Professional Documents
Culture Documents
Abstract-Previous theory is extended by including the temperature dependence and image force
correction and by a careful consideration of the electron energy ranges which give the main contri-
bution to the tunnel current. The effects on the volt-current characteristic of a non-parabolic energy
momentum relation in the insulator, and a conduction-band-edge effective electron mass in the
insulator which differs from that in the outer metal layers, have both been derived. The published
experimental dependence of the tunnel current on voltage and temperature, for Al-AlsOs-Al sand-
wiches, is in reasonable agreement with the theory if the effective electron mass in the insulator is
about a quarter of the free electron mass and the metal-insulator work function, for a 48 A thick
film, is about 1-l eV. Contrary to previous theory, it is predicted that the deviations (by about five
per cent) from the zero temperature and zero bias ohmic resistance will occur for temperatures in
excess of about 150°K and voltages in excess of about 0.1 V.
mass in the insulator which differs from that in the values of x for whichp, = 0, and
the metals is derived in Section 5. Here the effect
of a non-parabolic energy-momentum relation in Pf = P;+P,2 13)
the forbidden gap of the insulator through which
3. TUNNJ2LING VOLT-CURRENT CHARACTER-
the electron tunnels is also considered.
ISTICS (FREE ELECTRON MASS)
For parabolic ener~-momentum relations with
2. BASIC TUNNELING EQUATIONS
the free electron mass m, in each of the three
PRICE and RADCLIFFE@) have shown that the
regions equations (1) and (2) reduce to h
current density in the x-direction due to electron
tunneling from a conducting region 1 to a con-
ducting region 2, through the forbidden energy gap
of an insulator or semiconductor, is given by
0 0
where
-L _ CONDUCTION BAND
\~
XI i E. (P.)
There is a further condition due to the use of the expanding bl and cl as power series in V yields
W.K.B. approximation for calculating I’(&)
bl = blo-bllV-i&v~ (17)
[equation (S)]. Again following MURPHY and
GOOD,(~)we can write this in the form Cl = c10- c11V+crsP (18)
(f~-[l-kT(l-clkT)-f > R, P(Q) = e-1 (12b) on taking negative signs in front of the linear terms
for convenience. In Section 3.2 it will be shown,
where & is the maximum value of $1(x). [This is
by special examples, that subsequent coefficients
a slight generalization of the argument used by
in these expansions differ from preceding ones by
MURPHY and G00D(5f to derive their equation
a factor of order (l/x) where x is the barrier height,
@I).] In practice this condition merely limits the
above the Fermi level, at zero bias. Thus if cl0 is
applied voltage to a value just less than that
considerably greater than unity (as is the case for
leading to & = {I and also requires clkT to be
the examples treated in Section 4), cl can be re-
less than about one.
placed by cl0 in the exponent of the exponential
Integrating by parts in equation (4) leads to the
term since it would already be negligibly small
familiar form
compared to one for values of V that gave appreci-
4mnkT m able contributions to cl from terms other than the
j==4 - k3 %%J first. Under this condition, it is also reasonably
s accurate to replace cl by ~1s in the other terms
0
since the factor exp( -blV) will dominate the
1
f. + exp{(ti - &J/k T) voltage dependence of j. Thus, to terms of order
x in (13) V2 in b,
l~e~{(~l-~~-~)~kT~ m&T
‘Z=*
[exp(knV-klzl?l
Here the factor multiplying P(E& which has been ’ ‘JO sir&i&T)
called the supply function, represents the differ-
ence between the number of electrons per second
x (1 -exp[-cl0VJ) w
if
per unit area, having (x-directed) energy in the
range E, to Ez+dEo, incident on opposite sides cloX$l, klsV31
of the barrier. Inserting the value of P(&) from and where
equation (8) and carrying out the integration (c.f. .&a = A exp(-ho)(~wW-~ (W
Ref. 5) then leads to
is a constant (N-B. A is proportional to T2).
._ A exp( - bl) mlkT For barriers which are symmetrical with respect
’ - (c~kT)2 sin(~~~kT) (l-ex”-c’VI) (14) to the centre of the gap [x = &).&I equation (19)
can be even further simplified. Since, from
where equations (6) and (9),
s
X*1
i(0) sin(?rclkT)
(The last integral is zero by symmetry.) Thus, for where the image force correction is
symmetrical barriers
rclOk T
j kj0 exp( -blsVs) 2 sinh
sin(rcmkT)
and
(23)
For asymmetrical barriers, bls = bss and cl0 = $(z) = d ln(z)!/dz (29)
css but bll # bsr, nor will any other pair of co-
efficients be equal in general. In fact, it can be is GAUSS'S p.s.i. function. The quantity K (omitted
shown that equation (22) is generalized to in Ref. 2) is the appropriate high frequency di-
electric constant of the insulator; it will be set
h+b21 = Cl0 equal to the square of the refractive index.
(22')
Evaluation of the various integrals in the pre-
To consider the low-voltage ohmic region in vious section with this expression for 4 would
more detail, we expand the current density given require a considerable computer program. We
by equation (19) as a power series in V. Then have instead chosen to evaluate the volt-current
characteristic for three simple barrier profiles.
j/V = G(T)[l+ V(~I-BQO) Our reasons for this are :
(a) The potential near the metal surfaces will
+ ~2(8~~,-~h1a0+~ C;o-b12)...] (24)
not be correctly given by equations (27) and (28)
where in any case.
G(T) = jloc&clokT/sin(?rclokT)] (25) (b) The use of a parabolic energy band in the
gap region makes a detailed computer solution of
is the zero-bias conductance. Thus the coefficients doubtful value for application to thin film sand-
of the linear terms for voltages applied in opposite wiches, especially in view of (c) below.
directions are equal but of opposite sign. For a (c) The main features of the volt-current
symmetric barrier the coefficient of the linear characteristic can be seen for the simple barrier
term on the right-hand side of equation (24) models and good agreement with presently avail-
disappears, while that of the quadratic term be- able experimental results has been obtained.
comes equal to (+)k2,, (neglecting 61s in com-
parison). However, here we must be more careful The function &(x) has a maximum at x = +L
and consider the effect of putting cl equal to crs in given by
equation (14). Inclusion of the linear term in the ML) = -(n2/KL) ln 2 (30)
expansion of cl [equation (18)] leads to an addi-
tional linear term First &(x) will be replaced by the parabolic form
(N.B. q2 = 14.4 eV A so that all the inequalities By changing to our notation (using Table 1) it will
are very reasonable.) be seen that the expression for G(0) of HOLM and
Substituting this simplified image force barrier KIRSTEIN~ differs from equation (39) by having
potential [equations (27) and (31)] into equation the additional factor 2[1 +(l/claXr)]. The factor 2
(5) gives (c.f. equation (6)) arises because they calculated 0 from an equation
similar to equation (28) but evaluated as a function
-In P(E,) = naL[( V+ X1- X$ of Es, not just at the Fermi level. This implies a
change in the shape of the barrier with changing
+4e(x,+ c,-gv)y8e3~2 (36) electron energy. The quantity cn~Xr in the other
Thus
factor is usually very much greater than one.
blo = mtL[(X1 - X2)2+ 4&,]/803/2
BET&S(~) earlier result is actually very similar;
bll = ~a@ - (x, - &)]/‘t83/2 it differs from ours by the factor 2(X~/Xr)l’~ which
(37) arises partly because of a different method of
612 = mL/8tW
evaluating the pre-exponential coefficient.
Cl0 = 7raL/2lw Next we replace &(x) by the triangular form
and all other coefficients are zero. Thus equation &b(x) = -(q2/KL) ln 2-E[x-&LI (42)
(19) is not restricted to small values of I’ for a
parabolic barrier. The rectification ratio is given and consider symmetrical barriers only for the rest
bY (V > 0) of this section; the extension to asymmetric
barriers is obvious. The coefficient F is determined
1J$$- / =exp [“zL(:i2x2)“] (38)
similarly to 0 by equating tunneling paths [c.f.
equation (33)]. Thus
(The second term in the square brackets is usually (5)” (1-g;) > F (53)
negligible compared to unity.)
Expanding b as a power series in V leads to the Since the quantity on the right-hand side is very
values of the coefficients listed in Table 1. much less than one, this inequality essentially
Finally replacing +6(x) by the rectangular form limits V to a value slightly less than Xr.
For V in excess of Xr the last term in each
d&) = -(q2/xL) ln 2 Xl < x < xs (48) equation (50) and (51) is deleted. The limit on V
+--co x = Xl, x = xs is then given by inequality (12b) which leads to
where (V/L) < 8 aX$/2 (54)
Xl/L k l-(X& k fj2/4KLX
< 1 (49)
The expression for j then has the familiar form for
In the more usual form(T) of this approximation field emission from metals@) with neglect of the
x1 = 0 and xs = L. We have identified x1 and xs image force and with x1 = 0.
with the turning values given by the image-force In Table 1 the coefficients of the leading powers
barrier so that the results can readily be compared of V in the expansion of b have again been listed.
with those for the previous two simplified barrier In summary, equation (23) represents the volt-
shapes. Substituting into equation (5) and carrying current characteristic for the three simple barrier
out the integration leads to shapes with the coefficients, listed in Table I,
differing only by numerical factors. The equation
vxl 313 is only applicable to the triangular or the rectang-
-In P(E,) = f $ X,+C~-~
ular barrier if cubic and higher order terms in the
[i >
expansion of b may be neglected. For the p.arabolic
barrier, however, the voltage is not limited by this
requirement since b13 and the higher coefficients
are all zero. Considerable differences in the shapes
the last term being deleted when V > (XT+ of the volt-current characteristic for the various
l%)L/xs. Thus, using equation (8), and expanding barriers arise when V 2 X,.
BETHE'S replacement of the image force
fl =
$[ (+ !$)-“‘_ (xr-!y2] be approximately rectangular unless there are
appreciable space charges in the insulator.
The conditions (12) for the expansion will now 4. COMF'ARISON WITH EXPERIMENTAL
be considered separately for small (V < X,)and REsuLTs
large [V M 0(X,)]voltages. For small voltages, It will now be shown that the theoretical volt-
1184 R. STRATTON
current characteristic [equation (23)] can be fitted line in the inset in Fig. 2. The two points are the
to existing experimental data on Al-AlsO~Al thin experimental values taken from FISHER and
film sandwiches by a suitable choice of the co- GIAVER’S paper and are in remarkably good agree-
efficients bla, blr (= &la) and brs. ment with the theoretical values. From the value
The experimental points in Fig. 2 were obtained ofja and brl we can deduce bra, using equation (20),
by FISHER and GIAVER(~~) (their film 1 in Fig. 2) at
room temperature for a film 48 h thick. By plotting blo = 27.5
Table 2. (L = 48 A)
.- --
025 I I I I I I I
0.20
0.15
"E
is
=s
010
0.05
__ limit on EL,
ufiper - for a given E,is which should be compared with equation (4).
Integrating by parts then leads to - . .
(nac/m)E= (1-G (57)
Thus, from equations (1) and (56),
4mn&T Q)
j=q- cW#W&--'yP(yEz)l
h3 s
0
which is similar to equation (13) for the vacuum From the last three equations it is easily seen that
gap except for the correction term involving Kane’s band structure degenerates to Franz’s
yP(yE%). If m, > m, y is negative and P(yE,) de- when m, < m. We have chosen Franz’s band
creases as E, increases from zero. Hence, the cor- structure rather than Kane’s for our calculation
rection term will involve the tunneling probability since this leads to known (elliptic) integrals.
for electrons from the bottom of the conduction Further, it can be shown that the two band
band in metal 1 and will be extremely small. Thus structures actually lead to identical values of
the effective masses deduced in the previous P(E,pypz) for p-n junction tunneling. This
section could correspond to rnf if m, > m_ strengthens our belief that Kane’s band structure
If mC < ml,and P(E,) is expanded in the form will not lead to results qualitatively different from
of equation (8) a treatment similar to that of those we calculate for Fanz’s.
Section 4 in Ref. (10) leads to Substituting for pa into equation (2) from
equation (61) leads to
exp( - by)
sin:ICtiT) (I- exp( - cl V G.
i = Ai [ (qkT)2
lnP= -ai [E, - (E- YVO+ (E- W%>11’2
dx
s
Y exp( -A) Xl
- - exp{ - rGV) . (65)
(roq sinIL:T) (’ I
(60) crz
= - E1’2 “~(EF#“(+-EI)‘/~ dx (66)
where b; and h are defined similarly to bl and cl g’ z1
[equations (9) and (lo)] except that the barrier
where El and Ea (Es > El) are the roots of the
height is increased by (l-y&. The correction
quadratic equation
term becomes appreciable when y is close to one
or (m,/mi) is much less than one. +(2E+E,#+(E2+EEg--EEJ = 0 (67)
Next, instead of assuming a parabolic energy
momentum relation for the insulator, we will use for c#. The turning values x1 and xa are the roots of
the empirical relation proposed by FRANZ the equation C/J(X)= El.* [N.B.: If Eg $ E,,
El --f E-El and Es + E+ Eg so that equation
p2/2ml: = (E-W + (E- W%I (61) (66) degenerates to equation (55) for a parabolic
band as expected.]
where Eg is the width of the forbidden energy gap
As an example we now consider the parabolic
and C$now represents the bottom of the conduction
potential barrier given by equations (6) (27) and
band for the insulator. This equation yields a
(31). Then equation (66) can be rewritten in the
parabolic energy-momentum dependence near the
form
bottom of the conduction band and the top of the
valence band with both the hole and electron mass
equal to rnf. Franz’s empirical band structure is
closely related to the band structure that ICANE
lnP=
2cqL
-- E1,2
s
;z;+zs)“s(zf -es) dz (68)
g 0
derived for InSb which can be written as
where
.z = (4L)+4(V+x1-x2)/8-~
1+ (69)
is the maximum value of $. The integral can be Furthermore, the supply function for the lower
evaluated in terms of elliptic integrals and [using energy electrons will be greater. However, it
equation (67)] seems most likely that Xr/Eg will be considerably
4EL 314
In P = -;$sE$? l-l”---- k%(k) (72) I
i Eg )
where
@ -= 4
z
(Ilm-E-Wg
+A (74)
2;+22, E,[l f (4EL/Eg)]1’2 2
InP = -:%:(9,-E)a(k)
EL ET -045
x l+O -, (7%
[ ( EB E&&z - Ez) )I
v(K) = [(l-2P)E-(l-P)K]/[2a(l-k2)] will be obeyed, say to within about five per cent,
for voltages up to about l&i [c.f. equation (24)]
= - 8 k2(1+# k2)+0 (k6) (79)
i.e. about 80 mV for the sandwiches analyzed in
and &0,&l, &a and ho are the coefficients defined Section 4. At a given voltage, the tunnel resistance
by equation (37). (c.f. Fig. 5). Using the expansions will decrease by about 5 per cent from its zero
of CTand p for small k2 we then find temperature value when T is about (2mlk)-1 [c.f.
equation (16)] i.e. about 150°K using the data of
bl0 = %o[l- Q(wq7)I Section 4. These results should be contrasted with
the statement in Ref. 3 that a one per cent deviation
bll = &I[1 + +%(X1 - x2)2/w7H from Ohm’s law will only occur for voltages in
bl2 = ZlZ[l - g(&/&) (80)
excess of 0.5 V and that the temperature depend-
ence is negligible below the “region of thermionic
+ ~(@7){1 -(X1 - X2)/W
emission”.
-*x1 - X2)2/&7Ql 1 In Section 6 we derive the effect of deviations
from the free electron mass model on the volt-
Although the general trend is to increase the current characteristic. If the energy bands are still
theoretical value of bls (N.B. 0 +4X1/7) it is un- parabolic and the effective mass in the metals is not
likely to be sufficient to remove the discrepancy much less than that in the insulator the previous
between theory and experiment noted in Table 2, results hold with the free electron mass replaced
even if k2 were too large for the expansion of p(k) by the insulator effective electron mass. Other-
and o(k) to be valid. wise, the current is reduced [c.f. equation (60)].
The non-parabolic relation between energy and
6. CONCLUSIONS momentum in the forbidden gap of the insulator
In the free electron mass approximation the volt- should only be important if the metal-insulator
current characteristic is given by equation (14) work function is an appreciable fraction of the for-
[provided condition (12) applies] where bl and cl bidden energy gap. If it is, then the current density
can be derived from equations (9) and (10) for any will be larger than the value estimated on the basis
potential barrier shape. The reasonably good of parabolic energy bands [c.f. equations (14)
agreement between the measured volt-current and (80)].
characteristics of Al-AlsOa-Al sandwiches at
various temperatures, and the theoretical predic-
tions, based on a rectangular barrier, support the REFERENCES
idea that tunneling was observed. However, the 1. FRENKELJ., Phys. Rew. 36, 1604 (1930). See also
relatively large layer thicknesses (for tunneling) EHRENBERC W. and H~NL H.. 2. Phvs._ 68._ 289
(1931).
estimated by FISHERand GIAVER@~)lead to low
2. SOMMERFELDA. and BETHE H. Handbuch der
effective masses (= 0.3 m) for the oxide. An in- Physik (Edited by GEIGERH. und SCHEELK.),
dependent estimate of this effective mass would Vol. XXIV/2, p. 450 Springer, Berlin (1933).
be very desirable. The value of the reduced HOLM R. and KIR~TEINB., 2. Tech. Pkys. 16,488
metal-insulation work function (X, = 1-O eV) (19351.
Fiw~& R. and NORDHEIML., Proc. Roy. Sot. A119,
involves only the ratio @l&l) and should thus 173 (1928). NORDHEIML. ibid A121. 626 (19281.
be reasonably accurate if the theory applies. MURPHYE. L. and GOOD R. H. JR., Piys. Rekr. lOi,
Further the value of the unreduced metal-insulator 1464 (1956).
work function (X = 1.1 eV) is only slightly de- GOOD R. H. and MUELLER E. W., Handbuch der
Physik (Edited by FL~GGES.), Vol. XXI, p. 176.
pendent on the thickness. Since the volt-current
Springer, Berlin (1956).
characteristic for a parabolic potential barrier 7. HOLM R.,J. appl. Phys. 22, 509 (1951).
applies for all voltages up to breakdown, a detailed 8. PRICEP. J. and RADCLIFFE J. M., IBM J. Res. Dew.
computation for other barriers shapes will only be 3,364 (1959).
HARRISON W. A., Pkys. Rev. 123, 85 (1961).
useful when sufficient high voltage experimental
109: STRATTON R., Phys. Rev. 125, 67 (1962).
data has been obtained. 11. FISCHER J. C. and GIAVERI., J. appl. Phys. 32,127
For a symmetrical potential barrier, Ohm’s law (1961).
1190 R. STRATTON
12. MEAD C. A., Phys. Rev. Letters 6, 545 (1961). a > b. Then
13. PIMBLEY W. T., J. appl. Pkys. to be published.
14. FRANZ W., Handbwh der Physik. (Edited by 2 dlnP
FL~GE S.)Vol XVII, p. 155. Springer, Berlin
(1956). -,dE, = C?J jc[+;:.,,
15. KANE E. O., J. Phys. Chem. Solids 1, 249 (1957).
and
2dzlnP 1 a du
---=-
APPENDIX cc dE; a-b sb [+1--51+%]“2
Derivation of the coeficients in equation (8)
From equation (5) x [(+I - &-I- lX)1’2{(dx2/d.EX) - (da/dEz)} (A.3)
dx 1
1 d In P(Ez) 1 xp x2-33
--
W)
0: dE, = i s [#I - I&+ Q/2 -Z(+l-51+@2
+(a-u) 2
which leads immediately to the value for CL Further X
differentiation of the integral leads to a difference be- Now
tween divergent quantities [N.B. XI and xs are functions
of Ez through the relations &(xr,s) = &I. This can be d&(x) d&r d-xl 1
-=-~r-=-
circumvented by introducing the linear transformation (A.4)
dxl bl d&z 4'da)