Solid-State

Electronics

Pergamon

Press 1963. Vol. 6, pp. 459-462.

Printed in Great Britain

THE GEOMETRIC FACTOR IN SEMICONDUCTOR FOUR-PROBE RESISTIVITY MEASUREMENTS

Abstract-The application of the Green-function method to the mathematical problem connected with four-probe semiconductor resistivity measurements has led to a quite general solution of this problem. In the particular case of a linear array of probes lying on parallelepiped samples, the solution can be reduced to a more compact form. The solution is further simplified for the case of thin samples, where a slight approximation leads to the possibility of presenting it in closed form. R&urn&-Lapplication de la methode de la fonction de Green au probleme mathematique associe aux mesures de resistivite du semiconducteur a quatre sondes a men6 B une solution quasi-get&ale de ce probleme. Dans le cas particulier dun ensemble lineaire de sondes se trouvant sur des Cchantillons de parallelepipedes, la solution peut etre reduite a une forme plus compacte. La solution est encore plus simple dans le cas dechantillons fins, oti une leg&e approximation mene a la possibilite de la presenter dans une forme fermee. Zusammenfassung-Die Anwendung der Green-Funktion zur Losung des mathematischen Problems bei der Messung des spezifischen Widerstandes eines Halbleiters mit vier Sonden hat eine allgemeine Losung ermiiglicht. Im Sonderfall einer linearen Anordnung der Sonden auf Parallelflach-formigen Proben wird die Losung einfacher. Sie wird weiter vereinfacht, wenn man diinne, rechteckige Proben in Betracht zieht; hier fiihrt eine geringfiigige Annaherung zur Moglichkeit der Darstellung in geschlossener Form. 1. INTRODUCTION IN
THE

four-probe of current electric being two points

method

one

measures that into the

the appear

difference between electric (Fig. formula

potentials introduced is

of the semiconductor set of two points

surface, sample by the

by way of another 1). The

on the surface

resistivity

calculated

V p=LI
where contacts The tank

(1)
dimension of length. of the four sample

L is a factor
are given, value of

with

the

If the relative size and shape.

positions

and distances

L is a function

of the from

L can be deduced
However,

electrolytic work has

measurement.

some

been done in the past on the analytical calculation of the L function;(l-5) in all cases the method of electrical images was used. In the following, a general expression is derived
459

for L by making use of the well-known the Green function. The particular

method of case of a

parallelepiped sample is treated in detail. The results that follow were obtained several years ago; in as much as the author knows, they have not been published yet.
2. CALCULATION OF L BY THE METHOD GREEN-

and I- = r-(i,)c-(r?) = 7[0 1 r,A;l[jjL(rJ) -/k(r41 solution of the problem (!I

-.fk(f7)I[h(~3) \vhich giws the general

FUNCTION

(I,!

I = 2 1 ,~\,l[f~!ri)-.~.(r~)l[.~~(l.3)-.fl,.(~.~~)l
3. THE PARALLELEPIPED SAMPLE

H\- assuming, as usual, that the linear eztcnsion 0i the measuring contacts is verv much smaller than the probe spacing s, one can n-rite Poissons equation for the electric potential

(1Oi in the

I:or

the symmetrical
the analysis result

Fig. 2, folloning

situation of .qppendis

illustrated .! gives

\-I-(?.)

= 2&@(r-r,)-6(r-r_j]

(2, r(r) s q+ q sin! < :/3(. - 3 (1,) : = ~~hc /, *, -_i <, B . cash 37 _x __cI (l!! whew

\vhere i;(r) is the Dirac unity function. must satisfy the bountlar!, condikon: grad ,, C7(l.)il: = (!

(3)

X being the surface that encloses the geometric body formed by the sample and its mirror image reflected in the xOy plane. (See Fig. 1.) Let fk(r) be the normalized eigenfunctions and Xk the eigenvalues of the operator defined b! equations (2) and (3) :

11

b
\ _ -:

2(

.x

FE

-;

.s

(12)
5

It may that

be assumed,

without

loss (I 3 3s

of generality, (13)

Gyqr) = -hXlfk-(l.) fk satisfying

(4)

c 6 b < ci;

(3)

Taking into account these inequalities one can, by making a slight approximation, effect a partial summation of the infinite series (1 l)-see Appendix I<-obtaining the following result:

Del-eloping

the functions

7 and 6 in series of fk, (6) &_ s

(7)
(8)

The relative error introduced by using (14) instead of (11) is smaller than 0.19 per cent if 2c < a and b < a, and decreases rapidly with increasing a/c and a/b. In the case a < 2c < 2a,

FACTOR

IN SEMICONDUCTOR ----

FOUR-PROBE

RESISTIVITY

MEASUREMENTS

461

-----

which was obtained by the method of images and is suitable for computation when X, y, z are large (x, y, z > 4-5). The two formulae (11) and (17) are thus complementary. However, the series in (11) is double, while that in (17) is triple, also, the convergence of (11) is markedly better owing to the terms decreasing exponentially.
4. CONCLUSIONS

I(26 z

,O,@/

$-,u,o)

3i$
UF

,$Ul ,5,Ly

FIG. 2

the error is larger, but this case is seldom encountered in practice. The convergence of the infinite series in (14) becomes more and more rapid, and its sum decreases, when c/b decreases; in many practical cases c/b is negligible and then the first (logarithmic) terms in (14) give a good approximation to the exact result. In the particular case of a very thin sample (c < s), one obtains the result in closed form :

By the Green-function method, it is possible to obtain a general expression, in the form of a series of the geometric factor defined by formula (1). In the particular case of the parallelepiped sample, with linear configuration of the probes, formulae suitable for numerical computation can be derived. For a very thin rectangular sample, an approximate solution in closed form is proposed that approaches very closely the exact one. Other narticular cases can be studied in a similar manner, but increasing analytical difficulties are likely to be encountered.
APPENDIX A

The following eigenfunctions are easily obtained from (2) and (3) for the case of a parallelepiped sample:

f mm -N mpq sin

(2m+l)n
----x

IL

cos

p.27r

--y

. cos--x

q?T
&.I)

where X, 31, .s are indicated in Fig. 2 and should not be confused with the symbols defined in (12). The normalizing factor N7,ipphas the following values:

lx l-

exp[-+2a/b)] exp[-n(2n--4$)/b] Nmpq = (

The numerical results published for this case by SMITSW are easily verified (within the error mentioned above) by means of this formula. It is interesting to compare the formula (11) with another infinite series form of (L)-1, namely

(p = 0,

4 # 0 or
P # 0, 4 = 0) 64.2)

(P = 0,
The corresponding

P = 0)

eigenvalues are given by

xnzpq = nz[q;z+;+$]

(4.3)

x{[(py>2+(qZ)2+(mx- [(py)2-t (qZ)+

1)21-l/2 (17)

(WZX- 2)2]-12}

Although not all eigenfunctions of the problem have the form (A.l), symmetry allows us to ignore any other eigenfunction.

462 Inserting into (10) we get

A.

MIRCEA for the second and the fourth term in (B.2). For the third term, in the worst case (a = b = c) the relative error will be smaller than exp( - n) N 0.043. By means of (B.2) it is possible to obtain the sum of the first three terms in (B.l). The second term is written as follows:

(I;)-1

= ;;$ -m P

3, -02

z,, -a3

sin(2m+
X

l)~(s/2a) . sin(2m+

l)r(3s/2u) (A.4)

(2m + ly+py4&/b2)

-t q2(n/c2)

By transforming the product of sines into a difference of cosines, then making use of the identity

c
0 (L)-1

cc

cos (2m + 1)x

77 sin CL(T/~ -X)

\Ve apply to each of the four terms in (B.3) the formula

?n (2??2+1)2+22

4x

cash cr(77/2) (0 < s < n) (AS)

exp( - mx) 111 rrl

= ln[l-

exp( -x)]

(B.4)

and transforming again the difference of hyperbolic sines into a product of hyperbolic sine and cosine, we arrive at expression (11).
APPENDIX \l;e expand (11) in four terms: B

and obtain

-In m coshpn(s/b)(n/s-

(l{13)

exp[-77(4s/b)]}{lexp[-77(2s/b)]}(l-

exp[-17(2n-2s)/b].) exp[-7r(2n_4s)!b]j

= ;[1+2

cT, sinhpr(s/b) 1*

pr(s/b) . coshpm/b

m +2x
1

sinh qn(s/2c) . cash qm(n/s- 3) g qa(s/2c) - cash q+/2c)

+4

cc
1 plq approximations

to

cc

sinh r(s/b)[p2+ q?(b/2c)]l 7T(s/b)[p2 + q2(6/2c)2]12

* .

~s~~~~~~:~q~~~~~~~~~llz

(B. 1 )i

We make the following

coshpr i 21 z exp
1 E ?exp n i 1 - exp[ -77(2a/b)] -l (B.5)
in the same

coshqrz

i (

qrz b

;
1

x 1-

exp[-n(2a-4s)/b]

cash:-,p

+q

2 l/2 ) ]

The third term in (B.l) can be treated manner, finally obtaining (14). REFERENCES

1. L. B. VALDES, Transistor Technology Vol. I, Chap. 12.

x !xpir;ip:q2c:;Jil The relative error will be smaller than exp(-22n) 2: 0.0019

(B.2)

2. 3. 4. 5.

A. F. D. F.

Van Nostrand, New York (1958). UHLIR, Bell Syst. Tech. J. 34, 103 (195.5). SMITS, Bell Syst. Tech. J. 37, 711 (1958). E. VAUGHAN, Brit. J. Appl. Phys. 12,414 (1961). KEYWELL and J. DOROSHESKI, Rev. Sci. Instrum. 31, 833 (1960).