Professional Documents
Culture Documents
Improved Estimation of The Resistivity of Pure Copper and Electrical Determination of Thin Copper ®LM Dimensions
Improved Estimation of The Resistivity of Pure Copper and Electrical Determination of Thin Copper ®LM Dimensions
www.elsevier.com/locate/microrel
Abstract
Improved values for the resistivity, q, of pure, bulk copper from 50 to 1200 K, and their con®dence intervals, are
developed by extending the analysis of Matula. A recommended value for dq=dT and its con®dence interval in the
temperature range of 290±425 K is developed for use with Matthiessen's rule to calculate the electrical thickness of thin
copper ®lms and the cross-sectional area of copper lines from resistance measurements at two temperatures. Error
analyses are used to estimate the uncertainty with which the electrical thickness and cross-sectional area can be de-
termined. Values for the temperature coecient of resistance of pure, bulk copper are also provided. Ó 2001 Elsevier
Science Ltd. All rights reserved.
The organization of the paper is as follows: Given a mean value for dqPB =dT for copper in a
certain temperature range, the thickness of a copper ®lm
1. A method is reviewed to show how the electrical and the cross-sectional area of a copper line can be
thickness of copper ®lms and the cross-sectional area calculated, respectively, from measurements of the sheet
of copper interconnect lines can be calculated from resistance and the line resistance at two temperatures
resistance measurements taken at two temperatures. within this range. Neglecting the eect of thermal ex-
These calculations require knowledge of the change pansion, ®lm thickness can be calculated from the
in resistivity of pure copper with temperature, dq= change in sheet resistance, RS , with temperature from
dT , which is approximately constant over the temper- 1
ature range where such resistance measurements can dq dRS
t PB :
3
be conveniently made. dT dT
2. Improved recommended values for the resistivity of
The sheet resistance can be determined from the re-
copper from 50 to 1200 K and their con®dence inter-
sistance of a van der Pauw cross test structure [5]. The
vals are developed and tabulated by using the exten-
cross-sectional area, A, of a line of length L can be
sive compilation of Matula [2] and extending his
calculated from a measurement of the change in resis-
analysis of the resistivity of very dilute alloy speci-
tance of the line, R, with temperature from
mens of bulk copper.
3. A recommended value for dq=dT and its 95% con- 1
dq dR
®dence interval for the temperature range of 290± A L PB :
4
dT dT
425 K (17±152°C) is developed from these improved
values for resistivity. This is the temperature range But Matthiessen's rule is not obeyed exactly. Devia-
within which it is recommended that electrical de- tions from Matthiessen's rule have been reported for a
terminations of the dimensions of copper lines be variety of metals, including copper [6]. Such deviations
made. impact the above two equations by the multiplicative
4. From the values of q
T and dq=dT for pure copper, factor
1 , where
dq=dT =
dqPB =dT . The pa-
the temperature coecient of resistance, TCR
T , for rameter is a correction factor that indicates how much
two reference temperatures are provided for use in the estimates of the metal ®lm thickness or the cross-
making qualitative assessments of the purity and de- sectional area is altered by a deviation from Matthies-
fect density of a copper specimen [4]. sen's rule.
5. And, estimates are made of the precision with which Deviations are often displayed in the literature by
the electrical thickness of ®lms and the electrical plotting the ratio of D
c; T =qr
c with temperature for
cross-sectional area of lines can be measured within dierent small concentrations, c, of an additive metal.
a given laboratory (repeatability). The residual resistivity, qr
c, serves as a scaling factor
which removes the dependence of the ratio on c and
permits a comparison of D
c; T for dierent impurities
2. Electrical measurement of interconnect dimensions in the host metal [6]. For increasing temperatures to
room temperature and beyond, the curves for dierent
A generic expression for the resistivity of a dilute dilute copper alloys of a given additive tend to converge.
alloy of a metal at temperature T may be given by Based on the data for separate additives (from 0.05 to
1.0 at.%) of gold, germanium, nickel, and tin [6±8], the
q
T qPB
T qr
c D
c; T ;
1 values for the ratio D
c; T =qr
c, which we will refer to
as c, fall within the range of approximately 0.04±0.13.
where qPB
T is the resistivity of the pure metal (also Values for the slope of the change of the ratio with
referred to later as intrinsic resistivity), qr
c the residual temperature, S
1=qr
cdD
c; T =dT , fall approxi-
resistivity of the metal, which is dependent only on the mately within the range of 0.0001 to 0.0007 °C 1 .
impurity and defect concentration in the metal, and One can express in terms of the slope, S, and the
D
c; T a residual resistivity that is dependent on tem- ratio c by starting with an expression for the tempera-
perature as well as on the impurity and defect concen- ture coecient of resistance:
tration in the metal.
1 dq
If Matthiessen's rule holds exactly, the third term is TCR
T
zero and the result is that the change in the resistivity q dT
with temperature of the pure metal, qPB
T , is equal to 1 dqPB dD
qPB
T qr
c D
c; T :
that for the metal alloy, q
T , namely that: dT dT
5
dqPB dq
:
2
dT dT Noting that
C.E. Schuster et al. / Microelectronics Reliability 41 (2001) 239±252 241
Table 1
Resistivity values of data sets 35, 36, and 124 from 50 to 1200 K
T (K) q (lX cm) Data set (#) T (K) q (lX cm) Data set (#)
50 0.0528 124 280.5 1.594 36
60 0.0978 124 285.3 1.626 36
70 0.156 124 290.3 1.663 36
80 0.218 124 290.9 1.664 36
85 0.248 35 290.9 1.667 36
90 0.282 35 292.1 1.67 36
90 0.283 124 292.4 1.676 36
100 0.35 35 293.1 1.682 36
100 0.353 124 295 1.7 124
110 0.418 35 300 1.725 35
120 0.488 35 309.2 1.787 36
120 0.493 124 325 1.893 35
130 0.558 35 325 1.892 36
140 0.631 35 331.5 1.936 36
140 0.638 124 342 2.005 36
150 0.702 35 350 2.062 35
160 0.778 124 352.6 2.075 36
175 0.876 35 375 2.229 35
180 0.923 124 381.6 2.269 36
200 1.047 35 382.2 2.275 36
200 1.06 124 433.1 2.617 36
220 1.2 124 482.4 2.955 36
225 1.219 35 666.4 4.224 36
250 1.389 35 678 4.324 36
250 1.4 124 768.8 4.98 36
273 1.55 124 865.1 5.692 36
273.16 1.546 35 972.7 6.525 36
275 1.556 35 1066.7 7.309 36
278.4 1.578 36 1066.9 7.268 36
279.3 1.586 36 1165.2 8.16 36
280.1 1.593 36
original Bloch±Gr uneisen form. Also, the replacement Values for the resistivity of liquid copper from 1357.6 to
of the Debye temperature, h, with an eective Debye 1700 K are also provided by Matula [2].
temperature,
h CT , permits the representation of
changes in the eective Debye temperature with tem- 3.3. Extension of Matula's approach
perature [2].
In arriving at the initial ®t to the data, Matula re- 3.3.1. Initial attempts and analysis
ported the following values for the coecients of A least-squares ®t of the G2
T form to the experi-
G2
T : A 1:8089 10 8 X m, B 5:9991 10 3 , mental data led to resistivity values essentially equiva-
C 0:04563, D 6:476 10 4 , h 310:8 K, and lent to those obtained from the initial ®t reported by
p 1:84. To obtain a ®nal ®t to the data, Matula re- Matula with the above mentioned parameter values.
ported smoothing a plot of the fractional deviations But, the values for the parameters were dierent. An
versus temperature and adding the result to the initial ®t examination of the ®tted G2
T curve of Matula showed
to obtain smoothed resistivity data from 50 to 1200 K. A systematic deviations from the experimental data at
correction for thermal expansion was then applied to the both ends of the temperature range. The deviations are
smoothed data. He provided recommended values for larger at the low temperatures than they are at the high
the intrinsic resistivity, qi
T , of copper at 26 tempera- temperatures. This is shown in Fig. 1 where the dashed
tures from 50 to 1200 K. To obtain values for the re- line is the initial ®tted curve of Matula (before smooth-
sistivity at temperatures between those for which the ing) and the dots are the experimental data from the
recommended values were provided, Matula suggested three data sets.
using a linear interpolation of loge qi
T versus loge T . These deviations are explainable by noting that a
Additional resistivity values are provided by Matula [2] least-squares analysis of the resistivity data gives equal
for temperatures down to 20 K and up to 1357.6 K. weight to all dierences between the resistivity data and
C.E. Schuster et al. / Microelectronics Reliability 41 (2001) 239±252 243
the ®tted curve. At even lower temperatures, these dif- 3.3.2. Final ®t of experimental data to G2
T form
ferences become smaller as the values of resistivity ap- A ®nal ®t to G2
T was conducted in the iterative
proach and are bounded by zero. But, the relative manner described above except that, like Matula, the
dierences can be substantial. At high temperatures, the experimental data (sets 35, 36, and 124) were ®rst con-
dierences tend to be larger. Hence, the ®t is dominated verted to intrinsic resistivity by subtracting the reported
by the dierences at the higher temperatures. residual resistivity from the data in each set. This pro-
To improve this situation, a least-squares ®t of the cedure led directly to a satisfactory ®t to the data over
loge of the resistivity was performed over the tempera- the entire temperature range given by G2
T NIST . The
ture range of 50±1200 K. The parameters of the G2
T best parameter estimates obtained for G2
T NIST are:
form were varied to minimize QL , where: 8
A 1:816013 10 X m;
Xh i2 B 2:404851 10 ; 3
QL loge q
T j loge
qR G2
T j :
10
j C 4:560643 10 2 ;
D 5:976831 10 3 ;
In this case, equal weight is given to the fractional dif- p 1:838419;
ferences between the resistivity data, q
T j , and the ®tted
curve. A constant was added to G2
T to represent the h 310:8 K;
presence of a residual resistivity, qR , in the values of the qR 0:00018 lX cm:
three data sets. This permitted the direct use of the total
resistivity values in the three data sets without making The small value for the parameter qR constitutes the
any assumptions about the residual resistivity. residue of the residual resistivity in the ®tting process.
To achieve convergence, the following ®tting proce- The G2
T NIST is plotted in Fig. 1 where it can be com-
dure was used. The starting value for qR was zero and pared with the initial ®t made by Matula and the ex-
the value for the Debye temperature, h, was kept con- perimental data.
stant and equal to the value used by Matula, i.e. 310.8 A conventional measure for the variability of the
K. The nonlinear parameters, C and p, were initially experimental data about a ®tted curve is the sum of the
held constant at the values reported by Matula and the squared deviations (SSD). This measure was used to
linear parameters, A, B, D, and qR , were allowed to evaluate the quality of the ®t of G2
T NIST to the ex-
vary. Then, C and p were permitted to vary while pa- perimental data, q
T j (data sets 35, 36, and 124), where
rameters A, B, D, and qR were held ®xed at their es- Xh i2
tablished values. This process was repeated by using the SSD loge q
T j loge
G2
T NIST j :
11
newly arrived at values for C and p until convergence j
was achieved.
An analysis of the results of this ®tting procedure The value obtained for SSD was 0.00123. The root mean
revealed that the standard error for the ®t is small, while square, RMS, of the deviations is given by the square
the standard errors for the parameters are very large. It root of SSD divided by the degrees of freedom, which
was gratifying to ®nd that the estimated value for the are 55 (number of data values, 61, minus the number of
244 C.E. Schuster et al. / Microelectronics Reliability 41 (2001) 239±252
parameters, 6). The root mean square value for the vals for the resistivity as a function of temperature were
NIST data is 0.00473. calculated in the way described in Appendix A.1. Be-
By way of comparison, the same measure (SSD) was tween 200 and 1200 K, the half-width of the 95% con-
used for Matula's recommended values for resistivity. ®dence interval of the ®t of G2
T NIST is less than 0.2%.
Using the recommended resistivity values of Matula to Below 200 K, the half-width increases steeply with de-
obtain log-interpolated values of resistivity at the tem- creasing temperature to a value of approximately 14% at
peratures of the experimental data, the SSD for the ®t to 50 K.
the experimental data was found to be 0.08477. The root The values provided in Table 2 are an improvement
mean square of the deviations is therefore 0.0393. This over those provided by Matula [2] for a number of
value is over eight times as large as the RMS value ob- reasons: (1) the value for the resistivity can be evaluated
tained for the ®t of G2
T NIST to the experimental data. at any temperature and the procedure for obtaining
these values can be duplicated by the reader, if desired;
3.3.3. Correction of G2
T NIST for thermal expansion (2) these values can be used to develop estimates for
The ®nal step was to make the small correction for dq=dT and their con®dence intervals, as discussed next;
thermal expansion of the G2
T NIST values, which was and (3) the recommended values provide an improved ®t
done by using to the experimental data, on the basis that the root mean
square of the dierences between the predicted and the
DL
qcorr:
T quncorr:
T 1 ;
12 experimental values for resistivity is signi®cantly smaller
L
T0 than that when using the resistivity values and the in-
where T0 293 K; DL L
T L
T0 , and it is as- terpolation procedure recommended by Matula.
sumed that the volume thermal expansion coecient is
equal to three times the linear thermal expansion coef-
®cient. Recommended values for DL=L
T0 , in percent,
were obtained from Table 12R in Ref. [14] where: 5. dq
T=dT of copper and its con®dence interval
Table 2
Resistivity
lX cm of pure, bulk copper and its one-standard-deviation con®dence interval from 50 to 1200 K. To obtain values for the
resistivity at a temperatures not given, use a linear interpolation of loge q
T versus loge T
T (K) q
T (lX cm) s.d. (lX cm) T (K) q
T (lX cm) s.d. (lX cm)
50 0.049986 0.00385 310 1.790548 0.000868
60 0.094907 0.00266 320 1.857888 0.000882
70 0.150515 0.002237 330 1.925271 0.000894
80 0.213125 0.001684 340 1.992711 0.000904
90 0.279940 0.001359 350 2.060219 0.000914
100 0.349075 0.001269 360 2.127807 0.000922
110 0.419366 0.001283 370 2.195486 0.000931
120 0.490088 0.001304 380 2.263264 0.000939
130 0.560828 0.0013 390 2.331152 0.000948
140 0.631358 0.001267 400 2.399157 0.000958
150 0.701565 0.001211 410 2.467287 0.000969
160 0.771403 0.001141 420 2.535549 0.000982
170 0.840866 0.001067 430 2.603949 0.000997
180 0.909970 0.000994 440 2.672494 0.001015
190 0.978746 0.000928 450 2.741189 0.001036
200 1.047228 0.000872 460 2.810040 0.00106
210 1.115452 0.000829 470 2.879053 0.001086
220 1.183454 0.0008 480 2.948231 0.001116
230 1.251271 0.000784 490 3.017579 0.001149
240 1.318935 0.000779 500 3.087102 0.001184
250 1.386475 0.000782 510 3.156804 0.001222
260 1.453922 0.000792 520 3.226688 0.001262
270 1.521299 0.000806 530 3.296759 0.001304
273 1.541502 0.000810 540 3.367019 0.001349
280 1.588630 0.000821 550 3.437473 0.001394
290 1.655936 0.000837 560 3.508123 0.001441
293 1.676126 0.000842 570 3.578973 0.001489
298 1.709776 0.000850 580 3.650025 0.001538
300 1.723236 0.000853 590 3.721283 0.001587
307 1.770352 0.000864 600 3.792749 0.001636
610 3.864427 0.001684 910 6.124288 0.002209
620 3.936318 0.001733 920 6.203606 0.002187
630 4.008426 0.00178 930 6.283202 0.002164
640 4.080752 0.001827 940 6.363078 0.002141
650 4.153300 0.001873 950 6.443236 0.002119
660 4.226071 0.001917 960 6.523677 0.002099
670 4.299068 0.00196 970 6.604404 0.002081
680 4.372294 0.002001 980 6.685418 0.002068
690 4.445750 0.00204 990 6.766721 0.00206
700 4.519438 0.002077 1000 6.848315 0.002059
710 4.593362 0.002112 1010 6.930202 0.002066
720 4.667523 0.002144 1020 7.012384 0.002084
730 4.741923 0.002174 1030 7.094863 0.002112
740 4.816564 0.002201 1040 7.177640 0.002154
750 4.891448 0.002226 1050 7.260718 0.00221
760 4.966578 0.002248 1060 7.344098 0.002281
770 5.041955 0.002266 1070 7.427783 0.002368
780 5.117581 0.002282 1080 7.511775 0.002471
790 5.193458 0.002294 1090 7.596074 0.002592
800 5.269589 0.002304 1100 7.680684 0.002729
810 5.345975 0.00231 1110 7.765607 0.002883
820 5.422618 0.002313 1120 7.850843 0.003055
830 5.499520 0.002313 1130 7.936396 0.003242
840 5.576683 0.002309 1140 8.022267 0.003446
850 5.654108 0.002303 1150 8.108458 0.003666
860 5.731799 0.002293 1160 8.194971 0.003902
(continued on next page)
246 C.E. Schuster et al. / Microelectronics Reliability 41 (2001) 239±252
Table 2 (continued )
T (K) q
T (lX cm) s.d. (lX cm) T (K) q
T (lX cm) s.d. (lX cm)
870 5.809756 0.002281 1170 8.281809 0.004153
880 5.887981 0.002266 1180 8.368973 0.00442
890 5.966477 0.002249 1190 8.456465 0.004702
900 6.045245 0.00223 1200 8.544288 0.004999
Fig. 2. dq=dT as a function of temperature for pure, bulk copper and its 95% con®dence interval (- - -).
dq 1 (TCR) for pure, bulk copper for 0°C and 25°C, with
0:006747 0:000075 lX cm K
1:1%:
dT their 95% con®dence intervals:
7. Uncertainty of ®lm thickness and line-area estimates line of known length, L, made at two temperatures. The
area, A, is given by
7.1. Introduction
1
dq R1 R0
Estimates of the mean electrical thickness of a thin AL :
19
dT T1 T0
copper ®lm and the mean electrical cross-sectional area
of a copper conductor can be obtained from measure- The relative uncertainty in the measurement of area is
ments of resistance of special test structures at two dierent from Eq. 18 only in the addition of the term
temperatures and the use of the value of dq=dT devel-
u
L=L2 under the square root.
oped in this paper. The structures considered here are The analysis used to obtain the value of 0.0055 for
those that are formed in processes involved in fabricat- u
q0 =q0 is described in Appendix A.2 of the Appendix A
ing semiconductor integrated circuits. The temperatures (see also Section 5). Estimates for u
R=DR were ob-
at which these resistance measurements are made are tained from the results of earlier within-laboratory
assumed to be within the range of room temperature to measurements to establish the repeatability of sheet re-
approximately 150°C. It is within this range that the sistance and line resistance measurements of aluminum
mean value for dq=dT and its uncertainty were calcu- alloy structures [16]. The analysis, discussed in detail in
lated. The more widely apart the measurement temper- Appendix A.2, revealed that the relative uncertainty for
atures are, within this temperature range, the less impact the sheet and line resistances are 0.0023 and 0.0033,
an error in estimating the temperature will have in es- respectively. Assuming an error of 0.5 lm in a line 250
timating the dimensions of the copper structure. lm long, the relative uncertainty contribution from L is
0.0020. Potentially the largest contributor to the uncer-
7.2. Thickness and area calculations tainty, if care is not taken, is the measurement of tem-
perature. An error of 0.5°C was assumed for each
An estimate of the mean electrical thickness of a thin temperature measurement, 130°C apart. This led to a
copper ®lm can be obtained from measurements of the relative uncertainty of 0.0038. The total relative uncer-
sheet resistance made at two temperatures. The value for tainties for ®lm thickness and line cross-sectional area
the sheet resistance of a copper ®lm can be obtained are estimated to be 0.84% and 0.98%, respectively.
from measuring the resistance of a van der Pauw There are two additional sources for uncertainty.
structure [5] fabricated from the ®lm of interest and One is a deviation from Matthiessen's rule of the me-
multiplying that value by p= loge 2 to obtain the sheet tallization being measured. As discussed in Section 2, an
resistance. uncertainty of no more than about 2 % is estimated. The
Because the sheet resistance, RS , is the ratio of the other is the high-resistivity shunt material used to encase
resistivity to the ®lm thickness, the mean electrical copper interconnects [17]. This layer is used to prevent
thickness of a copper ®lm, t, is given by copper from diusing out from the interconnect, to
1 promote adhesion of the copper to the inter-layer di-
dq RS1 RS0
t ;
17 electric, or both. Generally, the thickness of this layer is
dT T1 T0 much smaller than the dimensions of the copper inter-
where RS1 and RS0 are the resistances at temperatures T1 connect. Also, the resistivity of the layer material is apt
and T0 , respectively, and it is recognized that dq=dT is to be much greater than that of the copper. Both con-
very nearly constant over the temperature range. From ditions make it less critical to know exactly the ratio of
the theory of propagation of errors [15], the relative the thicknesses and resistivities in correcting for the
uncertainty of the measurement of thickness, u
t, is shunting eect in the resistance measurements involved
given as follows, where the factor 2 takes into account in the method [10].
the errors made in the measurements of resistance at the
two temperatures and where q0 is dq=dT :
v
"
u
u 2 2 2 # 8. Summary
u
t t u
q 0 u
T u
R
2 :
18
t q0 DT DR Values for the resistivity of pure, bulk copper from 50
to 1200 K were developed by extending and improving
An expression similar to Eq. (18) serves to estimate on the analysis of Matula [2]. These values provide an
the relative uncertainty in measuring the cross-sectional overall ®t to the experimental data that is superior to the
area of a copper line because the measurement proce- ®t of Matula. The half-width of the 95% con®dence in-
dure is the same. An estimate of the mean electrical terval for the ®t is less than 0.2% from 200 to 1200 K.
cross-section area of an interconnect line is obtained Below 200 K, the con®dence interval increases rapidly
from a four-terminal measurement of the resistance of a to 14%.
248 C.E. Schuster et al. / Microelectronics Reliability 41 (2001) 239±252
The method described here to measure copper ®lm where A 1:816013 lX cm, B 2:404851 10 3 , C
dimensions from resistance measurements at two tem- 4:560643 10 2 , D 5:976831 10 3 , h 310:8 K,
peratures assumes that Matthiessen's rule holds for p 1:838419, qR 1:803752 10 4 lX cm,
copper. It also requires a value for dq=dT for pure, bulk Z x
4 z5 ez
copper. Using the temperature dependence developed U
x 5 dz;
for the resistivity of pure copper, the mean value rec- x 0
ez 12
ommended for dq=dT between 290 and 425 K and is
and
0:006747 0:000037 lX cm K 1 , where one standard
deviation con®dence limits are provided. The relative h CT
x :
uncertainty is 0.55%. T
Including estimated errors in the measurement of
resistance, length, and temperature led to total relative The left hand side of Eq. (A.1) is referred to as
uncertainties in the measurement of ®lm thickness and G2
T NIST in Section 3.3.2. To simplify notation, q
T
line cross-sectional of 0.84% and 0.98%, respectively. (or simply q) will be used for this quantity in the ap-
Data in the literature indicate that deviation from pendix. The derivative of q with respect to temperature
Matthiessen's rule for dilute alloys of copper may con- will be denoted here as q0
T , or q0 .In terms of x, Eq.
tribute an error of less than approximately 2% to the (A.1) can be written as
method. An error analysis of all these sources indicated q
T qR A
1 B=x Dxp U
x:
that, in combination, the total relative uncertainty for
both ®lm thickness and line cross-sectional area are The parameters in Eq. (A.1) cannot be estimated si-
approximately the same and equal to 2.2%. Errors in the multaneously by least squares. This is an indication that
temperature measurements can be dicult to control this function is overparametrized. As a consequence, one
and can lead to a signi®cantly larger uncertainty than should not attempt to assign meaning to the values of
given here. the parameters because, with the exception of A and h
From the values of q and dq=dT for pure, bulk (the Debye temperature ®xed at Matula's value), these
copper, the maximum values for the temperature co- parameters have huge uncertainties. Many combinations
ecient of resistance at a given temperature can be of the parameters can lead to very nearly the same ®t.
calculated. At 0°C and at 25°C, they are 0:004368 However, the estimated resistivity and its derivative are
0:000049 lX cm °C 1 and 0:003936 0:000044 lX cm well determined, with reasonably small uncertainties.
°C 1 , respectively, where 95% con®dence limits are Denote the intrinsic resistivity data at temperature Tj by
provided. The introduction of impurities and imperfec- qj , for j 1; . . . ; n. The least squares estimates above
tions will reduce the TCR
T . were obtained by ®tting log
q
T to the logarithm of
the intrinsic resistivity data. All of the parameters except
c and p were estimated ®rst, then these values were ®xed
Appendix A. Uncertainty analysis for q
T, q0
T, and and c and p were allowed to vary. This process was re-
dimension estimates peated until convergence.
We now calculate the uncertainties in q
T , q0
T .
A.1. Uncertainty in
T and 0
T Denote standard uncertainties of a quantity q by u
q.
De®ne the vectors of partial derivatives
Intrinsic resistivity as a function of temperature is h iT
approximately v1 oq
ToA
oq
T
oB
oq
T
oC
oq
T
oD
oq
T
op
oq
T
oqR
and
q
T qR G2
T
p h iT
BT h CT h CT o2 q
T o2 q
T o2 q
T o2 q
T o2 q
T o2 q
T
qR A 1 D U ; v2 oT oA oT oB oT oC oT oD oT op oT oqR
;
h CT T T
where the superscript ``T'' indicates a transpose. Let J be
A:1 the Jacobian matrix for logq
T , that is
r^2b
0:000030102 ;
Strictly speaking, of course, q0 and its uncertainty are
both functions of temperature. The range of a 95%
r^2e
0:000049382 ;
con®dence band on q0
T over the temperature range
of 293 425 K is 0:006672 0:006821 lX cm K 1 . We
take the midpoint of this interval, 0.006747 lX cm K 1 , were obtained using restricted maximum likelihood [19].
to be an estimate of q0
T , and the ratio of the half- The uncertainty, u
R, is the square root of the sum of
width of this interval to the appropriate t-multiplier these variances, or
0:00007450 lX cm K 1 =2:005 0:00003716 lX cm K 1 q
as an estimate of u
q0
T . u
R
0:000030102
0:000049382
v
" 0:00005783 X:
u 2 2 #
u
t u u
T u
R
6 t
0:0055 2 2
:
t DT DR For these data, we also have
[14] Touloukian YS, Kirby RK, Taylor RE, Desai PD. Report: 1994 Internat'l. Integrated Reliability Workshop,
Thermal expansion metallic elements and alloys. Thermo- IEEE Catalogue No. 94TH0654-4.
physical properties of matter, vol. 12, The TPRC Data [17] Ryu C, Lee H, Kwon K-W, Loke ALS, Wong SS. Barriers
Series, 1975. for copper interconnections. Solid State Technol 1999:53.
[15] Ku HH. Notes on the use of propagation of error [18] Bevington PR, Robinson DK. Data reduction and error
formulas. J Res National Bureau Standards ± C. Engng analysis for the physical sciences. 2nd ed. New York:
Instrum 1966;70C (4). McGraw Hill; 1992. p. 41±60.
[16] Schat HA, Suehle JS, Albers J. JEDEC ``TCR'' Interlab- [19] Searle SR, Casella G, McCulloch CE. Variance compo-
oratory Experiment ± Lessons Learned, 12±19, Final nents. New York: Wiley; 1992.