ALGORITHM FOR ACCURATE CAPACITANCE M ATRIX M EASUREMENTS

OF THE M ULTICONDUCTOR STRUCTURE FOR VLSI INTERCONNECTIONS

Lech ZNAMIROWSKI Olgierd A. PALUSINSKI

DEPARTMENT OF AUTOMATIC CONTROL, DEPARTMENT OF ELECTRICAL
ELECTRONICS AND INFORMATICS AND COMPUTER ENGINEERING
Silesian University of Technology University of Arizona
Ul. Akademicka 16 TUCSON, AZ 85721, USA
44-100 GLIWICE, POLAND e-mail: [email protected]
e-mail: [email protected] Tel.: (520) 621-4928
Tel.: +48 32 37-29-55 Fax: (520) 621-8076
Fax: +48 32 37-27-33

ABSTRACT

The demand for computer simulation of multiconductor interconnections in VLSI structures necessitates that
such structures should be modelled. Calculations of the model capacitance matrix from measurements, for the
multiconductor transmission lines which are coupled interconnections in a VLSI structures, using the "two-terminal"
capacitances indirect measurement procedure is strongly corrupted by the measuring errors. Sensitivity analysis and
error propagation relations demonstrates it.
The paper presents alternative method for direct capacitance measurements in the multiconductor structure using
active separation of the capacitance network to increase the measurements accuracy.

1. INTRODUCTION

The interconnections between parts of a system in high-speed VLSI silicon structures are working as strongly
coupled multiconductor transmission lines. The computer simulation of multiconductor interconnection structures
necessitates that such structures should be modelled. To find the parameters of that model, one may start by deter-
mining the capacitance matrix of the multiconductor system. The remaining parameters, such as inductance or line
losses, are determined depending on the type of propagated waves and physical phenomena accompanying high speed
transmission [1, 10, 12].
One approach to find the capacitance matrix as a per-unit-length (PUL) capacitance matrix, is based on geo-
metrical data and the electrostatic properties of multiconductors strip lines [4, 7, 9, 10]. The second approach is based
on measurements. The capacitance matrix can be determined indirectly basing on the laboratory data and calculations
[14] or directly basing on the algorithm (presented in this paper) applying active separation of the capacitance network
[6, 11] for accurate capacitance matrix measurements.
 -2-

Consider a structure with n signal conductors and a ground conductor (denoted by #0) with cross-section
presented in Fig. 1.

Fig. 1. The cross-section of n-conductors geometry

with reference

In order to determine the PUL "two-terminal" capacitance matrix T=[Tij], let us at first consider capacitance
coefficients matrix i.e. the Maxwell matrix C for multilayer nonhomogeneous dielectric and n conductors (and the
ground) imbedded. We have a basic equation for the matrix C =[Cij] definition [4]:

Q= C V , (1)
where:
QT = [Q1, Q2, ... , Qn] - is a PUL charges vector on the n-conductor transmission line,
T
V = [V1, V2, ... , Vn] - is the corresponding potentials vector.

The diagonal elements of the matrix C are called coefficients of capacitance and the off-diagonal elements are
called the coefficients of electrostatic induction. Properties of the matrix C are discussed in [4].
Analysing the potential differences between conductors and using "two-terminal" capacitances Tij, one finds [7]:

T ij = - C ij for i ≠ j ,
n

∑C .
(2)
Tii = ij
j =1

These "two-terminal" capacitances Tij have simple interpretation presented in Fig. 2.

The goal of presented algorithm is determining the matrix T entries from direct measurements.

Fig. 2. Capacitances equivalent circuits for n conductors

with reference
 -3-

Solutions of many problems such as some identification methods, the minimization of functionals, the summa-
tion of Fourier series with approximately specified coefficients, some techniques for semiconductor data extraction
calculations or complex laboratory methods for indirect parameter estimation, strongly depend on algorithm solving
the problem. In general case, it is possible to classify algorithms as ill-posed or correctly-posed depending on prop-
erties of the operator which is the fundamental of the algorithm basis according to Hadamard ill-posed operator
equation definition [3]. Important method of approximate solution of ill-posed problems are regularizing algorithms
[8].
In [14] was analyzed an error propagation in computing the Maxwell matrix C (the results of analysis are directly
valid for "two-terminal" capacitance model i.e. the matrix T computing) of multiconductor transmmision line on
surface of the wafer for VLSI interconnections using laboratory data. The measurements was realised indirectly for
per-unit-length interconnections with the 2-D cross-section. As a conclusion it was pick out a great sensitivity of the
results on measuring data especially for off-diagonal elements of the matrix model. Calculations of the capacitance
model for multiconductor transmission line interconnections using laboratory data was simple solved, basing on
measurements/calculations but the great sensitivity on the measuring data was observed (ill-posed problem). Using
measurements to finding Maxwell or "two-terminal" matrices' elements requires extremely high accuracy of instru-
ment and very carefully constructed measurement station for obtaining the acceptable results. In this paper is
presented alternative algorithm for extract capacitances Tij, applying direct measurement method using instrumental
unity-gain amplifier for n-conductor interconnections structure and a ground conductor, which fulfills active separ-
ation of the capacitance network to increase the measurements accuracy. Results of this measurements can be used in
different areas of VLSI interconnections characterization [7, 13] and simulation [1, 2, 5, 12]. The Maxwell matrix C
can be determined using Eq. (9) (see § 3.2), basing on accurate measurements of the T matrix.

2. ALGORITHM

2.1. Notation
For convenience of the algorithm description (the conductors numbered from #1 to #n, and ground conductor
denoted by #0) we'll use for element Tij the following notation:
a) for i=j:
T jj ≡ T0 j
and j=1, 2, ..... n.

b) for i ≠ j :
the indexes of the Tij fulfill the inequality i<j (see § 3.2)

and i=0, 1, ..... n-1,

j=1, 2, ..... n.

2.2. Unity-gain Voltage Amplifier

The Unity-Gain voltage Amplifier (UGA) with input pin IN, output pin OUT and a common point G, is
presented in Fig. 3. The voltage gain of the UGA is equal unity, so the input voltage VIN=VOUT with very high input
 -4-

impedance of the UGA and very low output impedance. i and j are numbers of the conductors of the multiconductor
structure connected respectively to the pins G and IN of the UGA. Simultaneously, the G and IN pins are connected to
the input of the capacitance meter (pins B and A in Fig. 4). The UGA always fulfills active separation because the pins
IN and OUT have the same potentials, and a part of a circuit connected to the pins G and OUT is charged from the
output of the UGA.

Fig. 3. Unity-gain voltage amplifier

2.3. Measurement Algorithm

The algorithm for the measurements of the "two-terminal" Tij capacitances values with active separation which
fulfills condition separately measured value in a multi-conductors system with ground conductor, has a following
form:

Algorithm MATRIX-T-MEASUREMENT
input: n;
output: matrix T;
begin
for i=0 to n-1 do
for j=1 to n do
if i<j then
connect i-th conductor to the G pin of the UGA;
connect j-th conductor to the IN pin of the UGA;
finally, shunt together all remaining conductors and connect them to the OUT pin of the UGA;
measure Tij;
using notation (§ 2.1), find suitable entry of the matrix T;
end.

Let us use the graph representation of the muticonductor structure (Fig. 4, 5 and 6). The nodes of the graph
represent the conductors with a ground conductor and branches (bolded lines), represent the sufficient capacitances.
In a 2-conductors and ground conductor system (Fig. 4) in a case of measuring the T12 capacitance, the conductor #2
connected to the IN pin of the UGA and the ground conductor #0 connected to the OUT pin of the UGA guarantees,
that capacitor T22 has electrodes at the same potentials and is not charged. The T11 capacitor is charged from amplifier
and in consequency, the measured capacitance T12 is separated from the structure. The inputs of the capacitance meter
(B and A) "sees" only isolated capacitance T12.
Similar properties have the measuring structures presented in Fig. 5 and Fig. 6, for exemplary 3-conductors and
ground structure.
In general case, for a n-conductors and ground system, realising the steps of the algorithm MATRIX- T-MEA-
 -5-

SUREMENT we get for Tij capacitance measurements, the equivalent circuit presented in Fig. 7.

Fig. 4. Graph representation of T12 capacitance mea-

surement in a 2-conductors and ground system with
active separation

Fig. 5. Graph representation of T12 capacitance mea-

surement in a 3-conductors and ground system Fig. 6. Graph representation of T11 (T01) capacitance
measurement in a 3-conductors and ground system

The equivalent capacitance T2eq is not charged, and equivalent capacitance T1eq is charged from the amplifier. The
Tij capacitance is separated from the circuit.

3. ACCURACY OF THE MEASUREMENTS

3.1. Accuracy of the T Matrix Measurements

Let us assume that the relative error of a capacitance meter (compensated, full scale) is defined as δ . The
measurements of the values Tij entries of the matrix T when the range if instrument is properly chosen, will be nearly
the value δ, because the measurements are performed directly.
 -6-

Fig. 7. Equivalent circuit for Tij capacitance

measurement with active separation in a n-
conductors and ground system

3.2. Accuracy of the C Matrix Determined from the Direct Measurement of the Matrix T
From the properties of matrix C and Eq. (2) we have:

C ij = C ji (3)
and

Tij = T ji . (4)

Thanks to the symmetry we need to consider (n2-n)/2+n relations instead of n2 relations.

To simplify the notation we use the following vector definitions

C T = [C11,C12,C13,C14,... C1n,C22,C23,C24,... C2n,C33,C34,C35,... C3n,C44,... Cnn]

= [C1,C2,C3,... CN] (5)

and
T T = [T11,T12,T13,T14,... T1n,T22,T23,T24,... T2n,T33,T34,T35,... T3n,T44,... Tnn]
= [T1,T2,T3,... TN]. (6)

Vectors C and T have N=(n2-n)/2+n=n(n+1)/2 entries, symbol T denotes transposition.

Using relation (2) and symmetry that Tij=Tji, one may write:
n
T ij = ( δ ij - 1 ) C ij + δ ij ∑A
k =1
ik (7)

for i = 1, 2, ... , n
j = 1, 2, ... , n
and i ≤ j,
where
1 for i = j
äij =
0 for i ≠ j,

Cik for i ≤ k
Aik =
Cki for i > k.
 -7-

Using relations (5) and (6) we have:

T = DC (8)

where matrix D is given by (7).

Relation between the "two-terminal" vector T and the vector C representing elements of Maxwell matrix can be
found using Eq. (8):

C = D -1T = GT (9)

where
G=D -1 - inverse matrix of D (N×N).

Elements of the matrix C are represented by vector C according with the relation (5).

We can find [14] the absolute error in Ci calculation (Ci is i-th coordinate of the vector C):
N
∆ Ci = ∑| G
j=1
ij | | ∆ T j | (10)
where
ÄTj - absolute error at measurement of the j-th element of the vector T.
The relative error for coordinates of vector C has the form:

∆ Ci
δ Ci = . (11)
|C i|

From (9) one finds:

Ci = Gi T (12)

where

Gi - i-th row of the matrix G.

Then using (10) and (11) we have:

N
∆Ci
∑| G
1
δ Ci = = ij || ∆ T j |. (13)
|C i| | G i T | j=1

Using expression ∆T j
δTj =
|T j |

for relative error of the j-th element of vector T, at last we have

N

∑|G
1
δ Ci = i j| | äT j | | T j | . (14)
| GiT | j =1

Relation (14) describes relative error in Ci element calculation for given measurements of Tj elements of vector T and
their relative errors of measurements.
 -8-

4. EXAMPLE I

For a 4-conductors and ground system, connections of the active separation in the measurement circuit are
presented in Figs. 8 - 17, accordingly with the algorithm MATRIX- T-MEASUREMENT.

Fig. 8. T11 (T01) measurement in a 4-conductors and

ground system Fig. 9. T22 (T02) measurement

Fig. 11. T44 (T04) measurement

Fig. 10. T33 (T03) measurement

 -9-

Fig. 12. T12 measurement Fig. 13. T13 measurement

Fig. 14. T14 measurement

 - 10 -

Fig. 15. T23 measurement Fig. 16. T24 measurement

Fig. 17. T34 measurement

5. EXAMPLE II

For a 6-conductor and ground system, the connections of the UGA fulfilling active separation during the
measurements T matrix of multiconductor structure, are presented in Table 1.
 - 11 -

Table 1
Connections of the UGA pins to the conductors
for a 6-conductors and ground system

Connections of the UGA pins to the conductors

Measured capacitances G IN OUT

T11 (T01) #0 #1 #2, #3, #4, #5, #6

T22 (T02) #0 #2 #1, #3, #4, #5, #6
T33 (T03) #0 #3 #1, #2, #4, #5, #6
T44 (T04) #0 #4 #1, #2, #3, #5, #6
T55 (T05) #0 #5 #1, #2, #3, #4, #6
T66 (T06) #0 #6 #1, #2, #3, #4, #5
T12 #1 #2 #0, #3, #4, #5, #6
T13 #1 #3 #0, #2, #4, #5, #6
T14 #1 #4 #0, #2, #3, #5, #6
T15 #1 #5 #0, #2, #3, #4, #6
T16 #1 #6 #0, #2, #3, #4, #5
T23 #2 #3 #0, #1, #4, #5, #6
T24 #2 #4 #0, #1, #3, #5, #6
T25 #2 #5 #0, #1, #3, #4, #6
T26 #2 #6 #0, #1, #3, #4, #5
T34 #3 #4 #0, #1, #2, #5, #6
T35 #3 #5 #0, #1, #2, #4, #6
T36 #3 #6 #0. #1, #2, #4, #5
T45 #4 #5 #0, #1, #2, #3, #6
T46 #4 #6 #0, #1, #2, #3, #5
T56 #5 #6 #0, #1, #2, #3, #4

6. CONCLUSIONS

The algorithm for accurate "two-terminal" matrix measurements (per-unit-length) of the multiconductor trans-
mission lines for VLSI interconnections model has been presented. The algorithm uses active separation approach
significantly increasing the accuracy of measurements in contrary to indirect determining the capacitance matrix.
Furthermore, the Maxwell matrix computing from T matrix can be determined with significantly smaller error
propagation than in the case of the indirect measurements of the matrix T entries.
The presented algorithm depending on principle of measurement performed with the instrument connected to the
pins B and A (higher frequency signal, pulse excitation), can be applicated for extended characterization of the
 - 12 -

multiconductor transmission lines in VLSI interconnections modeling. Important application in this area is accurate
measurement such phenomena as dielectric absorption [12] which plays significant role in high speed structures and
thermal effects.

7. REFERENCES

[1] F. Y. Chang, "Transient Analysis of Lossless Coupled Transmission Lines in a Nonhomogeneous Dielectric Medi-
um," IEEE Trans. on Microwave Theory and Techniques, Vol. MTT-18, No. 9, September 1970, pp. 616-626.
[2] A. R. Djordjevic, T. K. Sarkar, R. F. Harrington, "Time-Domain Response of Multiconductor Transmission Lines,"
Proc. of the IEEE, Vol. 75, No. 6, June 1987, pp. 743-764.
[3] J. Hadamard, Le problème de Cauchy et les équations aux dérivée partielles lineaires hyperboliques,
Hermann, Paris 1932.
[4] D. W. Kammler, "Calculation of Characteristic Admittances and Coupling Coefficients for Strip Transmission
Lines," IEEE Trans. on Microwave Theory and Techniques, Vol. MTT-16, No. 11, November 1968, pp. 925-
937.
[5] O. A. Palusinski, J. C. Liao, J. L. Prince, A. C. Cangelaris, "Simulation of Transients in VLSI Packaging Intercon-
nections," IEEE Trans. on Comp., Hybrid, and Man. Technology, Vol. 13, No. 1, March 1990, pp. 160-166.
[6] K. Reiss, "Accurate Measurement of Small Coupling Capacitance under Influence of Large Ground Capacitances,"
(Private communication), University of Karlsruhe, Karlsruhe 1986.
[7] A. E. Ruehli (Ed.), Circuit Analysis, Simulation and Design. Advances in CAD for VLSI, Vol. 3, Part 2, North-
Holland, Amsterdam 1987.
[8] A. N. Tikhonov, A. V. Goncharsky (Eds.), Ill-Posed Problems in the Natural Sciences, Mir Publishers, Moscow
1987.
[9] W. T. Weeks, "Calculation of Coefficients of Capacitance of Multiconductor Transmission Lines in the Presence
of a Dielectric Interface," IEEE Trans. on Microwave Theory and Techn., Vol. MTT-18, No. 1, January 1970, pp.
35-43.
[10] C. Wei, R. F. Harrington, J. R. Mautz, T. K. Sarkar, "Multiconductor Transmission Lines in Multilayered Dielec-
tric Media," IEEE Trans. on Microwave Theory and Techniques, Vol. 32, No. 4, April 1984, pp. 439-450.
[11] R. Zielonko, A. Krolikowski, Diagnostic-Measurement Methods of Analog Electronic Circuits (in Polish),
Wydawnictwa Naukowo-Techniczne, Warsaw 1988.
[12] L. Znamirowski, "Dielectric Absorption Model Identification for Microstrip Lines in VLSI Structures," Mixed
Design of Integrated Circuits and Systems - Proceedings of the 4-th Int. Workshop, MIXDES'97, 12-14 June,
Poznan 1997.
[13] L. Znamirowski, "Scaling of Coupled Multiconductor Interconnections in VLSI/ULSI Structures," Mixed Design
of Integrated Circuits and Systems - Proceedings of the 5-th Int. Conference, MIXDES'98, 18-20 June, Lodz
1998.
[14] L. Znamirowski, O. A. Palusinski, "Analysis of Error Propagation in Computing the Maxwell Matrix of
Multiconductor Structure from Measurements," IEE Proceedings - Science, Measurement and Technology,
Vol. 145, No. 3, May 1998.