PHYSICAL REVIEW B

Self-Consistent

15 JUNE 1972

VOLUME 5, NUMBER 12

Results for n Ty-pe Si Inversion Layers

Frank Stern

IBM Watson Research Center, Yorktourn Heights, Nero York 10598

(Received 13 January 1972)

Self-consistent results for energy levels, populations, and charge distributions are given for
n-type inversion layers on p-type silicon. Quantum effects are taken into account in the
effective-mass approximation, and the envelope wave function is assumed to vanish at the surface. Approximate analytic results are given for some special cases. Numerical results are
given for representative surface orientations, bulk acceptor concentrations, inversion-layer
electron concentrations, and temperatures.

I. INTRODUCTION

Section II gives the equations to be solved, and

and compares them with the corresponding self-consistent
results. In Sec. IV we briefly describe the numerical calculation, and in Sec. V we give numerical
results for representative surface orientations,
bulk-carrier concentrations, temperatures, and
inversion-layer carrier concentrations.
Section
VI discusses the relation of the calculated results
to experiment and includes an estimate of the
changes in the calculated energies that result from
relaxation of the condition that the envelope wave
function vanish at the surface. Two appendixes
describe depletion-layer edge effects and give
integrals of powers of z times the square of the
Airy function.

Sec. III gives some approximate results

External electric fields or charges can change

the properties of semiconductor surfaces significantly, and the consequences of such changes have
been systematically explored for many years. '~
More recently there has been considerable work
on the quantum effects that arise whenband bending
confines the carriers to a narrow surface channel.
Some of these effects have been described elsewhere. ' In this paper we give the results of
numerical self-consistent calculations for the energy levels and subband populations of n-type inversion layers on p-type Si. %e use the notation
of Stern and Howard'; readers are referred to that
paper for an introduction to n-type inversion

layers.
The first numerical self-consistent calculations
for inversion layers were made by Howard. Selfconsistent results have also been obtained by
Maeda~ and by Stern. Self-consistent solutions
for accumulation layers were discussed by Duke, 9
and have been studied in detail by Appelbaum and

'

Baraff

'

'"

Our calculation is based on three major approximations. We assume (i) that the effective-mass
approximation is valid, so that we can neglect the
periodic potential and use the effective masses
and the dielectric constant of the perfect crystal;
and (ii) that the envelope wave function vanishes
at the surface. Neither approximation is likely
to be valid at high surface electric fields, for
which the calculated wave functions can extend
10 A or less into the semiconductor.
The third
major approximation (iii) is that we can neglect
surface states and can replace the effect of any
charges in the oxide or insulator adjacent to the
semiconductor by an equivalent electric field.
This approximation is reasonably well justified for
the interface between silicon and silicon oxide
when it is appropriately treated. i~ Estimates of
the errors that result from use of the effectivemass approximation have been made by Howard. '

II. EQUATIONS

The band bending at a semiconductor surface

can be characterized by an electrostatic potential
P(z}. In the effective-mass approximation, the
electronic wave function for the zth subband is the
product of the Bloch function at the bottom of the
conduction band and an envelope function'
y (

~ z)

(z)&f8s&lk&x+Qz&

where k, and kz are measured relative to the band

edge, 8 depends on k, and kz, ' and t, (z) is the solution of

d~f]
'z+
d

2~z [E, +eP(z)it', (z)=0.

(2)

We are interested in the bound solutions of (2), so

we require that t, (~) =0. In addition, we require
that t';(z) vanish at the surface, where z=0. This
should be a good approximation for the Si-SiOz
interface, for which the potential barrier for electrons is approximately 3 eV, until we reach electric fields strong enough to confine the electrons
within 1 nm or less.
An estimate of the error
arising from this approximation is given in Sec. VI.
Each eigenvalue E, found from the solution of
(2) is the bottom of a continuum of levels called

"

"

4891

F RANK STE RN

4892

where g is the dielectric constant of the semiconductor, N, is the carrier concentration in the
ith subband, given by '

a subband, with energy levels given by

E, (k) =E(+ 5

iP, /2m,

5 kz/2~,

where m, and ma are the principal effective masses

for motion parallel to the surface, which can be
obtained in a straightforward way from the bulk
masses. ' Values for these quantities and for the
mass m for motion perpendicular to the surface,
used in (2), are given in Table I for the surface
orientations used in this paper.
There can be as many as three values of m, for
a given surface orientation because the conduction
band of Si has six valleys along the (100) directions
of the Brillouin zone. In the effective-mass approximation, the valleys are degenerate in pairs,
but the degeneracy is found to be lifted in magnetoconductance experiments at high magnetic fields. '
The origin of that splitting has not yet been explained.
The valleys which present the highest mass for
motion perpendicular to the surface have the lowest kinetic energy and the lowest energy levels.
We label the subbands arising from these valleys
with indices 0, 1, 2,
. The (ill) surfaces have
only one ladder of subbands, since all the valleys
have the same orientation with respect to the surface. The {100)and (110) surfaces have a second
ladder of subbands, which we label
2, .
For a general surface orientation, those subbands
which arise from the smallest value of
are
labelled
in this notation.
The potential P(z) which appears in (2) is the
solution of Poisson's equation"

...

0, 1, ...

0, 1, 2,

d ~(t)

~ ~

[.()--Zeel())/

z.,

p.

TABLE

I.

Parameters used in the calculation.

Surface

n2
0.

Lower

(0)

(100)

Valleys
Degeneracy
Normal mass

(4)

Higher
4

Lower
4

Higher
2

All
6

m2

916

0. 190

0. 315

Q. 190

0. 258

mj
m2

0. 190
0. 190

0. 190
0. 916

0. 190
0. 553

0. 190
0. 916

0. 190
0. 674

Conductivity
massb

ms

'1

' 15

0' 283

0'315

0' 296

Density-ofstates mass
per valleyb

m&

0. 417

0. 324

0. 417

0. 358

Longitudinal

masses

0. 190

11.7

Dielectric constant
Permi ttivity
Fermi level
T (K)
E -E (ev)

1.04xIQ

C~=K~tp

for N&-AD=10

in the bulk,

0
12

l.

F/m

See Eq. (9).

300
0. 88

77

1. 11

All effective masses are in units of the free-electron

mass; they are based on the conduction-band masses
m&=0. 190m and m& =0. 916m given by
C. Hensel, H.
Hasegawa, and M. Nakayama [Phys. Rev. 138 A225
(1965)]. See also, Table I of Ref. 5.
b
ms mi m2/ (mi+m2); mfl = (mi m2)

J.

N, = (n, mKT/v"')~p[(Ez E&)/KT1,

Eo(x) = In(1+ e*), n, and mz are the valley degeneracy and the density-of-states effective mass per
valley, given in Table I, and E~ is the Fermi energy. pp, (z) is the charge density of the depletion
layer, which we take to be

pp, (z) = e(NN'),

0&z &z'

Z &Zff
pp, (z) = 0,
where z is the depletion-layer thickness, given by

z'= [2zsc zo~ p/e(N Np)]

Here P is the effective band bending from the bulk
to the surface, apart from the contribution of the
inversion layer itself; its value is given below.
The boundary conditions for Eq. (4) are that
dP/dz vanish for large z and that its value at the
surface be Ewhere

F, = e(N, + No, )/z,

zp,.

Np, = z, (NNe) is the number of charges per

unit area in the depletion layer and N, =gN, is the
total number of charges per unit area in the inversion layer. Note that the electric field just
outside the semiconductor is larger than the field
just inside by a factor z /z, , where z, is the
dielectric constant of the insulator. We have
specified dP/dz at the boundaries, so P itself is
uncertain by a constant. We sometimes choose jt'
to vanish in the bulk, but the energy levels obtained from the solution of (2) assume that P vanishes at the surface.
The assumption that the depletion charge is constant for a distance z from the surface and then
goes abruptly to zero fails in the transition region
from depletion to bulk, in which the field decays
to zero exponentially with a characteristic distance
In Appendix
given by the bulk screening length.
A, we discuss the correction to the depletion-layer
charge, and therefore to z, =N/(NNe), which
arises from this effect. When the correction to P
is taken tobe KT/e, thebandbe-ndingwhich enters
in (7) is

"

P, =[(E, Er), +Ez KT)/e+

P- e-N, z/zzo,

(0)
where the first term is the energy difference between the bottom of the conduction band in the bulk
and the Fermi level, E is the Fermi energy
relative to the nominal conduction-band edge at
the surface, and
is the average penetration
of the inversion-layer charge from the surface.
The substrate bias p, b is the voltage difference

..

SELF-CONSISTENT RESULTS FOR n- TYPE Si INVERSION.

between the source and drain electrodes, which
are diffused into the surface to make electrical
contact to the inversion layer, ' and the bulk; it
is normally equal to zero. The last term in Eq.
(9) is the potential drop across the inversion
layer. Figure 1 shows some of the quantities
which enter in Eq. (9}. Values of E, Ez in the
bulk are given in Table I for several values of T
for NND=10 ' m

'.

III. APPROXIMATE RESULTS

A. Triangular

Potential

Because of the way in which the Schrodinger

(2} and Poisson's equation (4}are coupled
it is necessary in general to solve these equations
numerically to obtain self-consistent results. We
can, however, find approximate results for some
simple limiting cases.
The simplest approximation is to replace the
potential P(z) in (2) by F~ for z & 0 and by an
infinite barrier for z & 0. This is sometimes
called the triangular-potential
It
approximation.
leads to the Airy equation with solutions'
equation

g, (z) =Ai ((2mpeF, /h' )'/P [z (E, /eF,

)]),

(10a)

(lob)
E, = (h /2m, )' ' [', z eF, (i+ -')]
The eigenvalues E, are asymptotic values for large
i, but they are amazingly close even for the ground
state i =0. The exact eigenvalues for the three
lowest states have i+ 4 in (10b} replaced by 0. 7587,
Another
1. 7540, and 2. 7525, respectively.

"

property of these solutions is that the average

separation of carriers in the ith subband from
the surface, defined by

z, = z &P(z) dz/J k, (z) dz,

(11)

Ep=ff'b /Sm, + (3e /zepb)

~ [Np, p, +

Limit

b= bp+

approximation which
we have just described is a reasonable approximation when there is little or no charge in the inversion layer, but fails when the charge density
per unit area in the inversion layer is comparable
to or exceeds that in the depletion layer. When
only one subband is occupied, i. e. , in the electric
quantum limit, a variational approach gives a good
estimate for the energy of the lowest subband.
Fang and Howard' used the trial eigenfunction
z e ~/

(12)

with a single undetermined parameter b. The energy of the lowest state is found after a straight-

forward calculation to be

(13)

&b,

bp= (12m, e

N"/z~epk )'

',

(14)

',

(15)

&b = 4N/3N*;
Zp

= Zpp+ ~zp

zpp=

(9z,,eph /4mpe N")'

5zp 4N/, z pp/9N

Ep = Epp+ 5Ep,
(3)p/p (e pg/ml/2

5Ep =

~N

z e }2/8 (N

(2Ne zpp/3zpcepN

) (Np, + pp

}/Np1/3

N,}.

(16)

Here N = N~, + ~~ N, and N is the net acceptor

concentration.
In the limit N, = 0, for which the

0.4

Vaepl

0.3

EF

The triangular-potential

t (z) (~ hp)&/p

N, (2/b) (NND)]

choice of b minimizes the total energy of the system, in which the coefficient &~ is replaced by, z.
The last term in (13) arises from the curvature
of the potential of the depletion charge density;
it will be small in Si, for which the inversion
layer is generally much thinner than the depletion
layer. We can therefore treat this term as a
small perturbation.
The average distance the charge density associated with the variational wave function (12}
penetrates into the semiconductor is found from
(11) to be zp= 3/b. We therefore find the following
variational results:

g)

Quantum

Pp

when the potential P vanishes at z = 0. The term

involving N, gives the interaction of the inversion-layer charge density with itself; the correct

is 2E, /3eFand that the average of z z is apzz, .

See Appendix 8 for the relevant integrals, including the normalization of (10a).
B. Variational Results for Electric

4893

Zo

(III) p- Si

0.2

NA

17m-2
Ninv = IQ

Np-l02lm 3

LLJ

N 4e I
p

O. l

- I 2x

Q I5m

-2

OK
I

OO

3
z (nm)

FIG. 1. Potential energy V(z) = erat (z) at 0 K for electrons near a (111) surface of p-type silicon with 10 net
acceptor ions per m3 and with 10" inversion-layer electrons per m
The Fermi energy' and the energy Eo
of the bottom of the lowest subband are shown. The dotdash curve is the potential energy with the inversionlayer contribution removed. The potential V(z) was used
to calculate the results in Table III.

F RANK STERN

4894

TABLE II. Comparison of variational (var) and self-consistent (sc) values

lowest subband, as a function of X&, the number of inversion-layer electrons
depletion-layer charges per unit area. The results are for a (100) surface at
the parameters taken from Table I. See Eqs. (15) and (16). All energies are

N.

..

(m-')

1 xlp

2 x]p'4
5 x10'4

1 x10~5
2

x10"

5 x10

1 x1Q'6
2

x10"

5 x10

1 x10"

N, e(~

')

&Eo

Eoo

1.214 x10
1.213 x10
1.213 x10

M. 052
0 050

19.91
20. 65
21 31

-0 ' 048
M. 043

l. 213 x 1p"

23 ~ 54
26. 96

1.212 x10
1.211 x1p"
1.210 x10
1.207 x10
1.204 x 10~'
1.198 x10
1. 192 x1Q

-0 029
-0 017
-0 ' 011
-0 ' 006
-0 ' 003
-0 ' 002

6zo

32 35
48 81
71 ~ 50
108~ 7
195 1

7 ~ 072
7 008
6 945
6 768
6 511
6 093
5 273
4. 522
3 760
2 858

306 8

2 ~ 294

0 ~ 018
0 ~ 018
0 017
0 ~ 015
0 012
0 ' 009
0 ~ 004
0 ~ 002
0 ~ 001
0 ~ 000
0 ~ 000

21 34

20 ~ 28
22 48

49. 81
72. 74

zoo

18 76
19.53

Eo (sc)

19~ 86
20 ~ 60
23 ~ 50
26 92
33 ~ 28

-Q. 037

33 31
49 83
72 75
110~ 4
197 9

311

Eo (var)

110 4
~

197. 9
311 2
~

potential approaches the triangular potential, the

variational values of both Zoo and zo are 5. 9%
larger than the corresponding exact results. The
trial function (12) has too slow a decay at large
values of z; it gives an average of z~ equal to
fzo, compared to the value pzo found for the Airy
function.
The variational results are compared with numerical self-consistent calculations in Table II
for several values of N, /N~, , and the ratios of
the variational and self-consistent values of Eo
and of zo are plotted in Fig. 2. Because of the
scaling property discussed in Sec. IIID, the results in Fig. 2 are valid for all electric-quantumlimit cases provided the correction terms in (15)
and (16) are sufficiently small.
C. Excited States in Electric Quantum Limit

Approximate energy levels for the excited states

can be obtained in the electric quantum limit by
treating the inversion-layer potential and the cur-

08

I.

06

1.

t- I04
I.02

25. 96
~
~

~
~

II

4
inv

5 20

50

80

~depl

FIG. 2. Ratios of variational to self-consistent values

of the energy Eo and of the average penetration zo of the
lowest subband as functions of the ratio of the inversionlayer charge density to the depletion-layer charge density
in the electric quantum limit, when only the lowest subband is occupied. Note the change of scale atN)11+/Npy = 5.

~
~
~

~
~

zo (var)

7. 09

zo

(sc)

2. 86

6 69
6 ~ 63
6 ~ 56
6 38
6 ~ 13
5. 72
4. 95
4 ~ 26
3 56
2 72

2 ~ 29

2 19

7 ~ 03

6. 96
6. 78
6. 52
6. 10
5 ~ 28

4. 52

3. 76

~
~
~

vature of the depletion potential as perturbations.

The inversion-layer charge density is assumed to
be a sheet located a distance zo from the surface.
If we neglect the oscillations in the Airy function'
and assume that the probability density for the
excited states varies as (a, z) '~, zo where a,
=E, z/eEand E', =eN~e/K zp thenwe find
that the energy levels for the excited states are
approximately given by

E(=E~ q e Ed")E&n zo/4

4E f, z/15eE', zz+

eE,zo,

(17)

E,

where
=eN, /vE p allcl Z g is the energy obtained from (10b) by using the depletion-layer field
E" as the surface field. The second term in (17)
is the approximate lowering of the excited-state
energy by the inversion-layer potential well. The
third term gives the approximate contribution of
the depletion-potential
curvature to the energy levels; it is much more important for the excited
states than is the corresponding term for the lowest
subband, given in (16), because of the larger spatial
extent of the excited states. The last term in (17)
gives the inversion-layer contribution to the potential energy at the surface.
Equation (17) applies to the electric quantum limit, but it may still have some validity when more
than one subband is occupied, provided zo is replaced by
z~~

I.OO

of energy Eo and spatial extent zo for the

per unit area, and Ng~g, the number of
0 K, with 10 m net acceptors, and with
in meV and all distances in nm.

=Q&N(z&/Ng-

(1S)

the average distance of all the inversion-layer electrons from the surface.
The quantity zo in (17) can be found by combining
the variational expression (15) and the correction
factor in Fig. 2 The magnitudes of the terms in
(17) and of the energy obtained from a numerical
self-consistent calculation are given in Table III
for excited states of a sample inversion layer.
~

SELF-CONSISTENT RESULTS FOR n- TYPE Si INVERSION. . .

D. Scaling

In the electric quantum limit, Eqs. (2) and (4)

can be written in dimensionless form, with (K ep8 /
2mpePN~)'iP as the unit of distance and (I /2m, )'
x (Np, ep/w ap)pip as the unit of energy, as first
pointed out by Howard. 6 If the inve rsion- laye r thickness is much smaller than the depletion-layer thickness, then the only parameter characterizing the
dimensionless equations is N, /Np, Th. us solutions for other semiconductors can be obtained
from those given in this paper for Si in the electric
There
quantum limit by appropriate rescaling
does not appear to be a simple scaling relation when
more than one subband is occupied.
IU. DESCRIPTION OF CALCULATION

The self-consistent calculation starts with an

initial estimate for the potential P(z) and then solves
Eqs. (2) and (4) successively until the output potential from (4) agrees with the input potential in (2)
to within specified limits. The general method used
to achieve self-consistency has been described elsewhere. ~' Here we give only a brief description of
the calculation.
The trial potential is obtained by assuming that
all the carriers are in the lowest subband, for which
we use the variational approximation described above.
The trial eigenvalues for the initial integration of
(2) are obtained by an approximation similar to (17).
The trial potential generally leads to good convergence except at high temperatures,
for which the
assumption that all carriers are in the lowest subTABLE III. Energies of excited states for a (111) surface at 0 K, given in meV. Columns 2-4 correspond to
the first three terms on the right-hand side in Eq. (17).
The inversion-layer contribution to the surface potential
energy is eE)~~0=338. 5 meV. The fifth column is the
approximate energy calculated from (17), while the last
column gives the energy from a numqrical self-consistent calculation. We use N&~& =1.2 ~10 m (corresponding to NA-ND=10
m ) and Nfg~ 10 7 m
the constants are taken from Table I. The numerical calculation gives z0=2. 189 nm and ED=306. 8 meV; the Fermi
level is at 317.9 meV.
~

1
2

3
4
5
6

7
8

9
10
15
20

Ef, d
32. 43

43. 83
53. 90
63. 10

71.67
79. 75
87. 45
94. 82
101.91
108.76
140.30
168.61

T2

-10.53

7. 79

-6. 34
-5. 41
-4. 76
-4. 28

-3.91
-3.60
-3.35
-3. 14
-2. 43
-2. 03

T3

-0. 13
-0.23
-0.35
-0.48
-0.62
-0.77
-0.93

-1.09
-1.26
-1.44

-2. 39

-3.45

E &(approx)
360. 3
374. 3
385. 7
395. 7
404. S
413. 2
421. 1
42S. 6
435. 8
442. 7
474. 0
501. 6

E;(sc)
361. 9
375. 4
386. 5
396. 3
405. 2

413. 5
421. 4
428. 8
436. 0
442. 8
474. 0

501. 6

4895

band is particularly bad. At high temperatures

we start with small values of N, and then gradually go to larger values, taking the results of each
case to construct the starting potential for the next.

The convergence criteria are that successive

eigenvalues in the solution of the Schr5dinger equation (2) with a given potential agree to within 10
eV and that the potentials in successive rounds of
the iteration differ by no more than the larger of
KT/2000, (Ez Ep)/10, or 10 eV. This does
not mean that the solutions are correct within these
limits. We have not made a systematic study of
the accuracy of our results, but it is thought to be
in most cases. Greatconsiderably better than
er accuracy is not needed for comparison with
experimental results because physical parameters
like the acceptor concentration, the position of the
Fermi level in the bulk, and the dielectric constant
are not known to greater accuracy.
The present program handles up to 25 subbands
numerically, and a crude scaling approximation
for the contribution of higher subbands to the potential is used if required. This is adequate at 300 K
for (100) n-type inversion layers on Si if N& 10 '
m . Fewer subbands are needed for the other
high-symmetry surfaces because they have a largerenergy splitting between the lowest subband and the
first excited subband. At lower temperatures the
number of subbands required drops sharply.
The
present version of the program automatically assigns the grid interval used in the numerical integration, and fixes the number of grid points
needed for each subband. Up to 1201 grid points
can be used, but will normally not be required
except for cases with very many subbands.
The computer program has evolved over a period
of years. It contains 13 subroutines and about
2000 FoRTRAN statements.
Execution of the program
requires 280K bytes of storage, much of it for the
wave functions themselves.
A run with 17 subbands
and 748 grid points, which took six iterations to
converge, required 5. 2 sec of central-processingunit time on an IBM System/360 Model 91 data pro-

1'

cessing system.
U. SELFXONSISTENT RESULTS

In this section we give some representative results of our self-consistent calculations to illustrate
properties and trends in the solutions. A set of
results complete enough to encompass the entire
range of temperatures and doping levels used by
experimenters would be too lengthy to be published
here. We hope nevertheless that this sampling will
be useful.
Figure 1 shows the potential energy at 0 K for a
(111) surface with N& N& = 10 ' m ' and with N,
=10'7 m, the same values as for Table III. This

is a case in

which

N, is

much larger than Nd,

F RANK STERN

4896

= 1. 2 x10" m ', so that the electric field at the surface is much larger than the field in the depletion
layer.
As N, increases from zero at low temperatures,
the separation E~ Ep of the Fermi level from the
bottom of the lowest subband increases faster than

30
NA

ND

' 102l ~-3

Io
C

the splitting E, Eo between the next subband and

the lowest subband. But after the Fermi level
crosses the bottom of the first excited subband, the
rate of change of Ey Ep decreases and the rate
of change of E, Ep increases; the Fermi level
seems to be pinned near the bottom of the second
subband.
This effect is shown in Fig. 3 for a
(100) surface at 0 K for a sample with NNz& 10
m
The results of the self-consistent calculations
are not easily characterized by a single parameter,
but one that is quite important is
the average
penetration of the inversion-layer charge density
from the surface. This quantity is shown in Fig.
4 for NNs = 10-' m 3 for the (100) and (111) surfaces as a function of the total number of charges
at
in the space-charge layer, i. e. , N, +Nd,
2, Vv, and 300 K. Near threshold, i. e. , when the
inversion layer is first populated,
is almost
the same for all three surfaces at 300 K, showing
that quantum effects are relatively unimportant
there. A calculation on the continuum model confirms that the results at 300 K are very close to
those obtained here; they differ substantially at

gNyepi

(77K
t

(a)

30
NA

z,

-ND IIO21gn ~

Ip
E
C

hl

lower temperatures.
As the temperature is increased, inversion-layer
electrons will be excited to higher-lying subbands,
with larger average values of z, and the inversionlayer contribution to the ptoential well increases.
The temperature dependence of zp and zand of the

~N~

(77K)

IPI5

IO16

Ninv+NAp

(b)

20

Io

0I

3x10@

FIG. 3. Energy-level splittings and Fermi energy at

0 K for a (100) surface inversion layer on p-type silicon
with 102 net acceptor ions per m3, showing the change in
slope when the Fermi level passes into the first excited
subband. Note that the bottoms of subbands 1 and 0'
cross where N, /N~~, = 5. 3. All energies are measured
from the bottom of the lowest subband.

elec-

trons into the semiconductor as a function of total density

of charges in the inversion and depletion layers for (a)
(100) and (b) (111) surfaces on p-type Si with 10 net
acceptor ions per m3. The values for a (110) surface are
very close to those for a (111) surface.

splittings is shown for a (100) surface

N~ = 10 ' m and N, = 10 m
A measure of the importance of quantum effects
is the distribution of carriers among the subbands.
Figure 6 shows the fraction of electrons in the
lowest subband for several temperatures for
No = 10 ' m ' as a function Of Ni +Nd pi for the
A
three high-symmetry surface orientations. Also
shown is the fraction of the electrons in all the
subbands in the ladder associated with the highest
value of m3, the mass for motion perpendicular
to the surface. When quantum effects are unimportant, this ratio should be the degeneracy for
this ladder, as given in Table I, divided by 6, the
total number of valleys. We see in Fig. 6 that
the fraction of electrons in the lowest ladder of subbands does indeed approach the expected value at
300 K when N, +N~, is small, but departs from
it both for low temperatures and for large values
of N, +N~, Thus, quantum effects must be considered even at room temperature if accurate rein

(100) Si OK

2)

FIG. 4. Average penetration of inversion-layer

energy-level

30

OIY
(rn

Fig.

for

N&

SELF-CONSISTENT RESULTS FOR n- TYPE Si INVERSION.

t'ai(dg/dz}
(lO

the one present at the Si-SiO& interface. ' If the

field just inside the semiconductor is 2 x10' V jm,
a relatively high field, and ms =0. 916m, then the
scaling distance a* for the abscissa of Fig. 7 is

NA

6 N

M
C)

4897

' the value 0. 2 nm appropriate for a. free

electron in a 1-eV barrier, a smaller barrier than

&c

..

~
H
~ (1)
4
UJ

%+
QJ0
)CL
+U

I.O

(a)

IN LOWEST VALLEYS
FRACTION IN LOWEST SUBBAND

(b)

I20

8-

---- FRACTION

cn

pfj'Z

0.

~ 0.6
IOO

4J

4.

p
p2 04
I-

80

O
K

LLI

02-

60=
Ep

40

{a)

200

IOO

300

I.O

TEMPERATURE (K)

FIG. 5. Temperature

(n

of {a) zo and z,
the average penetration of electrons in the lowest subband
and in all subbands, respectively, and (b) the energies
of the lowest seven subbands, for a (100) surface with
Note that levels 0'
Nz-iV&=10 m and N&~=10 m
and 1 cross at about 35 K. Note also that Nd, decreases
from 1.21x10 5 m at 0 K to 0. 96x10 m at 300 K. If
the surface field were held constant at the value for 0 K,
splittings at 300 K would be about 1% larger than the
values shown here for a fixed value of N(.
dependence

8
O

~~

0.8
0.6

LLI

z 04
O
t

02

-
- FRACTION

IN

LOWEST VALLEYS

FRACTION IN LOWEST SUBBAND

I

(b)
I.O

suits are to be obtained for high surface-charge

densities.

O.

VI. DISCUSSION

Sections III-V give representative results of selfconsistent calculations for n-type inversion layers
on Si. In this section we comment on the validity
of some of the approximations that we use.
The assumption that the envelope wave function
vanishes at the surface simplifies the calculation
considerably, but a more realistic boundary condition specifies the ratio of the wave function to its
derivative at the surface, a value which depends on
the nature and height of the barrier there. The energy change connected with this more general boundary condition is easily calculated for the triangular
potential, whose eigenfunctions and eigenvalues are
given in (10), by using tabulated values of the Airy
function and its derivative. ' The dependence of the
eigenvalue on the boundary condition is shown in

Fig. 7.
For a numerical estimate of the effect of more

general boundary conditions,

let us take for

$o
iCt

~~ 06
U

pm

Op
iO

c[ Z
K
u.

04

NA- ND

102jm

(III) Si

0.2
Nd~i (77K)

IOI5

ioj~
Ninv+ Nd

IOI7

(m-2)
I

(c)
FIG. 6. Fraction of carriers in the lowest subband,
and in all the subbands associated with the valleys having
the largest value of m3, the mass for motion perpendicular to the surface, for the (a) (100), (b) (110), and (c)
(111) surfaces. The fraction in the lowest valleys should
be 3 and ~, respectively, for the (100) and (110) surfaces
when quantum effects are negligible.
It is identically
unity for the (111) surface, for which all valleys have the
same value of m3.

F RANK STERN

4898
I

Ep

UJ

3
Ep

Q/0 (dg/dz)

IO

[, 0

FIG. 7. Effect of boundary conditions on energy levels

for the triangular potential. The abscissa is the scaled
ratio of the envelope wave function at the surface to its
derivative there. The scaling distance a* is
(8 /2m3eF~), where E~ is the electric field, and the
scaling energy E* is eE, a*. The energies for dg/dz = 0.
are marked at the right-hand side.
nm and the scaling factor E* for the energy
scale is 0. 155 eV. For a value of 0. 38 on the
abscissa of Fig. V the energy levels are reduced
by less than 15% from the values found with a wave
function which vanishes at the surface, and the
energy-level splittings are increased by less than

0. 52

We conclude that within the framework of the

approximation the errors made in
taking the wave function to vanish at the surface

effective-mass

are relatively sma11.

The effective-mass approximation

itself will be-

gin to fail when the calculated energies are comparable to the energy gaps in the bulk and when the
spatial extent of the wave function for the lowest
subband is comparable to atomic dimensions.
The
resulting errors have been estimated by Howard';
they become significant for wave functions whose
average penetration z is of order 2 nm or less.
Exchange and correlation effects in the inversion
layer have been ignored in our work, an approximation which requires further examination.
At the highest surface fields it may be necessary
to undertake a more fundamental calculation of the
effects of interfaces on wave functions and eigenvalues. This problem is somewhat like the problem
of low-energy electron diff raction, because both involve the matching of waves across an interface between two media. It may be possible to include
within the framework of a simple model both the
surface subbands associated with band bending
and the conventional surface states which arise
from the presence of the interface itself.
The most promising experimental method to
determine inversion-layer subband splittings for
comparison with the calculations appears to be
to look for structure in the photoconductivity ex-

citation spectrum. This method was first used by

Katayama et al. 22 for InSb, and has been used for
The latter authors
Si by Wheeler and Ra1ston.
used a laser with a fixed photon energy, and varied
the surface electric field to change the subband
splittings. We have calculated the energy levels
expected from the parameters they report, but
find considerably smaller subband splittings than
the 44 meV of their laser if we use their subband
Alternative assignments also did
assignments.
not fit. It is probably too early to draw conclusions
from this comparison for a single sample, but
more extensive measurements should allow a
comparison with self-consistent calculations to be
made. Experimental information on subband splittings in accumulation layers is provided by the tunneling measurements of Tsui.
ACKNOWLEDGMENTS

The author is indebted to A. B. Fowler and W.

E. Howard for many valuable discussions, and to
M. I. Nathan for comments on the manuscript.
APPENDIX A: DEPLETION-LAYER-EDGE EFFECTS

We have assumed in Eq. (7) that the depletionlayer charge density is constant for 0 &z &z~, and
then changes abruptly to zero. The actual charge
density decays smoothly, with a characteristic
length given by the bulk screening length. Near
the surface, this effect can be accounted for by
adding a constant to the band bending used to calculate z, in Eq. (7). We must now think of z~ not
as the abrupt termination of the depletion layer,
but as a measure of its charge density, defined by
z, = N~ /(N- Ne ). We give the required correction term in this Appendix, and also give values for
the screening length. The derivation is straightforward, and has been omitted. '
If the bulk is characterized by acceptor and
donor concentrations N& and N&, respectively, and
by hole concentration p, if all the minority impurities are ionized, and if there is a negligible
minority-carrier concentration, then we find that
the bulk-screening length L is given by 5

L =KI, KOKTN/e fN~p+(Ng&+p)(N-Nn-p)]

At high temperatures,
the familiar result

L =KeoKT/e

where

p-N-N&,

(NNn)-

while at low temperatures,

where P

I. =KeoffTN/e Ne(NNe) .

(Al)
we get
(A2)

N-N~,
(AS)

The surface-charge density K &o(dp/dz} can be

as a function of (t) by multiplying both sides
= dP/dz and integrating.
of Poisson's equation by @
found

We find

SELF-CONSISTENT RESULTS FOR n- TYPE Si INVERSION.

for the cases we consider,
in other calculations.

= (2/zzo)((N~ Nz)eg

KT[P+N in[N/(Nz+P)]})

(A4)

The term in KT is the correction term which takes the

edge effects into account. At high temperatures,

"

KT)],

= (2/zzo)[(Ng Np)(eg

(A5)

while at low temperatures,

].
(AS)

At very low temperatures it is necessary to consider the width of the acceptor level. If we characterize it by a Gaussian distribution whose full energy width at half-height is 4, we find the following
approximate results when KT b, :
4z z n/e'N

'"N

'"

(A7)

@"= (2/z, z, ) [(NN, }e-y

0. 45&N

Nz

'

(AS)

The numerical coefficients and exponents are approximately correct for Ne/N= 0. 01.
result (A5)
We have used the high-temperature
in Eq. (9) and in our calculations over the entire
temperature range. The difference between (A5)
andthe more exact results (A4) or (AS) is negligible
~A Many, Y. Goldstein,
and N. B. Grover, Semiconductor Surfaces (North-Holland,
Amsterdam, 1965).
2D. R. Frankl, Electrical Properties of Semiconductor Surfaces (Pergamon, New York, 1967).
R. F. Greene, in Solid State Surface Science, edited
by M. Green (Marcel Dekker, New York, 1969), Vol. 1,

p. 87.

F. Stern, in Proceedings of the Tenth International

Conference on the Physics of Semiconductors, Cambridge,
Massachusetts, 1970 (U. S. Atomic Energy Commission,
Oak Ridge, Tenn. , 1970), p. 451.
'F. Stern and W. E. Howard, Phys. Rev. 163, 816
(1967).
W. E. Howard (unpublished).
'H. Maeda (unpublished).
F. Stern, Phys. Rev. Letters 21, 1687 (1968).
9C. B. Duke, Phys. Rev. 159, 632 (1967).
' J. A. Appelbaum and G. A. Baraff, Phys. Rev. B
4,
1235 (1971); 4 1246 (1971).
~'G. A. Baraff and J. A. Appelbaum, J. Vac. Sci.
Technol. (to be published); Phys. Rev. B ~5 475 (1972).
The author is indebted to these authors for preprints.
See, for example, A. B. Fowler, F. F. Fang, and
F. Hochberg, IBM Res. Develop. 8, 427 (1964).
W. E. Howard (unpublished).
' R. Williams, Phys. Rev. 140, A569 (1965).
i'We use mks units in this paper. To convert to cgs
units, replace &0 by 1/4z. Note that 1 nm =10 A. .

J.

4899

but might be significant

B: INTEGRAI. S OF AIRY

FUNCTIONS

In connection with the normalization of the eigenfunction (10a) of the triangular-potential

well and
with the calculation of averages of powers of z, we
note the following results:

= (2/&&0) [(N~- Na)e p NKTln(N/ND)

L'-0

APPENDIX

..

Ai (z')dz'

=-z Ai

(Bl)

(z)+Ai (z),

f, z'Ai (z')dz' =-,'[- z Ai (z)+zAi'z(z)

Ai(z)Ai'(z)],

z Ai (z

(a2)

)dz'= ,'[-z Ai (z)+z Ai (z)

2z Ai(z) Ai'(z) +Ai'(z)]

(a3)

These results are easily verified if we use Ai (z)

=z Ai(z), where Ai and Ai denote first and second
derivatives, respectively, of Ai with respect to z.
For the boundary condition used in our work, the
lower limit of the integrals is (8 /2m~eF, )'i times
the (i + 1)st zero of the Airy function, ' for which
the results take a much simpler form. To obtain
the results quoted in Sec. III A, note that the z axis
here has its origin at the turning point where the
energy is equal to the potential energy, while in
the body of the paper the origin is at the surface.
The difference is a&=E, /eF, .
Fang, W. E. Howard, and P.
Phys. Rev. Letters 16, 901 (1966); and J.
Phys. Soc. Japan Suppl. 21, 331 (1966).
'The high-temperature
case has been treated by J. A.
Lewis, J. McKenna, and E. Wasserstrom [J. Appl.
Phys. 41, 4182 (1970)). See also Refs. 1 and 2.
See, for example, in Handbook of Mathematical Functions, edited by M. Abramowitz and I. A. Stegun (U. S.
GPO, Washington, D. C. , 1964), Chap. 10.
'~F. F. Fang and W. E. Howard, Phys. Rev. Letters
16 797 (1966).
~~This is equivalent to the classical result that the
probability density is inversely proportional to the velocity. See, for example, L. I. Schiff, Quantum Mechanics (McGraw-Hill, New York, 1955), 2nd ed. , pp. 65-66.
F. Stern, J. Compt. Phys. 6, 56 (1970).
22Y. Katayama, N. Kotera, and K. F. Komatsubara,
in Proceedings of the Tenth International Conference on
the Physics of Semiconductors, Cambridge, Massachusetts, 1970 (U. S. Atomic Energy Commission, Oak
Ridge, Tenn. , 1970), p. 464.
R. G. Wheeler and R. W. Ralston, Phys. Rev.
Letters ~27 925 (1971).
24D. C. Tsui, Phys. Rev. B 4, 4438 (1971).
'Corresponding results for n-type material are given
by H. Brooks [in Advances in Electronics and Electron
Physics, edited by L. Marton (Academic, New York,
1955), Vol. VII, p. 158).

A.

B. Fowler, F. F.

J. Stiles,