I NF RARE D LATTI CE BAN DS IN AlSb

from the expression given by Born and Huang" in terms ACKNOVf LEDGMENTS
of the dielectric constants and the fundamental crystal The authors wish to acknowledge the Compound
frequency. This value is in agreement with the one Semiconductor Research Group of Battelle Memorial
reported earlier by Picus et ul. ,15 although the input Institute for the samples of A1Sb. We thank R. W.
data which they used for their calculation were sig- Keyes for his interest and stimulating discussions during
nificantly different than the ones reported in this paper. the course of the work; W. J. Doherty and R. P. Kelisky
"' M. Born and K. Huang, Dynamical Theory of Crystal Lattices who wrote the '7090 program for the reflectivity calcula-
"
(Oxford University Press, New York, 1954).
G. Picus, E. Burstein, B. W. Henvis, and M. Hass, J. Phys. tion; and S. P. Keller for his helpful comments on the
Chem. Solids 8, 282 (1959). manuscript.

P H YS ICAL REVIEW VOLUME 127, NUMBER 1 JULY 1, 1962

Theory of Photoelectric Emission from Semiconductors

EVAN Q. K~z
Bell Telephone Laboratories, MNrray Hill, Rem Jersey
(Received February 23, 1962)

The yield vs energy relation is determined for a number of possible photoelectric production and escape
mechanisms involving volume and surface states in semiconductors. Calculations are based on density-of-
states considerations and involve energy-band expansions to lowest nonvanishing order about the threshold
point. The "direct" and "indirect" processes involving volume states have yields proportional to E Eg —
and (E Er)', respectiv—ely. Both processes appear to have been identi6ed experimentally by Gobeli and
Allen. The linear yield also requires, in addition to production by "direct" optical excitation, that the
observed photoelectrons escape without scattering in the volume or at the surface. Energy and angle dis-
tribution functions of the emitted carriers are also determined.

I. INTRODUCTION AND CONCLUSIONS Gobeli and Allen4 in silicon and by Scheer' in CdTe and
CdS. This mechanism should produce the highest
HE early theoretical work on the photoelectric
effect generally made the free-electron approxi- quantum yields near threshold but it also has the highest
mation for the energy band structure. Fan' has given a threshold energy. The threshold energy will consist of
the energy to free the electron, the energy of the hole
general quantum mechanical treatment of the volume
production of photoelectrons. In his more detailed work
TABLE I. Dependence of photoelectric yield, 7, on photon
he also made the free-electron approximation. energy, E, near threshold, Ez, for a variety of production and
Huntington and Apker' have generalized Makinson's' scattering mechanisms. G~ is the Fermi level measured from the
work on the surface production of photoelectrons by vacuum.
taking the band structure into account where Makinson
assumed "free" electrons with a surface barrier. DI RECT OPTICA I
UN SCATTERED Y (E-ET)
EXCITATION ELASTICALLY
SCATTERED Y (E- ET)
The present paper deals primarily with the form of VOLUME
PROCESSES
the yield vs energy curve near threshold for a general INDIRECT OPTICAL
EXCITATION
t
UNSCATTEREDe1,
l ELASTICAU Y
Y (E- ET)5/2
L SCATTERED
band structure and for a variety of photoelectron pro-
duction and scattering mechanisms. These results should VOLUME STATES)
SURFACE AS
j~
ROUGH g SURFACF Y (E- ET)S/2
MOMENTUM
be useful in inferring the production mechanism from ABSORBER PERFECT SURFACE Y-(E-E )+2
the yield curve. The principal results are summarized THRESHOLD,
Y (E —ET)
in Table I. The discussion follows the ordering in the
ET FI
DIRECT OPTICAL ~

EXCITATION

table. SURFACE URFACE BAND

THRESHOLD
ET=t~FI Y~(E~E )3/2
PROCESSES STATES
In Sec. II the volume production of photoelectrons by INDIRECTOPTICAL T I F~ I
Y~(E- ET)
"direct" optical transitions is studied. In the absence of
EXCITATION
] ) ] Y-(E- ET)S/2
volume or surface scattering the yield should be linear DISTRIBUTED IN Y-(E- ET)2
—
with energy, E E&, above threshold Ed. There is good
SURFACE
PERFECTION
STATES
ENERGYe ET l~F I
LOCALIZED IN Y~(E- ET)
ENERGY BELOW CF
evidence that this mechanism has been observed by

4 G. W. Gobeli and F. G. Allen, following paper LPhys. Rev. 127,

rH. Y. Fan, Phys. Rev. 68, 43 (1945).
2H. B. Huntington and L. Apker, Phys. Rev. 89, 352 (1953); 141 (1962)g.
H. B. Huntington, ibid. 89, 357 (1953}. ~
J. J. Scheer and J. van Laar, Philips Research Repts. 16, 323
' R. E. B. Makinson, Phys. Rev. 75, 1908 t,'1949). (,1961).
132 EVAN O. KAN E

which is usually not at the valence band maximum, and III-V's whose maxima are degenerate, the Huntington-
the kinetic energy of the emitted electron tangential to —
Apker extra power of E E; should not be included.
the emitting surface which is required by conservation In Sec. VA we also consider the effect of the at-
of tangential momentum. Since the surface will usually tenuation of thelight intensity, Ao~'e ', which maynot
always be negligible, since n has a value of 2)(10' cm '
~

be normal to a symmetry direction, the threshold may

often occur at a point on the symmetry axis so that no for Si at 4.5 eV according to Philipp and Taft. The
tangential energy would be required. In fact, tangential effect is to relax the selection rule, dk„=0, on mo-
momentum will often be a strong factor in forcing the mentum normal to the surface for "direct" transitions.
threshold to occur on the symmetry axis normal to the In Sec. VB a number of processes involving sur-
surface. However, for the L111j direction in silicon face states are discussed briefly. Four cases involve the
studied by Gobeli and Allen, the upper valence band is perfect surface with band states. The cases depend on
known to be degenerate in the absence of spin-orbit whether momentum tangential to the surface is or is not
interaction. Normal to the L111$ direction, the de- conserved ("direct" or "indirect" processes) and whether
generate bands separate with terms linear in k. Unless the threshold occurs at or away from the Fermi level.
the relatively small spin-orbit interaction negates the The yield energy dependences may be listed:
importance of the linear terms, the direct threshold
should occur at three equivalent points off the L111j
1. Direct; threshold EasA &z ~; I'-E —
~
Eas.
axis and tangential kinetic energy should be required at 2. Direct; threshold Bp~; I" (E — 8~~)'.
~ ~

threshold. 3. Indirect; threshold E,sN B~~; I' (E Es)'. —

In Sec. III the effect of surface or volume scattering ~

on the yield due to direct transitions in the volume is 4. Indirect; threshold hp~; I' (E 8p~)'.
~ ~

considered. The yield is found to rise quadratically, Two cases involve states due to imperfections. If the
(E Ed, )', above— the threshold, Eq, . Eq, will often be states are distributed in energy with constant density
lower than Ed because there is no transverse momentum at the Fermi level, the yield is proportional to
conservation requirement. (E Bz~)'. If they are lumped at a discrete value,
In Sec. IV indirect transitions in the volume are con- —El„ the yield goes as E—El, .
~

—
sidered. The yield is found to increase as (E E;) above
&

the indirect threshold, E;. In this process the hole is left A. Assumptions and Conventions
at the maximum in the valence band and there is no
tangential energy of the emitted electron at threshold.
The yield vs energy dependences quoted are based
In Sec. V the yield vs energy is calculated for a entirely on "density of state" considerations assuming
number of processes which may occur at the surface.
that matrix elements do not vary rapidly near threshold.
These processes are divided into two categories. The The only exception is Sec. V, where we discuss the
6rst category, treated in Sec. V A, consists of processes Huntington-Apker node. Symmetry "forbidden" matrix
elements may be important in practice but we do not
in which the hole is left in a "volume" state. In Sec. V 8
discuss them further in this paper. The general effect
we treat processes where the hole is left in a surface
would be to add an extra power of E to the relevant
state.
In category A, an imperfect surface may be the mo- yield vs energy relationships. When quantum yields are
written down, it is assumed that optical absorption
mentum absorber for an "indirect" process whose yield
would vary as (L& E,)"* just as fo—
varies slowly near threshold. In general, if a quantity is
r the volume indirect
not zero at threshold it is assumed to be slowly varying.
process. A perfect surface can only absorb "normal"
momentum. For processes of type "A" with a perfect Strict energy conservation is always assumed. In fact,
surface, the yield is proportional to (E —
"lifetime broadening" effects due to strong scattering
E;) assuming
&

that the valence maximum lies along the surface normal. mechanisms, particularly where "pair production" is
This need not be true in general, but is true for Si, Ge, possible, will act to smear out predicted thresholds.
and all III-V's as far as is k.nown, since their maxima Energy loss is treated on an "all or nothing" basis, see
lie at k=0. Sec. III B. This assumption is certainly not always
Huntington and Apker' have demonstrated that for a adequate.
valence band possessing a nondegenerate maximum at Taylor expansions to lowest nonvanishing order for
functions of the energy bands are always made about
k=0, the k=0 wave function possesses a node at the
surface. All transition probabilities at the surface are the threshold point. This approximation may go bad
then proportional to E— E, which would introduce an quickly when bands are close together in energy at the
extra power of E— E; into all yield-energy dependences relevant k point. The simple results, in general, apply
quoted above for surface mechanisms. We find that the only for energies small compared to appropriate band
Huntington-Apker node is not present for degenerate separations.
bands or for bands, where k„, — k„QO are the k- %e assume a complete set of states which are HIoch
coordinate projections normal to the emitting surface of states inside the crystal and join smoothly through the
equivalent maxima. In particular, for Si, Ge, and the H. R. Philipp and E. A. Taft, Phys. Rev. 120, 37 (1960).
 PHOTOEI. ECTIiI C EM I SS ION I:ROM SEM ICON 0UCTOIiS
image barrier to free electron states of equal energy and vacuum is on the right of the solid. We neglect the re-
equal tangential k vector. These states are indexed by Rection by the image field barrier which MacColl has
their Bloch k vector. All "transitions" and "scatterings" shown to be small. '
are between such states, either in the volume or at the The quantum yield, electrons per photon, denoted by
surface. I", is then given by
Transitions are considered between a valence band,
indexed 2, and a conduction band indexed 1. Sums over —Bg(k) —E)dk
different band pairs may be important, a, s in. Eq. (1), P
1,
2
~
A. py, (k) ~'b(hy(k)
but will be suppressed, when possible, to avoid pro-
liferation of symbols. Similarly, sums over phonon
modes will not be explicitly indicated. p i
A pg2(k) i'8(hg(k) —b2(k) —E)dk, (3)
The symbol E refers to photon energies while h is used z 1,2
to denote electronic energy levels with the vacuum where the prime denotes integration over those values
chosen as the zero of energy.
of k satisfying condition (2). The factor by which the
All calculations are made for a filled valence band,
yield is reduced due to scattering is treated in Sec. III F.
empty conduction band at T= 0 with no "band bend- Equation (1) assumes "direct" transitions with the
ing" at the surface.
customary neglect of the k vector of the light. We treat
"indirect" transitions in Sec. IV.
II. DIRECT TRANSITIONS IN THE VOLUME
The rate of photon absorption in a semiconductor in A. Energy Deyendence of Yield
the one-electron approximation is given by the "golden
rule" We are interested in the energy dependence near

ZV12 =— 2m e2 2U
threshold of the yield, V, in Eq. (3). At threshold, Ez,
we assume that the range of integration in collapses
to a single point, or set of points equivalent by sym-
j'
5 m'c' (2m)'
metry, kq.
The energy conservation relation,
A 1»(k) l'~(h~(k) —h~(k) —E)&k, (1)
.z.
I
E= 8, (k) —82(k),
~total = defines an "optical energy" surface. The critical escape
1,2 condition,
h, (k) =I'b'kP/2m,
A is the vector potential of the light, V is the volume. A
factor of 2 for spin is included, the integration goes over defines an "escape" surface. At threshold these two
one Brillouin zone (B.Z. ) and transitions are considered surfaces are tangent at the point kq for an optical
between pairs of bands with energy B,(k). The index 1 energy Ed. Let a be the common normal at kz directed
runs over all empty bands, 2 runs over all filled bands. toward increasing optical energy. These surfaces are
8 is the photon energy. depicted in Fig. 1. Let a, b, c be an orthonormal coordi-
If scattering and energy loss in the volume and at the nate system in k space with origin kq. We may make the
surface are neglected, the condition for escape may be following expansions which will be valid for small
written k„kg, k. :
& h)/Bk. )0, (2) hg 82 Etg+ VJC($] 8g) kk~+ quad( kbk, ), (5)
g
8, (k) = k'(kP+ k„")/2m.
~ ~

&2k)2
The energy zero is taken to be the vacuum. The k vector h~ — —ngk. +quad'(kb, k.). (6)
tangential to the surface, k~, is conserved but the normal 2m
k vector, k„, k„' in solid and vacuum, respectively, is not.
The positive group velocity condition assumes the The absence of terms linear in kb, k. in Eqs. (5) and (6)
is implied by the quadratic functions quad, quad . Posi-
tive o, 1 corresponds to a "lower" threshold which is the
customary situation with the yield increasing for ener-
gies above Eg. For negative o.1 we have an "upper"
threshold with yield increasing below Ji &. A different
type of "upper" threshold would be caused by a maxi-
mum in h~ —h2. Since V~(B~ —8~)=0 in this case, a
different expansion than that of Eqs. (5) and (6) is
FIG. i. The intersection
of the "optical energy" surface with the required.
"escape" surface. As E
is decreased the circle of intersection
shrinks to the point kq. a is normal to the escape surface at hg. 7 I . A. MacColl, Bell System Tech. J. 30, 588 (1951).
 E VAN 0. k AN:t'
If we substitute Eq. (5) in the integral in Eq. (3) we In this case k„', kb, k, are the k-vector components of the
can integrate over the variable k, . This integration re- emitted electrons. They must lie on the ellipsoidal yield
moves the 8 function and installs the energy conserva- surface given by Eq. (7). The surface character of the
tion relation; E= B~ —h2. If ee eliminate k between yield clearly means that in a given direction the emitted
Rqs. (5) and (6) using Eq. (2)& energy conservation may electron has a unique energy. This is to be contrasted
be wtitten with the case of scattering or "indirect" transitions dis-
8-Ee= p'k "+pbkb'+p, k, ', cussed later where the yield relation has a volume
character and a range of energies will correspond to a
Ss
( (7) single direction. This fact may prove useful in differ-
I
~~(@ —& ) I ~' entiating between different processes.
E2mn,
When kd is not normal to the surface, k~ and k, refer
Pb and P,. must have the same sign, which is positive or only to the internal electron. Let 8k= (O, kb, k, ) IW. e may
negative for a "lower" or "upper" threshold, respec- neglect k compared to k~, k, in the region where
tively. The orientation of axes 0 and t," has been chosen Eqs. (5) and (6) are valid expansions. ]
Then k„',
to simplify the quadratic form in kb, k, in Eq. (7). (ke+5k) are coordinates of the emitted electron, where
~

The integral in Eq. (3) is then over those values of the subscript, t, refers to the projection on the plane of
kb, k. which satisfy Eq. (7). The "differential yield" is the surface. The electrons are nearly tangential to the
proportional to the area dk~dk, . When the surface is surface in the direction kg~.
normal to a symmetry direction, the threshold point kd The integration over kb, k, in Eq. (3) is just the total
and 7'~(8, —hs) will often also be normal to the surface. area for which Eq. (7) can be satisfied. We find'

F'= y(E —Ee),

~gIA p»(4) I'

I
~b@r —&s I "(pbp. )-'* I
A p„(k) I'8(h, (k) —Ss(k) —E)dk
~ Z0

g is the number of points, kz, equivalent by symmetry. point, ke, and the gradient Vj, (8~ —Ss) will often occur
We have assumed that p&s(k) is slowly varying in the along this direction. This is because energy terms linear
vicinity of ke. If the total absorption is also slowly in k normal to the axis are required to vanish by sym-
varying, p will be nearly constant and the yield will rise metry. As noted earlier, k„', kb, and k, in Eq. (3) are
linearly with energy. Linear dependences of this sort then the k-vector components of the emitted electron.
have been observed by Gobeli and Allen4 and Scheer. ' The band parameters P', Pb, P, are probably most easily
determined from the energy of the emitted electron as a
function of direction, as given by Eq. (7). These
B. Nondegenerate Symmetric Bands
parameters also enter into the differential yield, dF,
When the emitting surface is normal to a symmetry which may be described with the following choice of
direction of twofold or higher symmetry, the threshold polar coordinates:

k„'= k cos8,
k~= k sine cosy, (10)

k. =k sin8 sing,
—sin8d8d y,
dQ=
(PbP. )'ed' k'
dY=

k2=
rr(cos8 cos'pI 1+ (pb/p')
(E Ee)— tans8$+cos8 sin'yL1+ (p, /p') tan'e))

{p'. coss8+pb sin'8 cossy+p, sin'8 sinsy)

(12)

dQ is the differential element of solid angle delned by electron is 5'k'/2m, which is a unique function of direc-
Eq. (10). 8 is measured from the surface normal. 7 is tion for a given photon energy, B.
defined in Eq. (8). Equation (12) is obtained by substi- If the symmetry axis has only twofold rotation sym-
tuting Eq. (9) in Eq. (7). The energy of the emitted s Equation (44) gives the reduction in yield due to scattering.
 PHOTOELECTRI C EM ISSION FROM SEM I CONDUCTORS

metry, energy is quadratic in k normal to the axes but In the [111) direction a new feature appears. The
the principal axes are not specified. The axis will in quadratic terms associated with A under the square root
general be different for conduction and valence bands which were riot present for the [111)direction will have
and for the relevant axes in Eq. (7). For "two-mirror the effect of displacing the threshold point away from
plane" symmetry the axes are specified. In this case we the [111)axis if 6 is small. The emitted electrons will
may write then have a hnite transverse kinetic energy at threshold.
6 = nika+ pibk 2'+Pi.k', If 6 is small and if the constant C in Eq. (16) were
negligible (there is no reason it should be), the energy
h2 n2ka+P2bkb +P2ckc Ed bands would be cylindrically symmetric and the thresh-
is normal to the emitting surface; energy zero is the old would occur on a circle rather than at three equiva-
vacuum. The quantities Pb, P„P' in Eq. (7) are easily lent points. In this case, the yield goes as (E Ez)' —and
found to be then abruptly changes to linear, E— E&, when the
(ni n2) (k /22ib) +n2 plb nlP2b emitted electrons are finally able to appear with zero
transverse energy. To illustrate these points we con-
sider a less general case, namely,
(n, n, ) (k—
'/2222)+nppi. n, p2. — (14) hi(k) =nik +pi(kb2+k. 2),
hi —82 ——
nk, +p(kb2+k, 2)&i [A(kb2+k. 2)]'+Ep, (19)
p'= ~
ni —n2 (&'/2222ni)
~
E= Bg —82. (20)
The number of equvalent threshold points, kz, will be ~ith the use of Eq. (19) the integral in Eq. (3) is easily
unity in the case where kd is normal to the emitting performed. The integral of dk over the 8 function gives
surface. a factor 1/n and requires energy conservation, [Eq.
(20)). The integral is then proportional to the area in
C. Degenerate Symmetric Bands kb, k, space which satisfies Eqs. (2), (18), (19), and (20).
Eliminating k, between these equations, we have
In cubic materials the [111)and [100) directions are
symmetry directions of special interest. They both (kb2+k, 2) &
possess doubly degenerate bands which split under spin-
orbit interaction. Group theoretical considerations show aiA&+2 (A+b[E (n/ni)(k2/2—222)k ")}'*
that the forms of E vs k in the vicinity of the symmetry (21)
axes are (b/2)
[100) direction f k n
= 4 p —~pi—
b—
E(1)= k, +P(k.2yk„2) 2222 n,
a[8i(ka'+k ')'+82k, 'k '+ (6/2)']'* (15)
The two & signs are independent, the 6rst refers to
[111]direction Eq. (19). The coeKcient bn/ni must be positive in order
E(k) =nk +P(kb'+k ')~[A(kb'+k ') that unbounded values of k„" and (kb2+k. 2)' do not
' — occur. We also assume b positive so that we get the
+8(kb'+k, ')'+C(k, 3k, kb')+ (6/2)']& (16) normal case of a "lower" threshold. The threshold is
k. = k. (1/VS, 1/v3, 1/V3), easily seen to be given by
kb ——kb(1/v2, —1/W2, 0), (17) Eg —(A/b)+ Ep, — (23)
k, = k, (1/+6, 1/Q6, —2/Q6).
(kb2+k 2)g'*= 2A&/b,
Equations (17) define the vectors k„kb, k, in terms of
k2
their x, y, s components. 6 is the spin-orbit splitting. A —pi ~4A
linear k dependence along the symmetry axis is assumed.
k.g= ~
(25)
2222 1nib2
In what follows, we assume the surface is normal to
the symmetry direction in question. If the computation The solution clearly involves transitions corresponding
leading to Eq. (8) is carried out using Eq. (15) and to the ( —) case in Eq. (19). For E&Ep, only the ( —)
ignoring 6, the same linear dependence of yield on case is involved and the permissible area is given by
photon energy is found. This is due to the quadratic
character of the energy dependence on transverse mo- Farea= 1'62r[Ab(E —Eq)]'*/b2 E E&p&0 (26)— .
mentum. The angular dependence of the square root in For E&0, both & bands contribute and the total per-
Eq. (15) enters into the constant y. Two regions of missible area summed over both and +
transitions is—
different linear slope might be found for E— E~&&5 and
for E— Eq))A in favorable circumstances. F —
area= 82r[A+b(E Ed))/b', E Ep) 0. (27)—
 136 E VAN O. KAN E

Hence, the yield abruptly switches from square root to A. Internal Kinetic Energy
linear (but with continuous value and slope) as the
permissible area touches the symmetry axis. „
The threshoM, Ed, is defined as the minimum photon
energy for which the conduction electron is energetically
As remarked earlier, this solution may not have much
significance since there is no good reason for ignoring
able to escape. We make our energy-band expansions
the cubic terms in Eq. (16). With the inclusion of cubic about the point, kd„defined by
terms the threshold would reduce to a set of three 8, (kg, ) = +)i~,
equivalent points off the 111j axis which would give a
Eg, = 8~(kg, ) —82(kg, ).
I

linear yield as shown earlier. However, one would expect

a small range for this linear dependence because of the
Energy zero is still the vacuum and 8, (k) is the con-
proximity of the other band. —is for phonon. emission and ab-
duction band. +,
III. DIRECT VOLUME TRANSITIONS sorption, respectively; Sco=0 for elastic scattering. We
WITH SCATTERING expand E about kz, in the same way as in Eq. (5):
In the event that the threshold energy for direct 8j
transitions, Eq, corresponds to a k vector, kq, which is
89 Edg+ V+8/ I
8$ I Qg k~+quad(kp)k, ). (30)
not normal to the emitting surface, there will be a lower We also expand 8~(k),
energy threshold, Ed„resulting from surface or volume
scattering which allows the electron to escape with very 8, (k) = n, k.+ quad'(k b, k, ) a5a&,
little momentum tangential to the surface. The yield —
e = 8, (k)w5(o.
(31)
near threshold will be much lower in this case and will
rise like (E Ez, )' a—s compared to (E Ed) since o— nly e is the energy of the emitted electrons. The absence of
a small fraction of the scattered electrons will lie within linear terms in k&, k, is a necessary consequence of the
the solid angle permitting emission. assumption that kq, is a threshold. The procedure is
The total yield, F&„in the scattering case depends on entirely analogous to that used for Eqs. (5) and (6).
the product of four factors which we shall consider in Substituting Eq. (30) in the integral in Eq. (3), we
turn. integrate over k„ to remove the 6 function. Then k, can
I'"=p p f'f (E E')'— (28) be eliminated between Eqs. (30) and (31), giving
f, (E Eq, ) is t—
hat fraction of the internally excited E Eg, $V g—
(8g — e+Pgkg'+P, k,'
82) g„,/n, $— (32).
electrons which have sufficient kinetic energy to escape.
p, is the probability that the electron is scattered with- Axes have been chosen for b and c to simplify the quad-
out substantial loss of energy. f, (E Ez, ) is the fr—
action ratic form in Eq. (32). pq, p„and Vq(8~ —82)/n~ must
of the scattered electrons which lie within the escape be positive for a "normal" threshold. We use Eq. (32)
cone determined by the external tangential kinetic to define an "internal" differential yield, dI';, for elec-
energy. p, is the probability that the scattered electron trons of energy e in an energy range de. Using Eq. (3),
can escape without substantial loss of energy. we calculate

dI', =f, (l |7.(8i —8~) I .. /~~)de (33)

~gI & p»(4 ) I'(&~/I ~~(8j.—8~) ~") I

(p~p. ): IA I-(k) I'~(E —8 (k)+8 (k))dk

0( C( (Qg/ I
7 Q(8$ '82) I ~„)(E—E~.). (35)

The total fraction of electrons having sufficient kinetic principal vectors of the solid. For low-index planes such
energy to escape is f, (E Ez, ) as contai—
ned in Eq. (28). as L111$ and L100j in the face-centered cubic lattices,
g is the number of distinct points kz, which are equiva- the principal vectors of the surface are just the principal
lent by symmetry. vectors of the volume projected on the surface plane;
hence, no new scattering is obtained. Since the usual
B. Probability of Scattering surfaces are low index, it seems unlikely that "surface
%e should first mention and then dismiss the possi- umklapp" will be important.
bility of "surface umklapp" scattering. By "surface Note added irl, proof In the case of surface super- ".

omklapp" we mean scattering by a principal vector of lattice" structures such as those found in germanium,
the two-dimensional surface lattice. For high-index "surface umklapp" could be more important.
surfaces these vectors may be much smaller than the We divide scattering mechanisms into two types:
 PHOTOELECTRIC EM ISS ION FROM SEM ICON DUCTORS 137

(a) Phonons, impurities, or imperfections in the volume band structure information contained in the unseat™
or the surface. (b) Electron-electron collisions resulting tered case has been washed out by scattering.
in pair production. Within one band gap of threshold,
category b necessarily leads to loss of the excited elec- D. "Escape" Probability
tron. m and mb represent the total scattering rate of
The final factor to be considered is the probability
mechanisms u and b, respectively. We assume that
that the electron travels from the point of origin to the
energy losses can be ignored for processes of type u surface without su8ering an energy losing collision. This
(obviously not always true for phonons), and that type
we call the escape probability, p, .
b scattering always removes the electron from considera-
The escape probability may be written
tion. Then the probability, p„ that one or more scat-
terings of type u precede a scattering of type b may —g— t e
WQ
(40)
easily be shown to be
where t is the time required to reach the surface. If the
p, = w. /(w. +w t, ). (36) electron moves without scattering, this time is

C. "Escape" Cone (41)

ah, /ak„i'
We now calculate the fraction of scattered electrons i

which are able to escape. Their tangential momentum, where x is the distance from the surface of the point of
k&', must satisfy the energy conservation relation origin. If many quasi-elastic scatterings occur, the
e=5'(k '2+k "2)/2m (37) electron will diffuse to the surface and the escape
probability will be lowered.
where e is the total kinetic energy of the emitted elec- The absorption probability per unit length is pro-
trons. A detailed treatment of the scattering seems —
portional to exp( x/l ). If we crudely ignore the
uncalled for; hence, we assume that the scattered par- electrons motion prior to the first scattering and average
ticles are distributed uniformly over the energy shell over x for 3 given by Eq. (41), we find
8(Bi(k)%5co—e). This is equivalent to assuming a mo-
mentum independent scattering matrix element. For the p =~/(l+~ ) (42)
case m, ))mb, so that many quasi-elastic scatterings I, = (88,/8 k)1/Awb (43)
occur, the assumption should be quite accurate.
The fractional "volume" of the energy shell satisfying
Eq. (37) is easily seen to be 2f, L V'~(8, —82) k„, ~/ni]e,
where
~
E. General
The yield in the scattering cases rises like (Z Ez,
as compared to the linear rise (Z Ed, ) of the —
)'—
mPn, /( 7g(8, —h, ) g„[]
'.

~ Z0
~(hi(&) )d&
(38) mentum-conserving
f.
comparable within a few volts of threshold as (8 Eg,
mo-
case. The two processes will become

approaches unity. Of course, the approximations leading

to the linear and quadratic energy dependences will not
)—
k„ is the point on the energy shell 8(h&) normal to the generally be applicable for more than a volt above
emitting surface and assumed to be normal to the energy
threshold.
shell also. f, is assumed not to vary rapidly with energy
near threshold; The total yield factor introduced by the
F. Escape for Unscattered Electrons
escape momentum condition, Eq. (37), is In computing the yield for direct transitions without
scattering we should include an escape factor, 1 —p, ', so
f, (g +~,) that Eq. (8) becomes
where an average has been taken over all e for which
n= (1—p. )v(& —&.)
escape is energetically possible, i.e., values of e satisfying
'
Eq. (35). p, is the probability of any type of scattering. Re-
The differential yield is easily computed to be placing w& by w +wb in Eq. (43) and using Eq. (42), we
find
d V~cosNQ ed&, (39) p. = t. /(f +~ )
where 8 is measured from the normal and c is the emitted lhi) 1
electron's kinetic energy. dQ is given by Eq. (10). For a
given direction, a range of energies from a=0 to Bk„)fi(w. +wg)
—
&= {&i/ Vk( hi 82) g„,}(8 Eg, ) may be found in con-—
~ ~

IV. INDIRECT TRANSITIONS IN THE VOLUME

trast to the unscattered case where a single energy
occurred. The fact that e,„/E — Ed, still attests to the We have so far considered processes where the ex-
initial production by direct transitions, but most of the cited electron is produced by a direct optical transition.
 138 EVAN O. KANE
The threshold energy then contains the energy necessary s denotes a possible angular dependence of effective
to excite the electron above the vacuum zero plus the mass. The "indirect" threshold energy, E; is then given
energy of the hole. Since the threshold will seldom by
correspond to a hole located at the valence band maxi- E, = —e2&ACO
mum, a lower threshold is generally possible for processes
where the initial excitation is "indirect. " with + for phonon emission, —
for absorption. The
The rate of "indirect" excitations is given by the density of states may then be written
second-order "golden rule"

wg3 —
—2z p Q hg2h23
8(Eg —E,). (46)
V2m(m2*) & V'
g] —g2
3 2
C=
15~9&'(a h, /ak„)
1, 2, 3 index the initial, intermediate, and final states, m2* is a density of states mass which has the well-known
respectively. The initial state is the unexcited crystal value
plus a photon; the 6nal state is the electron at k~, the m, *'*=g (m. mmmm, ) '*
(54)
hole at k~, and no photon. The intermediate state is
formed by optical excitation producing a virtual elec- for g equivalent ellipsoids. The value of m2* in Eq. (51)
tron-hole pair both at either k~ or k, (ignoring the k is chosen, so that
vector of the light). A phonon or imperfection then
scatters either the electron or the hole to the final s(0, y)dQ/4n. = 1. (55)
condgur ation.
Assuming that matrix elements and energy denomi- Equation (55) is applicable for degenerate bands. The
nators are approximately constant for all possible transi- density of states for degenerate bands should also be
tions near threshold, we write Eq. (46) as summed over bands, although this has not been ex-
2m. 2V' hg2h23
plicitly indicated. The yield, I', will be proportional to
wi3= the density of states as given in Eq. (53).
t't (2x)6 2 Ey —E2 The differential yield may be written
drl~cosg dQ e(E —E,—e):de, 0(e&E—E,. (56)
b(h~(k~) —h2(k2) —E)dkgdk2. (47)
e is the energy of the emitted electron. The yield does
not depend on any band structure parameters except in
The symbols are the same as in Eq. (1). The quantum the factor of proportionality.
yield is given by the ratio of w» from Eq. (47) to w of
Eq. (1). V. TRANSITIONS AT THE SURFACE
The "escape" condition is We distinguish two main categories of optical transi-
= (ft'(kP+k„")/2m) tions occurring near the surface. The erst category is
hg(k) &h(o. (48)
that where the hole left behind is in a "volume" state
l't~ is the momentum tangential to the surface,
—
is for + and the second category where the hole is in a surface
state. In both cases we assume the electron is in a
phonon emission, for absorption, AM=0 for elastic
processes. Energy zero is the vacuum. We eliminate the "volume" type state characterized by asymptotic mo-
5 function in Eq. (47) by integrating over k„, the com- menta k„, k, in. the solid or k„', k, in the vacuum. Near
ponent of k~ normal to the surface. This installs the the surface the normal momentum, k„, is not constant.
energy conservation relation which we use, together For convenience we use the volume state density ap-
with Eq. (48), to integrate over k„ the tangential propriate to the k„degree of freedom. Alternatively, one
component of k~. could use the apparently different vacuum state density
appropriate to k„' but the volume amplitudes would
2V2 have to be renormalized. We take the state density for
t '(~) = & (h, (k, ) —h, (k2) —&)dk, dk2, (49) k„as a constant near threshold and always integrate out
(2m. )' this variable to remove the 5 function on energy. The
remaining degrees of freedom are then integrated over
nSV2 82 2 EWACO dk2 subject to strict energy conservation.

t.(&) = 16n't't'(8 hg/Bk )

(50) A. Transitions from "Volume" States
1. DiJlse Surface Transitions
In the neighborhood of an extremum, k, we write Equation (53) should apply for any form of indirect
h2(k2) = e2+ft'(tt J't 2m2*
)'s(0, y)/— (51) scattering mechanism including diffuse scattering at sur-
 P HOTOELECTRI C E M I SS ION F ROM SEM I CON DU CTORS 139

face imperfections, provided the final state involves a single function which can only be done by making the
hole free in the volume. combined amplitude zero.
In the case of a degenerate maximum, k and k' may
2. Specular Surface Tralsitiorts lie in different bands; for a group of equivalent extrema
k and k' may lie in different minima (if the minima have
Transitions made at a perfect surface must conserve
the same transverse momentum). In either of the above
momentum tangential to the surface (ignoring "surface"
situations P~ and P~ do not become identical as k tends
umklapp). The energy threshold will be the same as for to the extremum so there appears to be no reason to
indirect transitions for those cases where the valence
expect a node at the surface in these cases. Si, Ge, and
band maximum lies along the k vector normal to the
all known III-V's come in this category since their
surface. In this case, we write
valence extrema are degenerate at k=0.
Bg(k) =h'(kP+k„'")/2m, As Huntington and Apker have pointed out, in those
= —E,+ h'(kp+ k„') /2m~*.
materials where a node is required, the transition proba-
bility vanishes proportional to E—
82 (k) (57)
E; at the threshold
The yield integral is for processes where the hole is left at the valence
maximum. The "diffuse" and "specular" yields calcu-
lated in Secs. I and II would then vary as (E E,)'t' an— d
dk, dk„, (58) (E E,) '*, respe—ctively.

subject to the energy conservation relation 4. Effect of Light Attertuatiort

E E, =
2
kP ——*)
—
m m,
~+
m m, *
(59)
In all calculations so far we have ignored the k vector
of the light. Even if this were not done, no qualitative
change would result. When we consider the effect of the
attenuation of the light, however, the situation is
The electron momentum k„' has been integrated over, qualitatively different because of the "fuzzing out" of
as described above, with the constant state density momentum conservation.
factor being absorbed in the proportionality sign. In We assume the light attenuates according to
Eq. (59), m2* has its (negative) algebraic sign. The
emitted electron has its maximum kinetic energy when Ape " f', (61)
k„=0. The parameter no~* can be determined from Eq.
where x is normal to the surface. The matrix elements
(59) by measuring this maximum energy at any angle
not normal to the surface. The total and differential for optical transitions are then easily seen to be pro-
yields are easily found by integrating over dk„dk„ in portional to
Eq. (58) using Eq. (59) and Eq. (10)

dF
I ~ (E—E~)~
~ e cosOdQde/(E
= eL1 —(m/m2*) sin'0),
e—
E,—e) ~— (60)
instead of being proportional to a 5 function of the total
momentum change.
If Eq. (62) could be trusted a long way in the
"
"wings, the calculation would be quite analogous to
&max= + +i ~
the case of "specular" surface transitions as treated in
Sec. II. This conclusion is almost certainly wrong.
e is the kinetic energy of the emitted electron. Equation (61) is of classical origin. It may be derived
semiclassically by considering a wave packet moving
3. Huutimgtort Apker Xode- with the group velocity and decaying with a Weisskopf-
Huntington and Apker' have demonstrated that for a Wigner factor t.'"' where m is the rate of making ab-
nondegenerate valence band with maximum at k = 0 the sorptive transitions. This formulation suggests the
electrons close to k=0 will have a node at the surface. "energy broadening" approach which yields the energy
nonconservation factor
They show this by calculating the ratio of incident and
refiected amplitudes using matching conditions at the 1/4 (AE/h)'+ w2. (63)
surface. The incident BLoch wave is u~(r)e'~' and the
reflected wave is au~ (r)e'~", where k„'= — k„, k, '=k, Equations (63) and (62) are very similar. Equation (63)
which results from the requirements of transverse mo- is known from more accurate considerations to give fa, r
mentum and energy conservation. They find that they "
too much amplitude in the "wings, hence it appears
are unable to match at the surface (except for an that Eq (62) shou. ld likewise be mistrusted. The true
unusual special case) unless a — + — 1 as k — + 0, which effect is no doubt to give a slight fuzzing of the "direct"
gives a node at the surface. The reason this occurs is that yield characteristic. One might hope to study the
u~(r)e'~'~u~ (r)e'"" as k —+0 so that one is at- "fuzzing" effect with the use of expansions similar to
tempting to satisfy two matching conditions with a Eq. (13).The results are seen to diverge which could be
 E VAN 0. RAN E

corrected by more accurate expressions for B(k) but the condition that kd( is a threshold. This then requires
more likely necessitates the use of an improvement on k, dr —0. Linear terms in o.2(kb kbdr) are allowed since
Eq. (62). this quantity has only negative values for states which
are initially occupied.
B. Transitions from Surface States The yield integral is
We now consider rather briefly a variety of processes
in which the hole is left behind in a surface state. The
I' dkgdk, (69)
first four processes assume a perfect surface so that the
surface states possess well defined bands and tangential subject to energy conservation
momenta, k q, k, . The normal momentum of the excited
electron has been integrated out, leaving strict energy Ak
conservation.
2m
+
AkgdP

m
-")
&2
1. Direct Transitions; Threshold&
X (kb —kgb —p )k.',
hp
I I
)+ (70)
We assume for this case that the threshold point lies 2'
below the Fermi level and that tangential momentum which integrates to
is conserved. The appropriate energy band expansions
are
P'~(E I
br I)k (7l)

e) = 52(k 2+kb2+k 2)/2m 3. Indirect Transitions; Threshold& I

Br I

52k 2
In this case the surface band maximum, — E,q,
E el e2 +pb(kb kbds) located at k~, k, must lie below the Fermi level. The
2m
appropriate expansions are
+p. (k. k.ds)'—
+Eds
t' 5
e& is the energy of the electron in vacuum, c2 is the by= (k '+kb2+k. '),
I

energy of the hole. Equation (65) is energy conserva- &2m (72)

tion. Threshold occurs at kds, Eds. The absence of linear
terms in k b — —
k bds, k. k, ds is arequirement for irdsto be e2
— E,s+pb(k—b' kb )'+p. —
(k,
'
k, )'—
a threshold. The yield F, is proportional to
The yield integral is

Y dk6dk, dkb'dk, ',

(73)
+dS (66) y'~ (E—E,e) ~

This case is quite analogous to the case of direct volume

transitions. One difference is that the maximum normal 4. Indirect Transitions; Threshold= hrI I

energy, (h'k„'/2m) is equal to E Eds rather than- In this case, the energy band expansions are very
diRering from it by a constant factor. similar to those of Sec. II.

2. Direct Transitions; Threshold= BrI k2

I

e) — (k„'+kb'+k, b), (74)

In this case the threshold point lies at the Fermi level, 21S
We use the expansion

I
hr I+a2(kb ,',
kbdr)+p2k— (67)
I
~r I
+c(2(kb kbdr)+p2(ke kcdr)' (75).
The yield integral is
ph'
(k„'+k;+k. ) V dkt, dk, dkg'dk, ',
&2m
(76)
fh P'~ (E— h r ) 4
(k.'+2kbdr(kb —kbd F)+k.'). (68) I I

&2m
5. Distribbtted Imperfection States
We have chosen the axis k~ normal to the constant
energy contour of the hole. Linear terms in k, — k, dI; are We consider here the case where a variety of imper-
absent for e2 by this construction. Linear terms in fections gives rise to a distribution of localized states
(k, — k. dr) must be absent from e) —e2, hence from e), by with a constant density per unit energy in the vicinity
 PHOTOELECTRIC EMISSION FROM SEMICONDUCTORS
of the Fermi level. We write 6. Discrete Imperfection States
If the imperfection states are localized in energy at
ei=
A2
(k„'+kbs+k, '),
the value EI— .
below the Fermi level, the yield will be
2m rl- (E El,).—
ACKNOWLEDGMENTS
F dkbdk, p(es)des
The author would like to acknowledge a great deal of
stimulating discussion with G. W. Gobeli and F. G.
~ (E—Er)s (78) Allen.

P 8 YSI CAL REVIEW VOLUME 127, NUM HER 1 JULY 1, 1962

Direct and Indirect Excitation Processes in Photoelectric Emission from Silicon,

G. W. GOBELI AND F. G. ALLZN
Bel/ Telephone Laboratories, 3EINrray Hill, 37em Jersey
(Received February 23, 1962)

The spectral dependence of saturation photoelectric emission has been studied for atomically-clean
(111) silicon surfaces which were prepared by cleavage in high vacuum. The observed spectra, and their
dependence on sample doping, are interpreted as being due to a volume excitation process which is modified
by space charge band bending efI.'ects. Both direct and indirect optical excitation thresholds are observed,
at 5.45 eV and 5.15 eV, respectively, with the latter being equal to the electron amenity, x, plus the energy
gap, EG. The spectral dependence of the direct excitation process is in agreement with the theoretical model
developed by Kane, in which there is a complete absence of scattering either in the bulk or at the surface
for those excited electrons which are emitted. The indirect process is also in agreement with Kane's theory.
The dependence of the yield on sample doping, in conjunction with the theoretical model, may be used to
determine a direct-flight escape depth for excited electrons of 25 A+5 A for electron energies about 5.5 eV
above the valence-band maximum.

I. INTRODUCTION z-type samples. Work on alkali antimonides also indi-

PHOTOELECTRIC emission from metals has been cates that photoelectrons can originate at depths well
studied extensively, and it is usually assumed that beneath the surface. ' These results again indicate that
light absorption leading to photoelectric emission takes the volume eRect must be considered.
place only at the surface of the metal where the elec- Optical excitation leading to photoelectron emission
trons are subject to the surface effect. Some recent work in a volume eRect would be subject to the optical
on alkali metals, ' ' however, has indicated that photo- absorption selection rules of the lattice and would
electrons can originate at depths below the surface of therefore consist, in general, of an indirect or phonon-
several hundreds of angstroms. Such observations assisted transition and a direct transition in which the
indicate rather that a volume effect dominates the initial and final electron states differ in k vector only
emission. by the photon k vector.
The theory that photoelectric emission from semi- This paper discusses experimental measurements of
conductors is a volume eRect is generally accepted. The photoelectric emission from atomically clean silicon
eRect of band bending at the surface4 ' and the energy surfaces prepared by cleavage in high vacuum, as a
distribution of valence-band density of states' ' have function of the sample doping. The results confirm the
been discussed theoretically. Recently, some work on dominance of the volume photoelectric eRect. As
cesium-coated silicon surfaces' has substantiated the expected, p-type samples have a higher yield than n
theoretical expectation that p-type samples should types samples. In addition, the spectral yields from
exhibit a higher photoelectric quantum e%ciency than certain resistivity ranges show distinct structure, which
is interpreted as the onset of an eKcient direct excitation
' 395 (1959}.
' H. Thomas, Z. Physik 147,
H. Meyers and H. Thomas, Z. Physik 147, 419 (1959).
mechanism about 0.3 eV above the lower-energy in-
' S. Methfessel, Z. Physik 147, 442 (1959). direct excitation threshold. The spectral dependence of
' W. E. Spicer, J. Appl. Phys. 31, 2077 (1960); R. C. A. Review these two components is determined and the inQuence
19, 555 (1958). of the profile of the space-charge region and the mean
' L. Apker, E. Taft, and J. Dickey, Phys. Rev. 74, 1462 (1948).
6 D. Red6eld, Phys. Rev. 124, 1809 (1961). free path of the excited photoelectrons is examined.
' B. Huntington and L. Apker, Phys. Rev. 89, 352 (1953).
' H.
J. J. Scheer, Philips Research Repts. 15, 584 (1960). ' W. E. Spicer, Phys. Rev. 112, 114 (1958),