves away from the interface in
either direction. One has ®(z)~exp(—k |z |) for such
modes.
The finite slab, considered very thick for the moment,
supports two such modes, one on each surface. With a
thickness finite, the two modes couple to produce an odd-
or even-parity pair, split by interaction between the two
surfaces. We thus have dispersion relations »_(k) and
‎‫ اتات‬respectively.
The ‎‫ه‬- mode is described by the implicit dispersion re-
lation
elw_)=—coth (7kd) . (2.3a)
If elw)=€,+0}/(wj—w?), then as kd— oo, o_(k) ap-
proaches ‎‫ عط‬single-interface surface-mode frequency
w;=w5+Q;/(1+¢€,), while as kd—0, ©_(k)—aq, where
€(w) is infinite. If we have a metal, then wy=0, and so
©_(k)—0 as kd—0, with o_(k)=0,(3kd)'> when
kd <<1. This mode is an odd-parity mode in the sense
that the electrostatic potential is odd under reflection
through the midpoint of the film.
The ‎‫ به‬mode has the implicit dispersion relation
(2.3b)
(o, )=—tanh(Tkd) .
Then as kd— o0, @, (k) also approaches ‎‫ونه‬, while as
kd—0, (k) approaches the frequency for which e(w)
vanishes. This is ‎‫ عط‬LO-phonon frequency in a simple
ionic insulator, or the bulk-plasma frequency for a metal.
The electrostatic potential has even parity under reflection
through the midpoint of the film.
The results above are well known, and may be derived
in a few lines by noting that ¢(Xt) satisfies Laplace’s
equation everywhere, including inside the film [because
€(w)70]. Then by requiring that ¢ ‎‫ عط‬continuous at each
interface along with normal components of ]3=€(w)f3,
constraints which lead to the dispersion relations follow.
(ii) Bulk excitations. ‎‫ هط‬infinitely extended dielectric
medium supports bulk excitations of longitudinal charac-
ter (plasmons, LO phonons, and longitudinal excitions) at
those particular frequencies where €(w. . Here we are
no longer constrained to have V2$=0 within the medium,
since €(w) vanishes in Eq. (2.1b). For ‎‫ عط‬slab, elementary
arguments given elsewhere’ serve to delineate the proper-
ties of ‎‫ عط‬bulk excitations of a finite slab. One has a se-
quence of standing-wave resonances of the slab, each with
frequency such that e(w)=0, i.e., ‎‫ عط‬frequency ‎‫ له‬each
29 COLLECTIVE EXCITATIONS OF SEMI-INFINITE SUPERLATTICE STRUCTURES: 1697

slab resonance is identical to that of the relevant bulk ex- and must obey appropriate boundary conditions at each
citation. The boundary conditions require ¢>(Xt) and interface.
¢<(%,t) vanish identically, so ‎‫ عط‬modes generate no mac- Since we have translational invariance in the two direc-
roscopic electric field outside the slab itself. If the slab tions normal to the 2 axis, each normal mode is character-
surfaces are located in the planes z=0 and z=d, then ized by a two-dimensional wave vector K parallel to the xy
®(z)~sin(nmz/d), so there is a macroscopic field of plane, as in the case ‎‫ عط أله‬isolated slab. We assume each
standing-wave character within the slab. These con- material is isotropic, so that without loss generality, K
clusions will be altered by the presence of spatial disper- may ‎‫ عط‬taken parallel ‎‫ م‬the x axis. Thus, ¢(x,y,z,t) will
sion effects not examined here; these will be small in the have the form
long-wavelengths limit and in.a number of geometries of
practical interest. B(x,9,2,)=D(2)e’kx—o0) | 2.5)
and Eq. (2.4) leads to
B. Superlattice of infinite extent 2
2 )0)2(-0. 26
‘We consider here the structure shown in Fig. 1. Materi-
‎‫ لا‬A has a frequency-independent dielectric constant
Quite clearly, the general solution of this simple equation
€4(w), and thickness ‎‫ مرك‬while material B has a dielectric
is
constant €g(w) and thickness d,. For the purposes of ‎‫عط‬
present discussion, €4() and €z(w) will be supposed real; D(z)=A‫‏‬ ‫اتا _ ربك‬ 27
this restriction is relaxed later in the paper. The “unit
cells” of ‎‫ عط‬structure are designated by the index n, as il- Since the structure in Fig. 1 is periodic in the 2 direc-
lustrated in the figure. tion, our task is to synthesize the basic solutions in Eq.
As we saw in the preceding subsection, an isolated slab (2.7) so that proper boundary conditions are satisfied at
of material A or B has standing-wave resonances (bulk LO each interface, and so that the solution forms a Bloch
phonons, bulk plasmons, etc.) which generates a macro- wave, with respect ‎‫ ما‬translations normal ‎‫ عط ما‬interface.
scopic electric field at any frequency for which €4(w) ‎‫عه‬ Thus, if L =d, +d, is ‎‫ عط‬length of a unit cell, we require
€p(w) vanishes. The electric field associated with such ex- that ®(z) be written in the form
citations is totally confined within the slab, so when a su- D(2)=e*U,(2) , 2.8)
perlattice such as that in Fig. 1 is constructed, each con-
stituent still possesses bulk resonances identical to those in where for any integer n,
the isolated film. These are unaffected by the fact that
the film in question is now incorporated into a superlattice
Uy(z+nL)=U,(z) . 2.9
structure. First consider ‎‫ عط‬form of the electrostatic potential
Here we study the collective excitations of the whole within the nth slab of material 4, which extends from
structure which have frequency o such that neither €4(w) z=nL to z=nL +d;. One readily verifies that the most
nor €g(w) vanish. The electrostatic potential ¢(X,?) ‎‫معط‬ general solution of Eq. (2.6) which also satisfies Eq. (2.9)
satisfies Laplace’s equation everywhere, may be written
V2(,1)=0, (2.4) 1‫)ول‬2( ‫ هاا و‬- ‫ اله تامور بر) لله‬4 4 o—klz—nL))‫‏‬
nL<z<nL+d; (2.10)
80 that we have
D(z)=e""(4
‎‫ ") تامو‬g _o—klz-nD)) @11
€plw) A 0
In this discussion, ‎‫ و‬is a wave vector that will ultimately
€p () 8 ‎‫ووأ‬
6
0

enter the dispersion relation of the collective excitations of

the superlattice.
Similarly, one may write the most general form of the
€g(w) B scalar potential within the nth layer of medium B, which
extends from 2 =nL +4 to z=(n +1)L. We have
A

€g(w) 8 ‫ روات‬- ‫و إلا‬ TN‫‏‬ ‫_ ور‬ ‫لصتا‬ )1

There are four arbitrary constants, ‎‫ با‬,_4 B, and
B_, which appear in the scalar potential. To constrain
these and obtain an explicit dispersion relation for the col-
lective excitations we require that boundary conditions at
FIG. 1. Infinitely extended superlattice. The structure is the various interfaces be obeyed. These are that the elec-
made of alternating layers of material 4 and B, each character- trostatic potential be continuous across each interface,
ized by the relevant dielectric constant. The unit cells of the along with normal components of D. Through the use of
structure are indexed by n, as illustrated. the above forms, we apply these at z=nL and z=nL+d,,
1698 1 2. CAMLEY AND D. L. MILLS 29

and the Bloch property of the basic solution insures they collective excitations discussed here may be viewed as a
are satisfied everywhere else. linear superposition of surface polaritons, one localized at
Continuity of ®(z) at z=nL +, along with normal each interface in Fig. 1. When 4: and d, are finite, they
3, gives the two constraint equations couple because their fields overlap, resulting in normal
modes of the whole structure. Their nature is controlled
‫اعرد‬ ‫بد‬ ™™ _p,‫الإ ‏‬ (2.13 by Bloch’s theorem, which dictates the form assumed by
the resulting electrostatic potential.
and‫‏‬
Suppose kd; >>1 with kd, finite, i.e., let d;— o with
‫امات‬ —d_e M =ey(0)B,—B_), (.14)‫‏‬ d, fixed. Then one readily sees that ‎2‫و‬. )91.2( degenerates
to
while the same conditions applied‫= ع عد ‏‬nL give‫‏‬

A +A_=e~(B
" 1 B_e ™) @.15) ‫دا‬ coth )4/42(‫‏‬
32.2
and )@(4€ tanh ($kd,) , @23
‫— اانا‬A_)=eglwle
(B‫‏‬ ‫دعر‬ ,)™ the generalization of Egs. (2.3) ‎‫ ما‬the case where a finite
)61.2 slab of material B of thickness ‎‫ وك‬is embedded in medium
A. Of course, Eq. (2.19) is symmetric under interchange
By setting the appropriate 4X4 determinant formed of d; and ‎‫ ريك‬so that if ‎‫يك‬co‫ ج‬with ‎‫ رك‬finite, we obtain
from Egs. (2.13)—(2.16) ‎‫ ما‬zero, we obtain an implicit the dispersion relation of the surface modes of a slab of
dispersion relation for the collective modes of the super- medium 4 with thickness ‎‫ رك‬embedded in medium 2.
lattice. This reads Explicit expressions for the dispersion relations may be
€4(@) 2 . . obtained for a variety of special cases. For example, let
1+ |] | |sinh(kd,)sinh(kd,) 2
‎‫(هاو‬ )( @
es(0)=€)+—"—‫‏‬ ‫م‬ )224
‫ماب‬ ‫هوه‬
2+ ‫ ررم وج‬005] 34 )cosh(kd) —cos(gL)]=0 .‫‏‬ so that material 4 is possibly a polar semiconductor or a
nearly-free-electron metal (w,—0). Then let 55(10):5'5” )
2.17١
corresponding to a nonpolar semiconductor or an insula-
In general, this implicit dispersion relation must be tor. Then we have
solved numerically. For fixed k, this yields the frequency
o as a function of g, the wave vector of the collective exci- 2 2 e
wF(k,q)=wp+ %5
el s v (2.25)
tation in the direction normal to the stack. If we define
cosh(kd)cosh(kd)—cos(qL)
The model considered above has three characteristic
c(k,q)= frequencies. The first is ‎‫وه‬, which is the TO-phonon fre-
sinh(kd; sinh(kd ) ’ @19 quency if we view material 4 as a polar material. Then
we have the LO-phonon frequency of material 4, given by
a positive definite quantity for any choice of 1 or g, then
m§+flz/ef4"’, and finally there is the frequency
Eq. (2.17) admits solutions whenever
o3 =0f+02/(€° +€5”), the frequency of the surface
‫ماب‬ polariton which propagates on the single interface between
<< 2/111-(96122]+(961)- ٠ (2.19) a semi-infinite half-space filled with medium 4, with the
»‫(وهاو‬
other side formed from medium B. From Eq. (2.23), one
For the frequencies which emerge from Eq. (2.19) to be sees the w,(k,q) branch lies in the frequency regime
real, we must append the condition ‎‫ه> وه‬- ,< while ‎‫ هه عط‬branch ‎‫ كنا‬in the regime
‎‫> وه‬- > ‫مره‬. If medium 4 is a metal, then ‎‫وجوه‬, ‫قو‬
clkg)>1. (220
the surface-plasma frequency, and ‎‫ مرنه‬becomes the bulk-
It is then convenient ‎‫ ما‬parametrize ¢ (k,q) by writing plasma frequency of material A.
clk,q)=coshy(k,q) , (2.21) C. Semi-infinite superlattice; excitations localized
where, when Eq. (2.20) is satisfied, ‎‫ لوقتال‬is real. The im- near the surface
plicit dispersion relation in Eq. (2.19) then reads ‘We now consider the geometry illustrated in Fig. 2. We
‫ماب‬ have a superlattice structure identical to that in Fig. 1, but
ezlo
=—exp[+ylkq)], (2.22) the structure is terminated at the plane z=0, with the
half-space 2 > 0 filled with material that has a dielectric
so that our collective excitations occur only in frequency constant € ().
regimes where ‎‫ عط‬ratio €4(w)/€ep(w) is negative. As we remarked in Sec. I, terminating the superlattice
The frequency regime where €,4(w)/€p() is negative is, structure destroys its periodicity in the 2 direction, so that
in fact, the spectral region where surface polaritons may we no longer have collective excitations with Bloch char-
propagate on the interface between medium 4 and B. The acter in the 2 direction, as described by Egs. (2.8) and
 et‫‏‬ ‫احا‬ ‫ اس‬,229

‫م‬ 2
‫سم يم‬ 5‫و‬
€p‫‏‬
€, —
+ 1+E—‘}(e Mi_e-fLg_—0. (30)
FIG. 2. Semi-infinite superlattice. The structure is made of B
alternating layers of material 4 and B as in Fig. 1, but the struc-
ture is terminated with the half-space z > 0 filled with material We have dropped explicit reference ‎‫ ما‬the frequency
C. dependence of the dielectric constants for convience. For
Egs. (2.29) and (2.30) to be satisfied with 4, and 4_
nonzero, the appropriate 272 determinant must vanish,
(2.9). There will surely be modes which may ‎‫ عط‬viewed as leading to a constraint ‎‫ ده‬the attenuation constant 8. One
propagating up to the surface from z=+ ‎‫ رم‬then reflect- must have
ing back into the superlattice structure, with a reflection
coefficient that may be obtained by a suitable extension of cosh(BL ) =cosh(kd )cosh(kd,)
‎‫ عط‬discussion in the preceding subsection. Here we will
1 €4
not be concerned with these modes, but instead with a new
class of solutions that emerge. These modes have their ex- *2 —+Zi sinh(kd,)sinh(kd,) ,
A
(231)
citation localized in the near vicinity of the interface be- a relation equivalent ‎‫ م‬Eq. (2.17) with ‎‫ و‬replaced by 8.
tween material C and the superlattice. One may view
Further constraints are obtained by requiring the boun-
these new excitations as linear superpositions of surface dary conditions ‎‫ ما‬be satisfied at the interface z=0.
modes localized to a particular interface, with an envelope These give simply
function that decays exponentially as one moves into ‎‫عط‬
stack. C=A,+A_ (2.32)
The surface modes may be described by taking, for the
electrostatic potential in the slab of material 4 in the re- and
gion nL <z <nL +, ‎‫ عط‬form €C=¢q(d,—4_), (2.33)
‫ ون‬P4 kL) L g‫السسرصن_ ‏‬ )2.26( or with C eliminated, we have a new equation which in-
while in medium B, in the region nL +d; <z<(n+1)L, volves 4, and 4 _,
we shall have
1+€
€,
k(z—nL —dy) —klz—nL—d;) 1-4 4+ 4_=0. 2.34)
‫(عات‬- "‫ل‬3
e +B_e ٠١ @21‫‏‬ 6 e

Recalling that k is ‎‫ عط‬magnitude of the wave vector We have three unknowns, ‎‫ با‬A_, and B, and ‎2‫و‬5.
parallel to the surface [Eq. (2.5)], the attenuation constant ,)92.2( ,)03.2( and (2.34) provide three constraint equa-
B will be determined in the subsequent analysis. Of tions. Upon combining Egs. (2.30) and (2.34), we have
course, we require Re(8) > 0.
In the region 2 <0, where material C resides, we have e~PL—e ™[ cosh(kd,)+Pysinhkd, ] , (2.35)
where

D(z)=Ceth . (2.28) ‎‫حم‬ &€ —epe, ,

€4(€,—€p)
In ‎‫ تناه‬applications of ‎‫ عط‬theory, we shall take €.(w)=1 while we may also combine Eq. (2.29) with Eq. (2.34) to
for simplicity, although the general formulas obtained find a second relation for exp(—BL),
below allow this parameter to take on any desired value.
We now turn to the boundary conditions. The four
conditions at the interface z=nL and 2 =nL +d, provide
‫و‬5 ‫ و‬+ [ cosh(kd, )+ Pysinh(kd ;)] , (2.36)‫‏‬
a set of relations for ‎‫ا يبا‬, B, and B_ identical to where
those in ‎2‫و‬5. ,)61.2(—)31.2( except ig is replaced every-
where by —B. It is convenient ‎‫ ما‬eliminate B, and B_
from these equations in order to reduce the set to two ‫وو‬
(2.37)
equations which involve only 4 , and A_. We have, once eqle.+ep)‫‏‬
1700 R. E. CAMLEY AND D. L. 111115 29

The right-hand side of Eq. (2.35) must equal the right- for the surface wave. Once this is solved, we must then
hand side ‎‫ له‬Eq. (2.36), so that we have the final con- check the value of B as found, say, from Eq. (2.35), to be
straint certain that Re(8) > 0.
At this point, we may turn to a number of special cases.
Let us suppose for definiteness that ‎‫ طامط‬5€ and €, may be
2 cosh(kd )sinh(kd) regarded as frequency independent in the spectral regime
of interest, while it is €4(w) that varies with frequency.
‎‫رط‬ .0=)
+sinh(kd,)[Pye" )832( Then after some manipulation, Eq. (2.38) may be used to
Equation (2.38) constitutes an implicit dispersion relation generate a quadratic equation satisfied by €, (w),
1
sinh(kd )[egcosh(kd;)
- ‎‫ المصتقي‬61 76[(2 )( + }€( —€2)cosh(kd ‎‫ ستول ر‬21 e 4 (@)
+ege sinh(kd )[epsinh(kd,) —€.cosh(kd,)]=0, )2.39(
so that the possible frequencies of the surface modes are such that
-1
‫ لهام‬26011 epooshikd,) —<,sinb(kd;)]‫‏‬
X (€} —€2)cosh(kd )sinh(kd, )+ { (€} — €2 cosh?(kd )sinh*(kd)

+ ‎62‫و‬ ‫قب‬4‫ص‬
:1 ; [) 2€p€.cosh(2kd, ) — (€} +€2)sinh(2kd,)]}'72) . (2.40)

A particularly striking special case is that for which
‎‫و‬6. For a semi-inifinite stack of films formed from a
“surface active medium” (i.e., one with a negative dielec-
tric constant in certain spectral regions) separated by vac-
uum, with vacuum above, we have €z =€, =1. Then Eq.
(2.40) reduces to the simple pair of statements
€4(w)="tep . (2.41a)
If we choose the upper sign, we see that ‎‫— حم‬k, an unac-
ceptable value. However, if
esl0)=—¢€p, (2.41b)
we have
4-4
B =k7 (2.41c)
which is quite acceptable if d; <
The condition in Eq. (2.41b) is precisely the same as that
which determines the surface-polariton frequency, in the
neglect of retardation, on the interface between a semi-
infinite slab of material B joined with a semi-infinite slab
of material 4. The penetration constant 8 differs, howev-
er, in that one has =k when d, =0, and only material 4
resides in the upper half-space z > 0. For our problem, we
see that B decreases ‎‫ ما‬zero as d,—d; from below, and no
surface wave exists when d, >d;. This mode is a precise
analog to the surface spin wave of the magnetic superlat-
tice discussed in earlier papers,™* and which has also been
studied experimentally.#> We shall explore this wave fur-
ther in the next subsection, where its relation ‎‫ ما‬the bulk
excitation spectrum of the superlattice structure will be
elucidated.
D. Some explicit examples of superlattice
excitation spectra
Here we present some numerical studies of the disper-
sion relations of collective excitations in semi-infinite su-
perlattice structures. In what follows, we suppose
€4(0)=1—0}/w?, with w,=15 eV. This corresponds to
a model of aluminum. For €,, we choose unity, so that we
have a semi-infinite stack of aluminum films separated by
a dielectric space with a dielectric constant ‎‫وى‬.
In Fig. 3 we show dispersion curves where d, =2d,, and
€p=1. We see that the bulk excitations described in Sec.
113 fall into two bands separated by a gap. Note that as
one scans through the frequency spectrum of bulk modes
with a finite wave vector K parallel ‎‫ ما‬the surface, the
modes tend to crowd together, forming a high density of
states near the boundary lines with gL ‎‫جح‬.
In Figs. 4(a) and 4(b), for two values of the parameter
kd,, we illustrate frequencies of ‎‫ عط‬collective modes ‎‫له‬
the semi-infinite superlattices as a function of the ratio
dy/d,. The o, and ‎‫ هه‬bulk bands [Eq. (2.23)] are al-
ways separated by a gap, except for the special case
dy=d,, where their lower bounds just touch. The surface
mode always lies within this gap, but it exists only when
dy > d,, as one sees from our earlier discussion.
A diverse variety of collective-mode spectra may ‎‫عط‬
found, depending on the nature of the two constituents.
Figure 5 illustrates the effect of an insulating spacer
placed between the Al model films. We still have e, =1 in
this figure, but now €p=3, a reasonable value for ‎‫ميلم‬
or some other wide-gap insulator. For large values of kd,,
‎‫ طامط‬the w, and w_ bulk modes are down-shifted in fre-
quency by virture of the screening provided by the oxide.
For large values ‎‫ له‬kdy, ‎‫ عط‬surface branch lies near the
surface-mode frequency appropriate to the Al/vacuum in-
terface, 10.6 eV. Thus, in contrast ‎‫ ما‬the earlier example,
the surface mode lies above the ‎‫ ب‬bulk-wave branch. As
29 COLLECTIVE EXCITATIONS OF SEMI-INFINITE SUPERLATTICE STRUCTURES: ... 1701

at smaller values of kd;, we find a surface mode that ‎‫كنا‬

15.0 in the gap between the w_ and ‎‫ ذه‬bulk branches. This
mode also merges with the @ continuum at a wave vec-
tor very close to that where the high-frequency branch
does. On ‎‫ عط‬low-frequency surface-mode branch, BL is
2.0
purely real, and vanishes as the critical wave vector is ap-
; [_SURFACE MODE proached from below.
K E In contrast ‎‫ عط ما‬simple case where ez=¢,, when
>o 9.0 L ~ €p>€, we find surface waves of ‎‫ عط‬structure when
z 4
4: <d,. This is illustrated in Fig. 6, where, for the choice
w 42 - +4, we give the excitation spectrum of ‎‫ عط‬superlat-
2
tice structure with ‎‫ لحي‬and eg=3. We have a high-
2@ 60 5 frequency branch of surface waves, again with B=im+X.
- This branch behaves in a manner qualitatively similar to
the upper branch in Fig. 5. We also have a second surface
wave in the gap between the w, and w_ bulk branches
3.0 -1 again (with 8 real as before), but now the branch ‎‫ عمل‬not
merge with the bulk excitations as kd, increases ‎‫ ما‬a criti-
cal value. As far as we can tell, the surface mode survives
0.0 1 1 1 1 in the limit ‎‫جرم‬ ‫ رم‬always “trapped” between ‎‫ عط‬two
0 1 2 3 4 5 bulk branches. Quite clearly, we have a rich spectrum of
kd elementary excitations in these structures, depending on
FIG. 3. For the case where material A is aluminum, and both the nature of the constituents and their relative thickness.
dispersion relation of the surface mode of the semi-infinite su-
perlattice, and those of the bulk modes described in Sec. II B of 111. INELASTIC ELECTRON SCATTERING
the text. FROM COLLECTIVE EXCITATIONS
OF SUPERLATTICES

kd, decreases, this mode merges with the w-mode con- In Sec. II we presented ‎‫ عط‬theory of the collective exci-
tinuum at kd = 1.5, for this choice ‎‫ اه‬.24 On this branch, tations of superlatlice structures. We next turn ‎‫ ما‬the
we have BL=im+X, were X assumes ‎‫ عط‬value 6.37 at question of how their properties may be studied. If we are
kd;=5.0, then decreases monotonically ‎‫ م‬reach zero as concerned with metallic superlattices, possibly with insu-
the surface mode merges with the bulk continuum. Then, lating spacers between adjacent metal films, then small-
(eV)
(eV)

FREQUENCY
FREQUENCY

d,/d,
FIG. 4. For the case €5 =¢, =1, we show, for (a) kd =1.0 and (b) kd =0.1, the frequency spectrum ‎‫ كه‬the bulk & and w_ bands
of ‎‫ عط‬semi-infinite superlattice along with the surface mode. The surface mode is absent when ‎‫ ررك > رك‬as discussed in the text.
 2071 R. 2. CAMLEY AND D. L. MILLS 29

back scattering off of the surface of a simple metal, the

electron interacts with ‎‫ عط‬excitations in the substrate
through ‎‫ عط‬fluctuating electric fields in the vacuum out-
side the substrate. The theory of the loss cross section
under this circumstances was discussed some years ago for
scattering off ‎‫ عط‬surface ‎‫ له‬a semi-infinite material,® ‎‫عه‬
for a material upon which an optically active overlayer is
present.'® A summary of this theory, along with a num-
ber of applications of the resulting formulas, may be
found in a recent book.!!
w (eV)

Here we wish to extend the theory to the superlattice

structures discussed in Sec. II. We could ‎‫ مل‬so by follow-
ing the quantum-mechanical analyses given earlier, but
Schaich'? has used a simple classical trajectory procedure
which reproduces one limiting form of the results provid-
ed by the full analysis. One must note that this limiting
form follows froni a full treatment only after certain as-
sumptions are introduced;!! these are ‎‫ له‬questionable va-
lidity when the energy loss of the electron is in the range
of several electron volts, but the classical trajectory
analysis yields a formula which surely contains the princi-
pal features expected in the data.
kd, The trajectory analysis proceeds by noting that the in-
FIG. 5. For ‎‫ عط‬case €3=3 and €, =1, we give ‎‫ عط‬collective- coming electron polarizes the substrate, here viewed as a
mode frequencies as a function of wave vector for the case dielectric medium. The induced polarization produces an
dy=2d,. Material 4 is aluminum, modeled as described in the electric field which does work on the electron as it ap-
text. proaches the crystal, and then exits after reflection off ihe
surface. One calculates the total work performed by the
angle inelastic electron scattering may serve as a suitable induced field to obtain the total energy loss suffered by
experimental probe since it has been used to study surface the electron, and an appropriate decomposition of this ex-
excitations ‎‫ ده‬metal surfaces.® Since the losses of interest pression yields the energy distribution of those electrons
lie in the range of several electron volts, very-high- which suffer an inelastic scattering.
resolution techniques are not required. In this section we Let the electron trajectory be described by
present the theory of electron-energy loss in a reflection X(1)=T% (1) +£2(1), with X the projection of X onto a
geometry for small-angle inelastic scattering from a semi- plane parallel ‎‫ م‬the surface. We may calculate the elec-
infinite superlattice. tric field generated by the induced polarization by solving
In the regime of small-angle deflections, where surface- Laplace’s equation ‎‫ ها‬the vacuum above ‎‫ عط‬sample,
plasmon contributions dominate the loss spectrum for
V2$(X,1)=—dme 8(X) — X (1)8(z—2(1)) , 6.0
14 T T T T T
with the electrostatic potential subject to appropriate
dz=2d, boundary conditions at the surface. We ignore the fact
that ‎‫ عط‬electron penetrates into the substrate. In the sub-
‎‫ا‬ 4 strate, since there is no free charge, we have
V2(X,1)=0. 3.2
or w, BRANCH 3
Also, ‎‫نار‬ ‫ اتح‬for - ‎‫ > م‬2 > + ‫ رم‬with V| the projec-
3 8 - tion of the electron’s velocity on a plane parallel to the
< surface.
3 ‘We may write
0 4

2 0
w_BRANCH
0 4 ‫[ حص‬ 8 )3.3(
2
2 1 1 1 1 1 and one easily that ¢(6”,z;t) satisfies
0 i 2 3 4 5
kd,
2 N 7‫‏‬
FIG. 6. For e;=3 and e,=1, we give the superlattice ‫معرةم| ادج‬ ‫ وم‬Qe‫‏‬
collective-mode spectrum as a function of wave vector for
dy=3d,. Again, material 4 is aluminum. 4.3
29 COLLECTIVE EXCITATIONS OF SEMI-INFINITE SUPERLATTICE STRUCTURES: 1703

in the vacuum outside the material and the homogeneous ‘where

version of Eq. (3.4) within the material. If the electron
‫ب‬ + ‎-Q,d,
‫قرحي‬ —pL, 2
strikes the surface at ¢ =0, then F= , (3.13b)
€4—€p ‫و‬912 d, _ o - d; %1 ‫‏‬
vt fort<0,
z()= —v,t fort>0. 3.5 with B determined from Eq. (2.30), after k is replaced by
92. The condition 168 < 0 is appended.
As in Sec. II, ‎‫ عط‬sample occupies the half-space 2 > 0. If From Egs. (3.12), we then find an expression for C,
we write —eA(F—1)
c, =RA . (3.14)
#(Qyz;0)= ff: ‎‫ا‬ ‫ا‬ 0.8 T (e4—DF—(g4+1)
It is a straightforward matter to use the electric field
one finds that da(a”,z;m) obeys generated from the term C., exp(Q)(z) in Eg. (3.10) to cal-
culate the work done on the electron by the induced field.
d? = 8 .9
‎‫ نمكي‬$(Q),z;0)= —Tl"gws (a)——v”‘Q”);zl— | s The work performed by this source has a magnitude
102.2 ‫م‬ 1 ‫ال‬
3.7 ‫ا‬ _ٍ ‫ا‬ ‫ال‬ ‫م اح‬ 01
)9 ]
‫زر‬
]‫رعره‬ ‫ل‬2)
and the homogeneous form in the medium.
A particular solution of Eq. (3.7) in ‎‫ عط‬half-space z >0 )51.3(
is ‘where‫‏‬
€4(0)(F—1)‫‏‬
¢,16,|,Z;0))=A(6“,mlcos (a)—VH-(')“)DL} ) 3.8)
1 R = T‫‏‬ 711 -]‫[جدهاي‬
where
T
‫و‬ ‫اندر‬
‎1+‫)ل(مابي‬7+1)/(27-1(
810
872
AQe)= 6.9)
‎)‫ه‬-79‫ار‬il
*+ The expression for the energy lost by the electron may
‎‫ عط‬cast into the form
We have ‎‫ ما‬append to Eq. (3.8) a general solution ‎‫ ما‬the
homogeneous form of Eq. (3.7), and then match the re-
sulting form to the solution of Laplace’s equations in the W= [ ‎‫لماطمقمة‬, 718
sample, so that the proper boundary conditions are obeyed
at each interface. We shall suppress reference ‎‫ م‬the
where P(w) is interpreted as the probability per unit fre-
dependence of various quantities on 6” and o for brevity.
quency that the electron has lost the energy #iw. We then
Thus, in the region z <0 below the sample, here assumed
to be vacuum (e, =1 in the notation of Sec. II), we have
have, noting that Im[R (Q)},)] is an odd function of ‎‫عط‬
quency,
9l2)=A cos ‎|)‫م‬- ‫] كار‬+c+e°fl', (.10
e} 22 Im[R(Q),@)
Pla)=22 212y Im(R )2 ‎‫امم‬ )81.3
1

while, as in Sec. II, the most general form of the potential 72 - 0,0( P+ (0= QPP
in the outermost superlattice layer is
Upon noting that R(Q),®) _(_lepends only on the magni-
‫د‬ ‫لير‬ ‫ ا‬27 0224 3.11 tude and note the direction of Qj, we may write

The term proportional to + in ‎2‫و‬. )01.3( generates the ‫ل‬

induced field which does work on the electron. ‫(ماط‬-‫ [ ع‬doyofimIR(Q),)]‫‏‬
Continuity of the total potential and normal component
of D at z=0 gives ‎‫ عط‬two conditions
A +A_=A+C, (3.12a) ‫نر‬ 0
‫و‬ ‫ه‬
])‫ه‬
+ )‫ر‬
‫ل‬
‫ه‬-22
9 ‫ر‬
‫ا‬
‫(ر‬‫قم‬
*
‫ا‬
and
)91.3(
‎‫ الهاي‬-4-١-©+ . (3.12b)
If ‎‫ عط‬second constituent ‎‫ له‬the superlattice is described
by ‎‫ عط‬dielectric constant ‎‫ روت‬as in Sec. II, then from ‎‫عط‬ The integral ‎‫ عه‬6 may be evaluated in closed form.
discussion given there we know that Since this integral is encountered frequently in the theory
of small-angle electron-energy loss, we shall quote the re-
A =FA_, (3.13) sult explicitly. Let 6; ‎‫ عط‬the angle ‎‫ له‬incidence of the
 1704 R. E. CAMLEY
electron beam measured relative to the normal to the sur-
face. Then
)| ‎‫هتقو‬ (3.20a)
and
vy =vgcost; , (3.20b)
PR I dQ) QfIm[R(Q,»)]
Tho'cos?0; Y E[(£2—1)*+4E%0s%0, 2
‫لا‬681) )25 - 1+ 5:2 ‫وم‬9‫!(ر‬/1)]2+ ,252059 15+ cos6; )(1+E2c0s%6;)‫‏‬
+£%in0;(3E%c0s0;
In the calculation of the loss cross section, the integral on
the magnitude of Q)| quite clearly must be performed nu-
merically. But our analytic evaluation of the angular in-
tegration in Eq. (3.19) reduces this to a simple task, out-
lined below, since the calculation involves integration over
only one variable.
1t should be noted that for R(Q)},®) to be nonzero, one
or more of the dielectric constants of the substrate must
have a nonzero imaginary part. In our study of the
dispersion relations in Sec. II, ‎‫ لله‬dielectric constants were
taken to be real. Thus, here the attenuation constant j, as
found from Egq. (2.31), has a nonzero real and imaginary
part, and care must ‎‫ عط‬taken to always choose 8 ‎‫ مد‬that
ReB>0. In the calculations reported in Sec. IV, the
aluminum film is modeled by the choice
2 d;=15 A and d,=20 A. Clearly visible is ‎‫ عط‬prominent
eqlw)=1— ‫مه‬
‫ )له‬, (32.3)
with y the conduction-electron relaxation rate.
Near-specular electron-energy-loss studies generally col-
lect not all the electrons scattered via the mechanism
described above, but rather collect those scattered within a
certain angular range about the specular direction. This
range is determined by the slit width of the spectrometer,
which generally subtends an angle of roughly 1°, as viewed
from the sample. In our calculations, we simulate this by
cutting off ‎‫ عط‬integral on Q) in Eq. (3.22) at the value
Qff'=k'"A6, with ‎‫ نال‬the wave vector ‎‫ له‬the incident
electron. In the scattering event, 70 is the momentum
transfer suffered by the electron projected onto a plane
parallel to the surface, so that this procedure assumes that
all electrons which suffer a momentum transfer (projected
onto a plane parallel to the surface) between 0 and Q f’ are
collected. One could envision a more realistic cutoff pro-
cedure, reflecting ‎‫ عط‬actual slit geometry, but this method
allows the integration on 6 ‎‫ م‬be performed analytically,
as described above. The results of the calculation are not
very sensitive ‎‫ عط ما‬details of ‎‫ عط‬cutoff procedure.
AND D. L. 111115 29
where v, is the speed of ‎‫ عط‬incoming electron. The result
is conveniently expressed in terms of the dimensionless
variable £ given by
£E=00Q) /0 . (321
The loss function then becomes,
—2 —i€ ‎‫ روم‬+) E4sin*0;]} ٠ 322
IV. NUMERICAL CALCULATION
OF THE ELECTRON-ENERGY-LOSS
CROSS SECTION—GENERAL DISCUSSION
We now turn to our numerical studies of the electron-
energy-loss cross section for scattering off a superlattice
structure.
‘We first consider the case where ‎‫ وى‬,1= since, as we see
from Sec. II, the excitation spectrum for this case is par-
ticularly simple. In the figures that follow, we choose the
incident-electron kinetic energy to be 200 eV, with a 45°
angle of incidence. The spectrometer slit widths are as-
sumed equal ‎‫ م‬,°1 as mentioned at the end of Sec. IV.
Material 4 is modeled through use of the dielectric func-
tion in Eq. (3.23), with ‎‫ رن‬51= eV and y=0.2 eV
In Fig. 7, we show the loss spectrum for two cases: The
first has ‎4‫ر‬02 A and d,=15 A, while the second has
surface-mode peak at 10.6 eV for the former case, while
only a ‎‫ عامط‬is present for the latter, near the
aluminum/vacuum surface-plasmon frequency of 10.6 eV.
The broad feature which rises dramatically with decreas-
ing energy loss has its origin in scattering off of the con-
tinuum of “bulk” excitations which, as one sees from Fig.
3, extend down to very low frequency when Q)d, is small.
This feature will be evident in data as a broadening of the
quasielastic peak, which varies as the thickness of the
aluminum film is changed. Such a broadening of the
quasielastic peak is evident in recent studies of thin films
of Ag deposited ‎‫ ذه‬GaAs;" here the whole feature is
shifted to much lower energies by virtue of the smaller
bulk-plasmon frequency ‎‫ له‬Ag and by softening of the
low-frequency collective modes of Ag produced by the
screening of the electric fields provided by the GaAs sub-
strate. A theory identical to that employed here provides
an excellent account ‎‫ عط له‬variation of ‎‫ عط‬broadening
with increasing Ag film thickness. Note, that as in the
earlier example,'® with increasing aluminum film thick-
ness, ‎‫ عط‬intensity of ‎‫ عط‬quasielastic background decreases
for films thicker than a few monolayers.
It is well known that if one scatters electrons off a
semi-infinite sample of aluminum, then the mechanism
29 COLLECTIVE EXCITATIONS 01 SEMI-INFINITE SUPERLATTICE STRUCTURES: 1705

T T an
8y=45°
Eq=200ev
[ 1ol dp=4/3d, €g=1.0 4
2 ‫ا‬ 2 ‫ا‬
=
2 2
5 5
‫ع‬ ‫وده‬ 040 ]
َّ
L o <
[ o
[= E 4= 4804
[Q @
« dp=96040
3 3 om0 ‎‫لد‬
23 ool 3 .
o a 4,=4804
42404
4,21203
0.001 L L
0.001 1 o 50 100 15.0
0.0 5.0 100 15.0 ENERGY LOSS (eV)
ENERGY LOSS (eV)
FIG. 8. Electron-energy-loss cross section for near-specular
FIG. 7. Electron-energy-loss cross section for near-specular scattering of 200-eV electrons from a semi-infinite superlattice
scattering of 200-eV electrons off of a semi-infinite superlattice of aluminum films with vacuum in between. In ‎‫ لله‬cases, the ra-
of very thin aluminum films interspersed with vacuum. Two tio d,/d, is fixed at ‎‫ عط‬value 5.
cases, as indicated, are considered.

considered here leads to a loss peak at the frequency of the constituent. This point is reinforced in Fig. 9, where we
surface mode of the aluminum/vacuum interface, 10.6 eV present calculations, again with eg=1, of the loss spec-
for our model. It is interesting to inquire how such a trum for several values of d;, but now with the ratio
spectrum evolves from that given in Fig. 7 if ‎‫ رك‬is in- d,/d, fixed at rather than +.
creased with the ratio d,/d; held fixed at a value greater In Fig. 10 we present calculations of the loss spectrum
than unity; ‎‫ عط‬point is that in ‎‫ عط‬limit ‎‫ رم جره‬we must for scattering from a semi-infinite stack of aluminum
have a loss peak at 10.6 eV as ‎‫ عط‬only feature in the spec- films, but now with each separated by a dielectric with a
trum, but the discussion of Sec. II shows we never have a dielectric constant €p=3. In each example, the low-
surface mode at this frequency for any finite value of d, if frequency loss peak has its origin in scattering off the sur-
42/41 is greater than unity (see Fig. 4). face excitation in the gap between the ‎‫ رت‬and w_ bulk ex-
We study this point in Fig. 8, where we show the
energy-loss cross section for several values ‎‫ رك له‬with the 1.00 - r r _
ratio 41/42 fixed and ‎‫ رك‬increasing to rather large values. r 6y=45° 1
For d, =120 A, we see a shoulder near 2.5 eV produced 7 F Eg=200eV 1
by scattering off ‎‫ عط‬w_ branch of excitations. As ‎‫ رك‬in- = | dy=4/3d, €g=1.0
creases, this shoulder evolves into a prominent peak, and
moves to progressively higher energy, but always below
5S oot ‎‫ل ووو‬
10.6 eV. Similarly, scattering from ‎‫ رن عط‬branch pro- o d,=4804
duces a loss peak above 10.6 eV which softens as ‎‫ رك‬in-
@
creases, always remaining above 10.6 eV. As ‎‫ رك‬increases, =
‎‫ عط‬two peaks coalesce, and the gap between them fills in. @
What is left in the limit d;—o0 with d,/d; fixed at a ©> 0.010) . B
value greater than unity is a single feature centered at 10.6 = di=9604 1
eV with a width controlled by the damping factor y in Eq. 2
(3.23). Note that as ‎‫ رك‬increases, the low-energy “tail” in [ 9604
the loss spectrum decreases dramatically, so that there will
be very little broadening of the quasielastic beam for 0.00١ L L
scattering off thick films. o 5.0 100 15.0
Upon comparing the dramatic differences between the ENERGY LOSS (eV)
energy-loss spectra in Figs. 7 and 8, one may appreciate FIG. 9. Electron-energy-loss cross section for near-specular
that even when the aluminum film is quite thick, the scattering of 200-¢V electrons from a semi-infinite superlattice
electron-energy-loss spectrum is influenced by the whole of aluminum films with vacuum in between. In ‎‫ لله‬cases, the ra-
structure and not just by the properties of its outermost tio dy/d) is fixed at the value 3.
 1706 R. E. CAMLEY AND D. L. 111115 29

T ‎‫وص‬ the Al/vacuum interface. The origin of this shift is the

‎‫مع‬ upward dispersion evident in Figs. 5 and 6. Again, one
0 ] appreciates from the figure that the loss spectrum reflects
UNITS)

d)= 9604 the nature of the structure below the outermost aluminum
4= 19204 film, even for rather thick films.
‘While we have confined our attention to the case of a
superlattice with a metallic and an insulating material as
P(w) (ARBITRARY

the two constituents of the semi-infinite superlattice in

T
o
3

our numerical examples, the theory quite clearly applies to

semiconducting superlattice structures as well. For this
reason, in Secs. II and III we have attempted ‎‫ ما‬present
the analysis in a sufficiently generalized form so that the
interested reader may apply the theory to any structure of
interest. We note, in fact, that electron-energy-loss spec-
o
Q
T

troscopy has been applied to the study of the surface exci-

0.0 5.0 10.0 15.0
ENERGY LOSS (eV)
FIG. 10. For e5=3, and two sets of values for ‎‫ رك‬and d,, we
show the energy-loss spectrum, with material 4 again chosen to
‎‫ عط‬aluminum.
citation branches. For the case d,=7d;, the surface
wave which exists only for a limited range of wave vectors
evidently scatters electrons with sufficient intensity to
produce a clear loss peak. The high-frequency feature in
‎‫ عط‬loss spectrum has its origin in scattering off of the
upper surfacc wave branch; the peak in each case is shift-
ed somewhat above the 10.6-¢V loss energy appropriate ‎‫ما‬
tations of doped GaAs,' so that it should also prove a
suitable probe of superlattices, although high-resolution
methods are now required since the losses of interest lie in
the range of a few tens of meV. Perhaps Raman spectro-
scopy will also prove a powerful probe of the modes dis-
cussed here in semiconducting systems. We currently
have semiconducting superlattices under study and the re-
sults will be reported elsewhere.
ACKNOWLEDGMENTS
This research has been supported by the U. 5. National
Acronautics and Space Administration through Contract
No. NAG3-250.

1See, for example, Z. Q. Zheng, C. M. Falco, J. B. Ketterson,
and I K. Schuller, Appl. Phys. Lett. 38, 424 (1981).
2A. L. Fetter, Ann. Phys. (N.Y.) 88, 1 (1974).
3R. E. Camley, T. 5. Rahman, and D. L. Mills, Phys. Rev. B 27,
261 (1983).
4P. Griinberg and R. Mika, Phys. Rev. B 27, 2955 (1983).
SM. Grimsditch, M. Khan, A. Kueny, and 1. K. Schuller, Phys.
Rev. Lett. 51,498 (1983).
6We understand the influence ‎‫ له‬retardation ‎‫ ذه‬the excitations
of superlattice structures is under investigation by G. Giuli-
ani, J. J. Quinn, and R. F. Wallis (unpublished).
7E. Evans and D. L. Mills, Phys. Rev. B 8, 4004 (1973).
8C. Powell, Phys. Rev. 175, 972 (1968).
9D. L. Mills, Surf. Sci. 48, 59 (1975).
101. Froitzheim, H. Ibach, and D. L. Mills, Phys. Rev. B 11,
1980 (1975).
1111. Tbach and D. L. Mills, Electron Energy Loss Spectroscopy
and Surface Vibrations (Academic, San Francisco, 1982),
Chap. 3.
12y, L. Schaich, Phys. Rev. B 24, 686 (1981).
1L, H. Dubois, G. P. Schwartz, R. E. Camley, and D. L. Mills
(unpublished).
14R. Matz and H. Luth, Phys. Rev. Lett. 46, 500 (1981).