International Journal of Thermophysics, Vol. 25, No.

1, January 2004 (© 2004)

Linear Thermal Expansion Coefficient of Silicon from

293 to 1000 K
Hiromichi Watanabe, 1 , 2 Naofumi Yamada, 1 and Masahiro Okaji 1, 3

Received September 2, 2003

As a part of the program to establish a thermal expansion standard, the linear

thermal expansion coefficients of single-crystal silicon have been determined in
the temperature range 293 to 1000 K using a dilatometer which consists of a
heterodyne laser Michelson interferometer and gold versus platinum thermo-
couple. The relative standard deviation of the measured values from those cal-
culated from the best least-squares fit was 0.21%. The relative expanded uncer-
tainty in the measurement was estimated to be 1.1 to 1.5% in the temperature
range, based upon an analysis of thirteen standard uncertainties. The present
data are compared with the data previously obtained by similar dilatometers
and the standard reference data for the thermal expansion coefficient of
silicon, recommended by the Committee on Data for Science and Technology
(CODATA). The present data are in good agreement with the most recently
reported data but not with the earlier high-temperature data used to evaluate
the standard reference data, which suggests a need for the reevaluation of the
standard reference data for the thermal expansion coefficient of silicon at tem-
peratures above 600 K.

KEY WORDS: gold versus platinum thermocouple; heterodyne interferom-

eter; reference material; silicon; standard reference data; thermal expansion.

1. INTRODUCTION
Silicon is an ideal reference material for thermal expansion measurements,
because it is isotropic in thermal expansion owing to its diamond-like crys-
tallographic structure and because it is readily available in an ultra-pure

1
Thermophysical Properties Section, National Metrology Institute of Japan, AIST, 1-1-1
Umezono, Tsukuba, Ibaraki 305-8563, Japan.
2
To whom correspondence should be addressed. E-mail: [email protected]
3
International Affairs Department, AIST, 1-1-1 Umezono, Tsukuba, Ibaraki 305-8563,
Japan.

221

0195-928X/04/0100-0221/0 © 2004 Plenum Publishing Corporation

222 Watanabe, Yamada, and Okaji

condition as a result of the needs of the electronics industries. During the

1970s, cooperative measurements made by various laboratories indicated
that the linear thermal expansion coefficients, a(T), for various samples of
semi-conductor grade, polycrystalline, and single-crystal silicon agreed
within 0.01 × 10 −6 K −1 in the temperature range 280 to 420 K [1]. In 1983,
therefore, Swenson made a critical evaluation of previously reported a(T)
for silicon and produced ‘‘recommended values’’ of a(T) as a function of
temperature T for the CODATA Task Group on Thermophysical Proper-
ties [2]. The CODATA recommended values were deduced from various
absolute determinations of a(T) for both single-crystal and polycrystalline
silicon at temperatures below 850 K, and thus the recommended values
were limited to the temperature range 90 to 850 K. In 1988, Okaji [3]
reported absolute measurements of a(T) on single-crystal silicon at tem-
peratures up to 1300 K using a laser interferometric dilatometer (LID 1).
The feature of LID 1 was a heterodyne laser Michelson interferometer [4],
which results in excellent resolution and accuracy for the measurement of
the length change. Swenson has included the data measured with LID 1 in
a reevaluation [1] of the CODATA recommended values of a(T) for
silicon for the temperature range 90 to 1300 K.
Recently, we have made refinements and improvements to most parts
of LID 1 and have determined a(T) of silicon using the modified dilato-
meter (LID 2) [5]. Comparison of a(T) measured with LID 1 and LID 2
for the same silicon sample reveals that values of a(T) measured with
LID 1 in the temperature range 700 to 1100 K are systematically smaller
than those with LID 2. The disagreement suggests that the measured data
for LID 1 in this temperature range include some systematic errors, which
can be reduced in the measurement with LID 2. Accordingly, there is
concern about the reliability of the standard reference data for a(T) of
silicon, because the present CODATA recommended values at tempera-
tures above 800 K were mainly attributed to the data measured with
LID 1.
Despite the effective improvements, data of a(T) measured with LID 2
are insufficient for the standard reference data, because of several experi-
mental shortcomings that may lead to serious systematic errors in the
temperature measurement below 700 K. In the previous work, a high-tem-
perature platinum resistance thermometer (HTPRT) was used to measure
the sample temperature and the furnace was evacuated to 10 −3 Pa. The
reduced heat transfer due to the absence of gas makes the thermal contact
between the sample and HTPRT become worse at lower temperatures. The
importance of the presence of thermal exchange gas at an appropriate
pressure has been pointed out in ASTM E 289 ‘‘Standard Test Method
for Linear Thermal Expansion of Rigid Solids with Interferometry.’’ In
Linear Thermal Expansion Coefficient of Silicon 223

addition, there was a possibility that the HTPRT was so long (30 mm in
length) that sufficient temperature uniformity within the sensor could not
be attained. As a result, we could not determine a(T) of silicon below
700 K using LID 2, and the absence of the data below 700 K influences the
reliability of our previous data measured using LID 2.
The objective of the present work is to determine reliable data for
linear thermal expansion coefficients of silicon in the temperature range
293 to 1000 K. This work is a part of a project to establish reference
materials for measurements of thermal expansion of solids. To reduce the
temperature difference between the sample and thermometer, the mea-
surements were undertaken with the sample in helium at an appropriate
pressure. The temperature measurements were corrected for the residual
temperature difference between the sample and thermometer. The sample
temperatures were measured using an elemental thermocouple of which the
achievable accuracy was comparable to that of the HTPRT to avoid the
possible error associated with the size of the thermometer sensor. In addi-
tion, a detailed evaluation of experimental uncertainty was carried out.

2. EXPERIMENTAL
The present interferometric dilatometer (LID 3) was almost the same
as LID 2 [5]. Therefore, we describe here an outline of the measurements
and differences from the operations of LID 2.
In the present work, values of a(T) were calculated by applying

a(T)=DL/(L 0 DT), (1)

where DT is a given temperature change from Tn to Tn+1 , T is (Tn +Tn+1 )/2,

L 0 is the initial length of the sample at room temperature, T0 (293 K), and
DL is the change of sample length from L(Tn ) to L(Tn+1 ). We present here
the linear thermal expansion, e(T), calculated using the following expres-
sion in regard to the practical availability of thermal expansion data,

e(T)=(L(T) − L 0 )/L 0 . (2)

Values of DL, DT, and L 0 were measured with a heterodyne laser Michel-
son interferometer, a gold versus platinum (Au/Pt) thermocouple, and a
linear gauge, respectively. Details of the optical system of the interferom-
eter are described in our previous papers [5–7]. The interferometer
employed a double-path system so that

DL=
l Dhp
4 2p
1 2
, (3)
224 Watanabe, Yamada, and Okaji

where l is the wavelength of laser in the furnace and Dhp is the difference in
the phase of the beat signal for optical heterodyne interferometry measured
on changing the temperature from Tn to Tn+1 .
Two samples of single-crystal silicon in which the crystallographic axes
are oriented in [100] and [110] directions were prepared from an ingot
made using a floating-zone method, which is the same batch as that inves-
tigated in our previous works [3, 5]. Each sample was in the form of a
perpendicular block whose size was 10 mm × 7 mm × 20 mm. The (7 × 20)
surfaces were polished to be flat to within one tenth of a wavelength of
632.8 nm and parallel within 5 arc s. The sample was placed on the flat
base plate, which was also made of single-crystal silicon. The mirror face of
the base plate was polished as was that of the sample. A hole having a
diameter of 4 mm was bored down the symmetry axis through the optical
flats of the sample and the base plate.
The Au/Pt thermocouple consisted of gold and platinum wires having
a diameter of 0.25 mm and an insulation tube made of alumina. The
thermocouple was encased in a fused-quartz tube of 3 mm in diameter, and
it was inserted into the hole of the sample. The equation used to compute
the temperature was deduced from the standard reference function given in
ASTM E 1751, the thermocouple calibration at the metal freezing points of
zinc, aluminum, and silver, and the correction for the temperature differ-
ence (Ds ) between the sample (Ttrue ) and the thermocouple (Texp ). Ds as a
function of temperature was experimentally estimated using a ‘‘dummy’’
silicon sample with a differential thermocouple consisting of two platinum
and one platinum-13 mass% rhodium alloy wires, and a type R thermo-
couple in a mineral-insulated, metal-sheathed (MIMS) configuration.
Figure 1 shows a schematic diagram of the arrangement of the dummy
sample and thermocouples to measure Ds and T. The dummy sample had
two extra holes into which the MIMS thermocouples were inserted. To
attain good thermal contact, the diameter of the holes was made almost the
same as that of the metal sheath (1.6 mm). The following correction func-
tion was given by four repeated measurements of Ds in the temperature
range 293 to 1100 K:
Ds (T)=−0.070+0.00108(T − 293), (4)
which was used to compensate for Ds in the measurements of a(T).
The measurements of a(T) were made while the sample was heated
or cooled successively to constant temperatures between which there were
temperature intervals of approximately 50 or 25 K. In order to improve
thermal contact between the sample and thermocouple, an amount of
helium gas at a pressure of 80 Pa and a temperature of 293 K was sealed
in the furnace. After one measurement cycle of heating to approximately
Linear Thermal Expansion Coefficient of Silicon 225

Fig. 1. Schematic diagram of the arrangement of the

sample and differential thermocouple.

1050 K and then cooling to 293 K, the gas pressure at 293 K increased by about
30 Pa as a result of gas evolution from the furnace wall. Taking into account the
gas evolution resulting in the pressure increase from 80 to 110 Pa, we con-
sidered the average amount of helium gas during the experiment to be that at
a pressure of 95 Pa and a temperature of 293 K. Thus, the value of l was
calculated to be 632.991 nm using the refractive index of helium gas at 95 Pa
and 293 K. The temperature of the sample changed at a rate of less than
1.7 × 10 −2 K · s −1 from one constant temperature to the next one. The tempera-
ture was kept at each of the temperature steps for a time period longer than
3.6 × 10 3 s to attain thermal equilibrium. The optical window of the furnace was
in contact with a brass plate whose temperature was kept constant using two
thermoelectric transducers in order to reduce the uncertainty due to the tem-
perature distribution within the window through which the laser beams pass.

3. RESULTS
Figure 2 shows the experimental results for a(T) for the two silicon
samples. For each sample, the measurement in the temperature range 293
to 1025 K was repeated six times; three runs were carried out during
heating and cooling, respectively. All of the data were pooled, and the
following fifth-order polynomial function of temperature was obtained by
the method of least squares: at 293 [ T [ 1025

a(T)/(10 −6 K −1)=−3.0451+0.035705T − 7.981 × 10 −5 T 2

+9.5783 × 10 −8 T 3 − 5.8919 × 10 −11 T 4
+1.4614 × 10 −14 T 5. (5)
226 Watanabe, Yamada, and Okaji

Fig. 2. Linear thermal expansion coefficient, a(T), of

silicon as a function of temperature.

The polynomial function is shown as a solid line in Fig. 2. The standard

deviation of this fit with 192 data points and its relative value is
7.5 × 10 −9 K −1 and 0.21%, respectively. The error bar attached to the line
illustrates the expanded uncertainty (k=2), an estimation of which will be
described in Section 4.1. Figure 3 shows the difference of the data points
from a(T) calculated using Eq. (5). The unfilled and filled circles and
triangles shown in Fig. 3 correspond to a(T) obtained during heating and
cooling for the [100] and [110] samples, respectively. Inspection of Fig. 3
indicates that there was no appreciable deviation between values of a(T)
for the two samples but that values of a(T) measured (ca. 1000 K) during
heating at the upper limit of the temperature were systematically larger
than those measured during cooling.
Table I gives the data of a(T) and e(T) calculated from the fitting
functions and their expanded uncertainties. The values of e(T) listed in the
table were calculated from the following equation based upon a least-
squares analysis of the experimental data of e(T) from the two samples: at
300 [ T [ 1025

e(T) × 10 6=−426.68 − 0.090926T+0.0065359T 2 − 4.7219 × 10 −6 T 3

+1.3822 × 10 −9 T 4. (6)

The standard deviation of this fit is 1.1 × 10 −6 with 204 data points.
Linear Thermal Expansion Coefficient of Silicon 227

Fig. 3. Differences of the present results for a(T) from

the best least-squares fit.

Table I. Results of Thermal Expansion Measurements and Expanded Uncertainties

for Silicon in Temperature Range 293 to 1000 K

Linear Expanded Linear thermal Expanded

thermal uncertainty expansion uncertainty
Temperature expansion (k=2) coefficient (k=2)
T (K) e(T) × 10 6 de(T) × 10 6 a(T) × 10 6 (K −1) da(T) × 10 6 (K −1)

293 0 null 2.57 0.038

300 18.0 1.6 2.63 0.039
350 160.4 1.7 2.97 0.040
400 315.9 1.9 3.24 0.042
450 482.3 2.2 3.44 0.043
500 658.0 2.6 3.60 0.045
550 841.3 2.9 3.73 0.045
600 1031 3.3 3.84 0.046
650 1226 3.7 3.93 0.047
700 1425 4.1 4.00 0.044
750 1627 4.5 4.07 0.045
800 1832 5.0 4.14 0.045
850 2040 5.4 4.19 0.046
900 2250 5.8 4.24 0.046
950 2463 6.3 4.29 0.048
1000 2679 6.7 4.33 0.048
228 Watanabe, Yamada, and Okaji

4. DISCUSSION

4.1. Estimates of Uncertainties

4.1.1. Outline of Estimates

Tables II and III give the respective uncertainty budgets of the linear
thermal expansion coefficient (a) and linear thermal expansion (e) for
silicon measured at 1000 K. The relative combined standard uncertainty
(da/a) listed in Table II is calculated using

da/a=`C {d(DL)/DL} 2+C {d(DT)/DT} 2+C (dL 0 /L 0 ) 2 (7)

where d(DL), d(DT), and dL 0 are the standard uncertainties of DL, DT,
and L 0 , respectively. The magnitudes of uncertainties in Table II were
estimated under the measurement conditions that DT was 50 K, L 0 was
10 mm, a was calculated from Eq. (5) at 1000 K, and DL was given by

DL=aL 0 DT. (8)

On the other hand, the relative combined standard uncertainty (de/e) listed
in Table III is calculated by using

de/e=`C {d(DL)/DL} 2+C (a dT/e) 2+C (dL 0 /L 0 ) 2 (9)

where dT is the standard uncertainty of the temperature. The second term

in the square root of Eq. (9) accounts for the strong dependence of e on
temperature. In the estimation of de/e, the value of e was calculated from
Eq. (6), L 0 was 10 mm, and DL was given by

DL=eL 0 . (10)

In both the uncertainty estimations we attempted to address the standard

uncertainties corresponding to the thirteen original uncertainties listed in
the two tables. The original uncertainties are divided into three categories,
as described below.

4.1.2. Uncertainties Associated with Measurements of Length Change

The first and second uncertainties listed are due to the uncertainty (dl)
of the wavelength of the laser in the furnace. The wavelength was cali-
brated by a beat frequency comparison to an iodine stabilized reference
laser. The beat frequency comparison was carried out over 720 minutes at a
data acquisition rate of six points per minute. The calibration uncertainty
Linear Thermal Expansion Coefficient of Silicon 229

Table II. Uncertainty Budget of Linear Thermal Expansion Coefficient, a,

for Silicon at 1000 K

Measurement conditions

T (K) DT (K) L 0 (m) a ( K −1) DL (m) a

1000 50 1.0 × 10 −2 4.33 × 10 −6 2.16 × 10 −6

Relative
Original Probability uncertainty
Uncertainty source Type uncertainty distribution Divisor (%) Uncertainty

Terms associated with DL dl (m) d d(DL)/DL

1. Calibration uncertainty of
laser wavelength A 2.53 × 10 −16 normal 1 3.99 × 10 −8
2. Instability of refraction
index of gas A 3.32 × 10 −15 normal 1 5.25 × 10 −7
dhm (rad)
3. Nonparallelism of the sample top
surface and the base plate B 1.88 × 10 −3 rectangular 1.73 1.03 × 10 −4
dhp (rad)
4. Uncertainty of measurement of
phase with lock-in amplifier B 1.75 × 10 −2 rectangular 1.73 1.66 × 10 −2
d(DL) (m)
5. Temperature distribution within
optical components A 4.68 × 10 −9 normal 1 2.87 × 10 −1

Terms associated with DT dT (K) d(DT)/DT

6. Calibration uncertainty
of thermocouple B 3.03 × 10 −1 normal 2 4.28 × 10 −1
7. Uncertainty of correction for
temperature difference between
sample and thermocouple A 5.81 × 10 −2 normal 1 1.64 × 10 −1
8. Instability of ice point B 2.00 × 10 −2 rectangular 1.73 3.27 × 10 −2
dVemf (mV)
9. Uncertainty of measurement
of emf with DMM B 1.60 × 10 −3 rectangular 1.73 1.23 × 10 −1

Terms associated with L 0 dL 0 (m) dL 0 /L 0

10. Uncertainty in reading of
linear gauge B 4.00 × 10 −8 normal 2 2.00 × 10 −4
11. Nonparallelism of sample A 3.17 × 10 −8 normal 1 3.17 × 10 −4
12. Temperature difference from T0 B 1.73 × 10 −7 rectangular 1.73 1.00 × 10 −3
13. Resolution of linear gauge B 5.00 × 10 −7 rectangular 1.73 2.89 × 10 −3

da/a b
Combined standard uncertainty 0.556 2.41 × 10 −8
Expanded uncertainty (k=2) 1.11 4.82 × 10 −8

a
DL=aL 0 DT.
b
da/a=`; {d(DL)/DL} 2+; {d(DT)/DT} 2+; (dL 0 /L 0 ) 2.
230 Watanabe, Yamada, and Okaji

Table III. Uncertainty Budget of Linear Thermal Expansion, e, for Silicon at 1000 K

Measurement conditions

T (K) T0 (K) L 0 (m) e DL (m) a a ( K −1)

1000 293 1.0 × 10 −2 2.679 × 10 −3 2.679 × 10 −5 4.33 × 10 −6

Relative
Original Probability uncertainty
Uncertainty source Type uncertainty distribution Divisor (%) Uncertainty

Terms associated with DL dl (m) d d(DL)/DL

1. Calibration uncertainty of
−16
laser wavelength A 2.53 × 10 normal 1 3.99 × 10 −8
2. Instability of refraction
index of gas A 3.32 × 10 −15 normal 1 5.25 × 10 −7
dhm (rad)
3. Nonparallelism of the sample top
surface and the base plate B 1.88 × 10 −3 rectangular 1.73 1.03 × 10 −4
dhp (rad)
4. Uncertainty of measurement of
phase with lock-in amplifier B 1.75 × 10 −2 rectangular 1.73 1.34 × 10 −3
d(DL) (m)
5. Temperature distribution within
optical components A 3.26 × 10 −8 normal 1 1.23 × 10 −1

Terms associated with T dT (K) a dT/e

6. Calibration uncertainty
of thermocouple B 3.03 × 10 −1 normal 2 2.45 × 10 −2
7. Uncertainty of correction for
temperature difference between
sample and thermocouple A 5.81 × 10 −2 normal 1 9.39 × 10 −3
8. Instability of ice point B 2.00 × 10 −2 rectangular 1.73 1.87 × 10 −3
dVemf (mV)
9. Uncertainty of measurement
of emf with DMM B 1.60 × 10 −3 rectangular 1.73 7.03 × 10 −3

Terms associated with L 0 dL 0 (m) dL 0 /L 0

10. Uncertainty in reading of
linear gauge B 4.00 × 10 −8 normal 2 2.00 × 10 −4
11. Nonparallelism of sample A 3.17 × 10 −8 normal 1 3.17 × 10 −4
12. Temperature difference from T0 B 2.19 × 10 −7 rectangular 1.73 1.26 × 10 −3
13. Resolution of linear gauge B 5.00 × 10 −7 rectangular 1.73 2.89 × 10 −3

de/e b
Combined standard uncertainty 0.126 3.37 × 10 −6
Expanded uncertainty (k=2) 0.252 6.75 × 10 −6

a
DL=L − L 0 =eL 0 .
b
de/e=`; {d(DL)/DL} 2+; (a dT/e) 2+; (dL 0 /L 0 ) 2.
Linear Thermal Expansion Coefficient of Silicon 231

of the wavelength was estimated to be 2.53 × 10 −16 m, which was the stan-
dard deviation of the measured wavelengths from the mean. The instability
of the refraction index of the helium gas in the furnace was caused by the
gas evolution from the inside wall of the furnace. The standard uncertainty
of the laser wavelength due to the change in the refraction index was esti-
mated to be 3.32 × 10 −15 m from the observed pressure change from 80 to
110 Pa after one cycle of heating and cooling. The relative uncertainties
due to the first and second sources are calculated as follows:

d(DL)/DL=dl/(ld), (11)

where d is the divisor determined from the probability distribution and

coverage factor of the original uncertainty.
The third uncertainty is attributed to a possible nonparallelism of the
sample top surface and the base plate. For the double-path Michelson
interferometer, the relative standard uncertainty caused by the tilt of the
sample top surface to the base plate is roughly estimated as follows:

d(DL)/DL=−dh 2m /(2d), (12)

where dhm is the tilt angle between the sample top surface and the base
plate in radians. The double-path interferometer considerably reduces the
requirement of the parallelism, because the incident and reflected laser
beams are maintained mutually parallel even in the case where there is tilt
of the optical system or nonparallelism between the top and bottom surfa-
ces of the sample [8]. However, the value of dhm in the present work was
limited to 1.88 × 10 −3 rad (0.108 deg) based upon the following inequality:

2l tan(2dh) [ (DLB +DPD )/2, (13)

where l was the distance (600 mm) between the sample and the cube corner
prism used to return the laser beams to the sample, DLB was the diameter
(3 mm) of the laser beams, and DPD was the diameter (6 mm) of the cir-
cular sensing area of the photodetector. The left side of the inequality
equals the distance by which the nonparallelism of dhm makes the reflected
beam shift from parallel. The interference signal dissipates if the distance
becomes greater than that given by the right side of the inequality. There-
fore, the relative uncertainty due to the third source was calculated by
substituting the upper limit of dhm into Eq. (12) on the assumption that the
probability distribution was a rectangular one.
The fourth uncertainty is associated with the uncertainty of the mea-
surement of the phase (hp ) of the beat signal with the lock-in amplifier used
232 Watanabe, Yamada, and Okaji

in the present work. The magnitude of the fourth original uncertainty (dhp )
was reported as 1.75 × 10 −2 rad (1 degree) from the manufacturer of the
lock-in amplifier (SRS, Model SR830). The phase shift (Dhp ) correspond-
ing to DL is obtained by the two determinations of hp , and thus dhp mul-
tiplied by `2 equals the uncertainty of Dhp . Therefore, the relative uncer-
tainty is calculated using

d(DL)
DL
=
`2 l dhp
d DL 4 2p
. 1 2 (14)

The fifth uncertainty is associated with the temperature distribution

within optical components through which the laser beams pass. It is
impossible to eliminate temperature distributions within all the optics that
are irradiated by the thermal radiation from the furnace. The residual
temperature distribution causes a difference in refractive index within the
optics, which leads to the apparent length change [9]. Figure 4 shows a
typical experimental result of the apparent length change for the case where
only the base plate was subjected to the same stepwise heating cycle as that
carried out in the thermal expansion measurement. The filled circles
and error bars correspond to the mean value and the fluctuation of the
apparent length change while the temperature of the plate was held con-
stant, respectively. The slope of the line between the two circles is referred
to as the zero drift. The apparent length measurement from 293 to 1050 K

Fig. 4. Typical change in apparent length in the case

where only the base plate was set in the furnace.
Linear Thermal Expansion Coefficient of Silicon 233

was repeated two times during heating and cooling. The value of the
original uncertainty was calculated as follows:

d(DL)=`(`2 dL A ) 2+(DT · SA ) 2, (15)

where dL A and SA were the respective average values of the instability and
the magnitude of the zero drift obtained during the four measurement runs.

4.1.3. Uncertainties Associated with Temperature Measurements

First, it should be noted that the temperature uncertainty (dT) mul-
tiplied by `2 equals the uncertainty of DT, because DT is obtained from
the two determinations of T. The sixth original uncertainty is identical to
the expanded uncertainty (k=2) associated with the thermocouple cali-
bration at the three metal freezing points, the value of which was provided
from the calibration laboratory (Yamari Industries) of the thermocouple.
The value of the seventh original uncertainty was estimated as the standard
deviation of the measured Ds from the correction function given by Eq. (4).
The eighth uncertainty is associated with the fluctuation of the temperature
of the cold junction of the thermocouple. The temperature of the junction
was stabilized by an automatic ice-point cell. The long-term instability of
the ice point realized by the cell was used to estimate the eighth uncer-
tainty. The ninth uncertainty is associated with the uncertainty of the mea-
surement of the voltage (Vemf ) of the thermocouple emf with a digital multi-
meter (DMM) used in the present work. The magnitude of the ninth
original uncertainty (dVemf ) was calculated using the relation between the dc
voltage reading and accuracy, which was provided from the manufacturer
of the DMM (Keithley, Model 2001).

4.1.4. Uncertainties Associated with Measurements of Initial Length of

Sample
The tenth original uncertainty is identical to the expanded uncertainty
(k=2) of the length of the gauge block used for the correction of the
reading of the linear gauge carried out before and after the measurement
of L 0 . The value of the expanded uncertainty was provided from the
manufacturer (Mitsutoyo) of the gauge block. The eleventh original uncer-
tainty is estimated from the nonparallelism of the top and bottom surfaces
of the sample. The twelfth uncertainty is associated with the possible tem-
perature difference between T0 and the real temperature on the measure-
ments of L 0 . The listed value of the twelfth original uncertainty was esti-
mated from the thermal expansion or contraction due to the expected
maximum temperature difference (2 K) on the assumption that the prob-
ability distribution was a rectangular one. The last original uncertainty is
234 Watanabe, Yamada, and Okaji

identical to the uncertainty due to the resolution of the linear gauge (Sony,
Model DZ521).

4.2. Comparisons with Previously Reported Data

Figure 5 shows four sets of the differences from the present results
(Eq. (5)) for the data previously measured with LID 1 and LID 2 and the
previous and present CODATA recommended data (CODATA 1 and
CODATA 2). The thick solid and broken lines correspond to the data for
LID 1 and LID 2, and the unfilled and filled circles indicate the CODATA
1 and CODATA 2, respectively. Two thin broken lines denote the positive
and negative expanded uncertainties in the present measurements.
It must be emphasized that the present results are in good agreement
with the data measured using LID 2 but not with those measured using
LID 1 in a temperature range 700 to 1000 K. From the characteristics of
LID 1, we can consider that the earlier high-temperature data included
systematic errors due to the following factors: (a) temperature difference
between sample and thermocouple, (b) long-term drift in the signal of the
thermocouple, and (c) the periodic nonlinearity due to imperfect separation
of the two optical frequencies. In the operation of LID 1, the sample was
rapidly heated with a radiant image furnace and a type R thermocouple
was used to measure the sample temperature. In general, it is difficult to

Fig. 5. Differences of a(T) measured with LID 1 and

LID 2 and the present and previous CODATA recom-
mended data from the present results.
Linear Thermal Expansion Coefficient of Silicon 235

uniformly heat the bulk sample and thermometer using radiant image fur-
naces. A fairly large temperature difference between the sample and ther-
mometer or a serious nonuniformity in temperature within the sample
would exist in the measurement. Compared to the Au/Pt thermocouple
and HTPRT, type R thermocouples are sufficiently sensitive but lack long-
term stability. It is considered that the use of a Zeeman laser as the light
source of the heterodyne interferometry in LID 1 increased the periodic
nonlinearity, more or less. In the operations of the later interferometers
(LIDs 2 and 3), the probe and reference laser beams were generated using
the combination of a stabilized He-Ne laser unit and a pair of acousto-
optical modulators to reduce the periodic nonlinearity. From the concerns
about the significant error, we can conclude that the standard reference
data should be reevaluated without the data measured using LID 1 at
temperatures above 600 K, although the difference between the present
data and CODATA 2 is within the magnitude of the expanded uncertainty
in the present work.

5. CONCLUSIONS
Linear thermal expansion data, a(T) and e(T), on single-crystal
silicon in the temperature range 293 to 1000 K have been determined using
a laser interferometric dilatometer. The expanded uncertainties (k=2) of
a(T) and e(T) in the measured temperature range were estimated to be
3.8 × 10 −8 to 4.8 × 10 −8 K −1 and 1.6 × 10 −6 to 6.7 × 10 −6, respectively. The
respective standard deviations of the measured a(T) and e(T) were esti-
mated to be 7.5 × 10 −9 K −1 and 1.1 × 10 −6, which were within the mag-
nitudes of their expanded uncertainties.
The present results are in good agreement with the most recently
reported data [5] but not with the earlier reported data at temperatures
above 700 K [3], both of which were measured with the dilatometers
similar to that used in the present work. The poor agreement with the
earlier reported data, which were used to estimate the present CODATA
recommended values [1], suggests the need for a reevaluation of the stan-
dard reference values of a(T) on silicon at temperatures above 600 K.

ACKNOWLEDGMENTS
The authors wish to thank Dr. Jun Ishikawa of the Time and
Frequency Division of our institute, who calibrated the wavelength of the
laser used in this work. This work is a result of the national research
program ‘‘Research on Measurement Technology and Reference Materials
for Thermophysical Properties of Solids’’ which was financially supported
236 Watanabe, Yamada, and Okaji

by the Promotion System for Intellectual Infrastructure of Research and

Development.

REFERENCES
1. G. K. White, in CINDAS Data Series on Material Properties, Vols. 1–4: Thermal Expansion
of Solids, R. E. Taylor, ed. (ASM International, Materials Park, OH, 1998), Chap. 11,
pp. 269–285.
2. C. A. Swenson, J. Phys. Chem. Ref. Data 12:179 (1983).
3. M. Okaji, Int. J. Thermophys. 9:1101 (1988).
4. M. Tanaka, T. Yamagami, and K. Nakayama, IEEE Trans. Instrum. Meas. 38:552 (1989).
5. H. Watanabe, N. Yamada, and M. Okaji, Int. J. Thermophys. 23:543 (2002).
6. N. Yamada and M. Okaji, High Temp. High Press. 29:89 (1997).
7. H. Watanabe, N. Yamada, and M. Okaji, Int. J. Thermophys. 22:1185 (2001).
8. M. Okaji and H. Imai, Prec. Eng. 7:206 (1985).
9. M. Okaji and H. Imai, J. Phys. E 20:887 (1987).