You are on page 1of 12

730 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO.

3, JUNE 2015

Temperature Dependence of the Elastic


Constants of Doped Silicon
Eldwin J. Ng, Vu A. Hong, Yushi Yang, Chae Hyuck Ahn, Camille L. M. Everhart, and Thomas W. Kenny

Abstract— Resonators fabricated in heavily doped silicon loop combined with a micro-oven [10] or a temperature
have been noted to have a reduced frequency-temperature sensor [11], [12]. In recent times, another method for tem-
dependence compared with lightly doped silicon. The resonant perature compensation has emerged: a reduced temperature
frequency of silicon microelectromechanical systems (MEMS)
resonators is largely governed by the material’s elastic properties, dependence has been observed for silicon resonators that have
which are known to depend on doping. In this paper, a suite been heavily doped with either p-type [13] or n-type [14], [15]
of different types and orientations of resonators were used to dopants. Typically, it is seen that a doping concentration
extract the first- and second-order temperature dependences of of 1e19 to over 1e20 cm−3 is necessary before frequency-
the elastic constants of p-doped silicon up to 1.7e20 cm−3 , and temperature compensation effects start to become significant.
n-doped up to 6.6e19 cm−3 . It is shown that these temperature-
dependent elastic constants may be used in finite element analysis A suggested mechanism is that heavy doping strains the crystal
to predict the frequency-temperature dependence of similarly lattice and shifts the electronic energy bands, resulting in a
doped silicon resonators. [2013-0331] flow of charge carriers to minimize the free energy, thereby
Index Terms— Doping, resonators, silicon, temperature changing the elastic properties [6], [13], [16], [17].
dependence. From previous work, it is observed that the frequency-
temperature dependences are not tied solely to the doping
I. I NTRODUCTION concentration; resonators with different mode and orientations
on the same wafer can exhibit markedly different temperature
S ILICON has been widely used as the base material
for MEMS applications because of the availability of
high quality wafers, established fabrication processes, and
dependences [14], making it difficult for a MEMS designer
to predict the frequency-temperature dependences for more
reliable techniques for integration and packaging. Excellent complex geometries.
stability has been observed for well-packaged silicon devices, However, because the resonant frequency is largely
due to silicon’s near-ideal crystalline properties that governed by a material’s elastic properties, it is possible to use
include the absence of observable fatigue and temperature known elastic properties and their temperature dependences
hysteresis [1]–[4]. Furthermore, the mechanical properties of to predict the frequency-temperature dependence of arbitrary
monocrystalline silicon are well characterized [5], [6] and resonant modes. The temperature dependence of the elastic
have been used successfully for the predictive modeling of properties has previously been extracted for lightly doped
numerous MEMS devices. silicon [6], [18], and more recent work has elucidated the
Resonant systems form an important group of MEMS heavily n-doped regime [19]. The heavy p-doped regime has
devices. When used as time-keeping devices, the frequency yet to be characterized, and in this work, we characterize
stability of resonators is extremely important, and a crucial six different dopings of silicon (up to 1.7e20 cm−3 p-type
issue that needs to be addressed is the frequency-temperature and 6.6e19 cm−3 n-type). Based on the method outlined
sensitivity, which is often around −30 ppm/°C for silicon. in [18], a suite of different types and orientations of resonators
Many methods have been employed to compensate for this were fabricated with the various dopings, and the doping and
temperature dependence, such as using a composite silicon- the temperature dependences of the elastic constants were
silicon dioxide structure [7]–[9], or an active phase-locked extracted. Further, it is demonstrated that the extracted material
properties can be used to predict the frequency-temperature
Manuscript received October 25, 2013; revised July 10, 2014; dependence of different types of resonators.
accepted August 5, 2014. Date of publication August 28, 2014; date
of current version June 1, 2015. This work was supported in part by
the Defense Advanced Research Projects Agency through the Precision
Navigation and Timing Program under Grant N66001-12-1-4260 and II. M ETHOD FOR E XTRACTING THE E LASTIC C ONSTANTS
through the Mesodynamic Architectures Program under Grant FA8650-
13-1-7301, and in part by the National Science Foundation through The elastic constants can be obtained by several methods,
the National Nanotechnology Infrastructure Network under Grant with the most commonly cited values measured using the
ECS-9731293. Subject Editor H. Seidel. transit times of pulse-echo techniques [6]. Resonant systems
The authors are with the Department of Mechanical Engineering, Stanford
University, Stanford, CA 94305 USA (e-mail: eldwin@mems.stanford.edu; can also be used to characterize the elastic constants. The
vuhong@mems.stanford.edu; yushiyang@mems.stanford.edu; ahn1229@ frequency of a resonator is dependent on its geometry and
mems.stanford.edu; everhart@mems.stanford.edu; tkenny@stanford.edu). the material properties. By measuring the resonant frequencies
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. of a variety of resonators and their temperature dependences,
Digital Object Identifier 10.1109/JMEMS.2014.2347205 and linking this together with finite element models of the
1057-7157 © 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted,
but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
NG et al.: TEMPERATURE DEPENDENCE OF THE ELASTIC CONSTANTS OF DOPED SILICON 731

or in compliance form as
⎡ ⎤ ⎡ ⎤⎡ ⎤
1 s11 s12 s12 0 0 0 σ1
⎢ 2 ⎥ ⎢ s12 s11 s12 0 0 0 ⎥ ⎢ σ2 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 3 ⎥ ⎢ s12 s12 s11 0 ⎥ ⎢ ⎥
⎢ ⎥=⎢ 0 0 ⎥ ⎢ σ3 ⎥ (3b)
⎢ 4 ⎥ ⎢ 0 0 ⎥ ⎢ ⎥
⎢ ⎥ ⎢ 0 0 s44 0 ⎥ ⎢ σ4 ⎥
⎣ 5 ⎦ ⎣ 0 0 0 0 s44 0 ⎦ ⎣ σ5 ⎦
6 0 0 0 0 0 s44 σ6

Fig. 1. Extraction method for the temperature dependence of the elastic where there are three independent components (s11 , s12 , s44 ),
constants. or (c11 , c12 , c44 ) that define the material.
The equivalent stiffnesses and compliances for some ideal
resonant modes are given in Table I. It is apparent that
resonant geometries, the elastic constants and their temperature
working in the compliance form compared to the stiffness
dependences can be extracted (Fig. 1).
form for these resonators is especially convenient due to the
Using a lumped element model, the resonant frequency can
linear relations. From the table, it is seen that a choice of
be expressed as
 uniaxial and Lamé modes [21] in the <100> and <110>

1 k 1 ceq 1 1 directions is particularly useful extracting the elastic constants
f = = = √ (1)
2π m Lef f ρ L e f f ρseq because they span the vector space well: uniaxial modes along
the <100> direction are dependent solely on s11 ; Lamé modes
where k is the equivalent spring constant, m is the equivalent
along the <110> direction are dependent solely on s44 ;
mass, L e f f is the effective length (or wavelength), ρ is the
the s12 term appears strongly in both the uniaxial mode in
material density, ceq is the equivalent material stiffness, and
the <110> direction and the Lamé mode in the <100>
seq = 1/ceq is the equivalent compliance. The frequency-
direction. These equations could be directly used in the
temperature dependence can be categorized into two types:
extraction of the elastic constants, however, because fabricated
thermal expansion (L and ρ), and elastic properties (seq ).
devices are expected to deviate slightly from these ideal
modes due to geometrical differences (such as the presence
A. Frequency Dependence on Thermal Expansion of supporting anchors, and release-etch holes), finite element
L and ρ are affected by thermal expansion, which for silicon analysis (Section II.C) was used to account for the actual
can be approximated with temperature coefficients of first and fabricated geometry.
second order, α (1) = 2.84 ppm/°C and α (2) = 8.5 ppb/°C2 It is first noted that the elastic constants do vary slightly
respectively, as reported in [18]. The variation in frequency with doping [6], resulting in a slight frequency shift.
due to temperature can then be written in fractional form with The elastic constants for doped silicon at the reference
a Taylor expansion, keeping up to the quadratic terms in T: temperature (s11,0 , s12,0 , s44,0 ) can be extracted by starting
  
f   f  with values for undoped silicon (s11,0 = 7.691e-12 Pa−1 ;
≈ (1 + α (1) T + α (2) (T )2 ) 1 + −1 s12,0 = −2.142e-12 Pa−1 ; s44,0 = 12.58e-12 Pa−1 ) and
f
0 f  0 α=0
(2) minimizing the least-squares fractional difference between
the finite-element simulated frequencies and the measured
where the subscript 0 is used to denote quantities at a frequencies using an iterative gradient descent. The extracted
reference temperature T0 (here 25 °C), and f0f |α=0 is the reference-temperature elastic constants for the various dopings
component due to the variation of elastic properties ceq (or can then be applied as the linearization point for subsequent
seq ) with temperature, with the thermal expansion assumed temperature dependence analysis.
to be zero. Consequently, the effect of thermal expansion can The temperature dependence of the elastic constants
be decoupled from the effect of the elastic properties. For all can then extracted by first relating the fractional depen-
dopings, the density at T0 is assumed to be 2330 kg m−3 [6]. dence of seq to that of the three independent elastic com-
pliances, using a multivariable Taylor expansion to 2nd
B. Frequency Dependence on Elastic Properties order about the reference-temperature compliance values
Focusing now on the elastic properties, it is known that (s11,0 , s12,0 , s44,0 ):
with the reference axis oriented along the <100> directions,     
seq si 1 si s j
the elastic stiffness properties of monocrystalline silicon (with ≈ μi + μi, j
cubic symmetry) can be represented as a 4th rank tensor that seq,0 i si,0 2 i j si,0 s j,0
can be simplified (ignoring higher order terms) as [20] (4)
⎡ ⎤ ⎡ ⎤⎡ ⎤
σ1 c11 c12 c12 0 0 0 1 where i and j each index the 11, 12 and 44 components,
⎢ σ2 ⎥ ⎢ c12 c11 c12 0 ⎥ ⎢ ⎥
⎢ ⎥ ⎢ 0 0 ⎥ ⎢ 2 ⎥ and
⎢ σ3 ⎥ ⎢ c12 c12 0 ⎥ ⎢ ⎥
⎢ ⎥=⎢ c11 0 0 ⎥ ⎢ 3 ⎥ (3a) 
seq
 
seq

⎢ σ4 ⎥ ⎢ 0 0 ⎥⎢⎥ ⎥ ∂ seq,0 ∂ 2 seq,0
⎢ ⎥ ⎢ 0 0 c44 0 ⎢ 4 ⎥
⎣ σ5 ⎦ ⎣ 0 0 0 0 c44 0 ⎦ ⎣ 5 ⎦ μi =   ; μi, j =    s 
σ6 0 0 0 0 0 c44 6 ∂ s
si,0
i
∂ ssi,0 ∂ s j,0
i j
732 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 3, JUNE 2015

TABLE I
E FFECTIVE E LASTIC C ONSTANTS OF C OMMON R ESONANT M ODES AND T HEIR F REQUENCY-C OMPLIANCE C OEFFICIENTS

The fractional frequency dependence on the equivalent with the a’s being the directional cosines between the stress
compliance can also be expanded from (1) as direction x and the reference axis (1, 2, 3) oriented along
     the <100> directions. With the linear form of (7), (4) is
 f  1 seq 3 seq 2
≈ − + (5) reduced to
f0 α=0 2 seq,0 8 seq,0
seq  si,0   si 
and after substituting (4) in, = λi (9)
    seq,0 i seq,0 si,0
 f  1 si
≈ − μ
f 0 α=0
i
2 i s and the coefficients γ in (6b) then correspond to
 i,0     
1 si s j
− μi, j 1
γi = − λi
si,0
(10a)
4 i j si,0 s j,0
    2 seq,0
3 si s j  
s j,0

+ μi μ j (6a) 3
γi, j = λi λ j
si,0
(10b)
8 i j si,0 s j,0
      8 seq,0 seq,0
si si s j
= γi + γi, j where values for uniaxial stress orientations along the <110>
i si,0 i j si,0 s j,0
(6b) and <100> directions have been calculated in Table I.
Similarly, square extensional modes can be approximated to
where the coefficients of the fractional compliances in (6a) be in biaxial stress with two superimposed uniaxial modes
have been grouped together as γi and γi, j in (6b). perpendicular to each other [22], and from (3b), the equiv-
Although (6b) can be derived directly from (1) by expanding alent compliance can be worked out assuming σ1 = σ2 .
with respect to the fractional compliances, the given relation Lamé modes are dependent on the shear modulus which is
between (6a) and (6b) is useful for analytically computing given for cubic crystals as a function of the perpendicular
the coefficients in some cases, such as uniaxial, square exten- directions of propagation and displacement, x and y [23]
sional, and Lamé resonant modes.
For uniaxial stress along an arbitrary direction x, the λ11 = 4ψ (11a)
equivalent compliance can be obtained from (3b) by means λ12 = −4ψ (11b)
of a tensor rotation [20]:
λ44 = (1 − 2ψ) (11c)
seq = λ11 s11 + λ12 s12 + λ44 s44 (7) ψ = ax1
2 2
a y1 + a x2
2 2
a y2 + ax3
2 2
a y3 (11d)
where
The coefficients γ for the Lamé and square extensional modes
λ11 = 1 − 2β (8a) have similarly been analytically calculated (Table I). It is also
λ12 = 2β (8b) useful to note that resonant orientations along the <100> and
λ44 = β (8c) <110> directions typically represent the extremes for (100)
silicon wafers according to (7) and (11), and are hence the
β = ax1 ax2 + a x2
2 2 2 2
ax3 + a x1
2 2
ax3 (8d)
most useful for characterizing the elastic properties.
NG et al.: TEMPERATURE DEPENDENCE OF THE ELASTIC CONSTANTS OF DOPED SILICON 733

C. Finite Element Analysis TABLE II


C HARACTERISTICS OF THE WAFERS IN T HIS A NALYSIS
For a general resonant mode, finite element analysis can be
directly used to simulate the coefficients γ in (6b). Each 2D
resonant geometry (including support tethers and release-etch
holes) as drawn in the mask layout was imported into COM-
SOL 4.3b. A lateral etch of 0.1 μm into the features (decreas-
ing the device dimensions) was taken into account based on
scanning electron microscope (SEM) cross-section images of
a typical trench etch. This 2D geometry was extruded to the
corresponding device layer thickness to form the 3D geometry.
A planar mesh of triangular elements was first generated across
the 2D geometry with a maximum element size of 5 μm,
and a higher mesh density (1 μm) at the support tethers and
release-etch holes. This planar mesh was then swept in the temperature coefficients pk and qk :
thickness direction with a maximum element size of 10 μm,
from the top to the bottom faces of the geometry. Quarter-  fk
≈ pk (T ) + qk (T )2 (14)
and half-symmetry conditions were exploited where possible f k,0
to reduce the computation time. Device support tethers were It is readily seen that by equating the first order temperature
terminated in a fixed condition at the anchor. The initial coefficients of (13) and (14),
elastic compliances were varied by a small amount si (1% ⎡ ⎤ ⎡ ⎤ ⎡ (1) ⎤
percent of si,0 , based on the expected temperature variation), pk1 γk1 ,11 γk1 ,12 γk1 ,44 Ts
and these elastic properties were substituted into COMSOL ⎢ pk ⎥ α (1) ⎢ γk ,11 γk ,12 γk ,44 ⎥ ⎢ 11 (1) ⎥
⎢ 2⎥− =⎣⎢ 2 2 2 ⎥ ⎢ T s 12 ⎥
while retaining the same mesh An eigenmode analysis was ⎣ pk3 ⎦ 2 γk3 ,11 γk3 ,12 γk3 ,44 ⎦ ⎣ ⎦ (15)
(1)
performed and the frequency change of the desired eigenmode ... ... ... ... T s 44
was noted for the si variation, thereby allowing for the
where the subscripts k1 , k2 , k3 denote different resonator
coefficients γ to be extracted. For the resonant geometries
designs. If more than three resonators are available, the
used in the extraction, a finer mesh with about 1.5 times the
unknown T si(1) are fitted in the least-squares sense. The second
density was then used to ensure that values of coefficients γ
had converged to ±0.0002. order coefficients T si(2) can be fitted similarly.

III. M EASUREMENT OF F REQUENCY-T EMPERATURE


D. Extraction of the Temperature Coefficients of Compliance
D EPENDENCES
Finally, the elastic compliances were approximated to have
A. Fabrication
a polynomial dependence on temperature:
The resonators in this work were fabricated using an
si (1) (2) epitaxial polysilicon encapsulation process (epi-seal). The
= T s i T + T s i (T )2 + · · · (12)
si,0 epi-seal process is a wafer-scale, hermetic encapsulation
(1) (2) process that seals the MEMS within an ultra-clean, sub-Pa
where T si and T si are the temperature coefficients of the
vacuum cavity that is free from native oxide [24], [25]. This
compliances of first and second order respectively and are the
results in high stability resonators with ppb-level frequency
parameters to be extracted for each doping type/concentration.
drift [4]. The process was originally proposed by researchers
The fractional frequency change with temperature can thus
at the Robert Bosch Research and Technology Center in
be derived by substituting (12) into (6b) and then into (2).
Palo Alto and then demonstrated in a close collaboration
Expanding the result and keeping terms up to the second order
with Stanford University. This collaboration is continuing to
in T gives
develop improvements and extensions to this process for many
 
f 1 (1) applications, while the baseline process has been brought into
≈ (T) α (1) + γi T s i commercial production by SiTime Inc.
f0 2 i

1 1  (1) 2 1 (1) (1)
In the baseline process, the device doping concentration and
+ (T)2 α (2) − α + α γi T s i variation is determined by Czochralski-grown wafers. As heav-
2 8 2 i
 ily boron-doped wafers were not available, epitaxial deposition
(2)
+ γi T s i + γi, j T s (1)
i T s (1)
j was used to fabricate the doped device layers on some starting
i (i, j )
(13) silicon-on-insulator (SOI) wafers [26] prior to the epi-seal
process. This also allows for large released areas without
As there are three independent compliances (s11 , s12 , s44 ), release-etch holes (unlike the baseline epi-seal process), and
at least three measured frequency-temperature curves from the doping concentration can be carefully controlled by gas
three different resonators are necessary to extract the T si . The flow rates. For the other dopings in this process, SOI wafers
measured frequency-temperature curve for a resonator k can directly purchased from wafer vendors were used. A summary
first be fitted in a least-squares sense with linear and quadratic of the wafers used in this analysis is provided in Table II.
734 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 3, JUNE 2015

no purge gas was used (air ambient). Frequency sweeps were


performed with an Agilent 8753ES Network Analyzer while
operating the devices in the linear regime, and the RTD
temperature was recorded at the time of the frequency sweep.
The resonant frequency was determined by the maximum point
of a fitted curve around the peak of the frequency sweep,
and the quality factor (Q) was computed using the −3 dB
bandwidth method. Frequency-temperature curves were thus
mapped out for the various doping concentrations and least-
squares quadratic curves were fitted with coefficients p and q
around T0 = 25 °C.

IV. R ESULTS
A. Comparison of Analytical and Finite Element Results
Several resonators were designed around the resonant modes
in the above analysis and are presented in Table III. Uniaxial
resonant modes include the length extensional (LE) and width
Fig. 2. Cross-section SEM of a typical device encapsulated with the epi-seal
process. extensional (WE) modes, as well as double-ended tuning
forks (DETF) [7]. Lamé, square extensional (SqExt), double
breathe-mode rings (Ring) [27] and disk resonating gyroscopes
The epi-seal fabrication technique has been detailed before (DRG) [28] represent typical resonant designs of interest, and
in [24], but in short, the process begins with deep reactive all these devices together span resonant frequencies of over
ion etching (DRIE) trenches (limited to 0.7–1.5 μm wide) two orders of magnitude Devices without release-etch holes
to define the resonator and electrical isolation on a device were fabricated only for the heavy boron-doped wafers and
layer. A 2 μm layer of oxide is deposited to bridge the DRIE not the other dues to limited release areas of the baseline epi-
trenches and then electrical contact vias are etched in the seal process (Section III.A).
oxide. A 6 μm layer of epitaxial polysilicon is deposited The devices were simulated for the corresponding device
over the oxide, and vent holes are then etched into the thickness in Table II, with the coefficients γ for a repre-
polysilicon. The oxide around the resonator is etched away sentative thickness listed in Table III. The values of the
through the vent holes with vapor HF, and the cavity is coefficients γ obtained by analysis (Table I) and finite element
sealed by depositing a thick (∼20 μm) layer of epitaxial simulation (Table III) are seen to be in good agreement,
polysilicon over the vents. It is this final seal in a clean, especially for the LE and Lamé modes without release-etch
high temperature hydrogen environment that confers the ideal holes. Release-etch holes, support tethers, and mode shape
packaging properties mentioned above. Electrical isolation and differences are seen to cause the coefficients of the actual
contacts through the encapsulation are formed, completing the devices to deviate slightly from the analytical model.
process. A cross-section SEM of the final device is shown in
Fig. 2. B. Quality Factors
The electrical resistivities of the device layers were cal-
culated by measuring electrical test structures in the device High quality factor (Q) resonators are necessary for accurate
layer located near the resonators to give a good estimate for frequency measurements. Typical measured Qs (Table III) for
the doping concentration. The device layer thicknesses were this work are in the range of several hundreds of thousands,
measured using a cross-section SEM of a location on the wafer indicating that the peak frequency can typically be determined
in proximity to the tested devices. with frequency sweeps accurately to well under a ppm. The
Q for most of these resonators are known to be limited by
thermoelastic dissipation [29]–[31], but could also be limited
B. Measurement Setup by the Akhieser effect [32] for the Lamé modes without
To obtain frequency-temperature curves, the encapsulated release-etch holes. Anchor losses could also be a limiting
resonators were wire bonded to a package and mounted on a factor for the WE resonators.
printed circuit board. These were then placed in a Thermotron
S-1.2 environmental chamber for temperature control and
allowed to stabilize at the initial temperature step (−40 °C) for C. Elastic Properties of Doped Silicon
60 min and subsequent temperature steps (5 °C increments to The frequency-temperature dependences of the various res-
85 °C) for 30 min each. The temperature was measured using onators were fitted with quadratic curves centered around
a platinum resistance temperature detector (RTD) placed near 25 °C – the coefficients are given in Table III and the plots in
the device, and it was observed that the temperature reached Fig. 3. Not all devices were fabricated on all wafers: a couple
stabilization within 30 min for the initial step and 10 min for of lower stiffness designs experienced stiction issues, leading
subsequent steps. Since these were encapsulated resonators, to gaps in the table.
NG et al.: TEMPERATURE DEPENDENCE OF THE ELASTIC CONSTANTS OF DOPED SILICON 735

TABLE III
R ESONANT M ODES : M ODELED AND M EASURED R ESULTS

The extracted doping dependences of the elastic constants were used for the elastic constant extraction, as these were
as well as their first and second order temperature dependences believed to best span the vector space of the elastic constants,
are listed in Table IV. Data from the shaded cells in Table III as well as being the least susceptible to process variations.
736 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 3, JUNE 2015

Fig. 3. Measured frequency-temperature curves of all the different resonators and doping types/concentrations in this work.

TABLE IV
D OPING D EPENDENCE OF THE E LASTIC C ONSTANTS AND T HEIR T EMPERATURE D EPENDENCES

The other data were used for verification of the extracted Devices with frequency-temperature turnover points (zero first-
values (Section V). order coefficients) can be achieved with either heavy p- or
The values in Table IV are plotted in Fig. 4 as a function of n-type doping with an appropriate choice of resonant mode
doping concentration, with increasing p-type character towards (Fig. 3). P-type doping is seen to affect the temperature depen-
the left and increasing n-type character to the right. Error dence of s44 significantly, allowing for frequency-temperature
bars are from the calculations in Section IV.D (Table V). turnover points for resonant modes that depend strongly
Values reported in literature [6], [18], [19] are also shown and on s44 . N-type doping achieves similar significant temperature
are generally in agreement with the values extracted in this compensation effects by means of s11 and s12 .
work. These values, together with fits to the available data, The results of the heavily p-doped devices are particularly
are available on our group’s website [33], and will be updated interesting for temperature compensation. Comparing devices
as new data emerges. The fits allow for frequency predictions with first-order compensation (turnover points) within the
at arbitrary doping concentrations within the given range. temperature range, the frequency-temperature curvatures for
It is seen that the resonant mode as well as the doping both p-doped Lamé modes oriented in the <110> direction
significantly affect the frequency-temperature dependence. are much smaller than those achieved by n-doped devices.
NG et al.: TEMPERATURE DEPENDENCE OF THE ELASTIC CONSTANTS OF DOPED SILICON 737

Fig. 4. The elastic constants of doped silicon and their temperature dependences, with p-type doping represented on the left with increasing doping and
n-type on the right. The extracted values in this work are compared with values reported in literature.

TABLE V
U NCERTAINITIES OF THE E LASTIC C ONSTANTS AND T HEIR T EMPERATURE D EPENDENCES

The curvature coefficients q are seen to be around resonators with high f-Q products of 2e13 Hz have been
−25 ppb/°C2 for the p-types, compared to about −80 ppb/°C2 demonstrated [26].
for n-type devices. Furthermore, even though the two heavy
p-doped concentrations are close to each other at 1.4 and
1.7e20 cm−3 , a significantly lower curvature of the heavier D. Uncertainty Analysis
p-doped device is observed (−21 ppb/°C2 vs −31 ppb/°C2 ). The uncertainties of the elastic constants and their tem-
It is expected that even heavier doping could yield lower perature dependences are listed in Table V. Uncertainties are
curvatures. It is also noted that the quality factors of these believed to arise from several sources:
Lamé mode resonators are also largely unaffected by the 1) Process Variation: Fabrication process variations around
heavy doping – heavily p-doped temperature-compensated the wafer such as in doping concentration and etch profiles are
738 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 3, JUNE 2015

known to cause a spread in the measured resonant frequency, 4) Temperature Inaccuracies: The chamber temperature
resulting in an uncertainty in the extracted values. The mea- varied as much as ± 1 °C at −40 °C. However, the temperature
sured resonant frequency at T0 for the resonant modes used used in this calculation was measured with an RTD (accuracy
in the elastic constant extraction was observed to fall within of ± 0.15 °C), taken at the time of the frequency measurement.
± 1000 ppm (95% confidence interval). This was propagated To investigate the temperature variation in the chamber, two
to the elastic constants by rewriting (1) as RTDs on circuit boards similar to the resonator setup were
1 placed at different locations in the chamber and measured
seq = (16) simultaneously. It was found that the two RTDs tracked each
f 2 ρ L 2e f f other to within ± 0.4 °C at −40 °C and within ± 0.2 °C
and using the relations in Table I to estimate the uncertainty from −30 °C to 85 °C. An uncertainty of ± 0.4 °C at
in si and then ci . the temperature extremes was propagated to the temperature
The uncertainties of the temperature coefficients were esti- coefficients using (14) and (15).
mated as the 95% confidence intervals of the fitted parameters 5) Linearization Point: The linearization points (c11,0 , c12,0
T si in the least-squares linear regression performed in (15), c44,0 ) that were used in the temperature dependence analysis
and these were converted to the corresponding uncertainties derives from the extracted values for the corresponding doping
for T ci . concentration. The uncertainty of the linearization point was
2) Dimensional Offsets: While the above section on process propagated to the temperature coefficients by repeating the
variations addresses the spread of the data, offset errors could finite element analysis using the error bounds of the extracted
also be incurred during fabrication. The DRIE process is elastic constants as the linearization points.
known etch laterally, causing the resonant geometry dimen- 6) Thermal Expansion and Density: The thermal expansion
sions to be smaller than designed, and hence L e f f in (16). coefficients and density of silicon have been assumed to be
While a lateral etch of 0.1 μm was accounted for to capture constant across all dopings in this analysis. Measurements in
this offset (Section II.C), there could be some error in this literature have shown minimal effects due to doping, resulting
value, and the lateral dimensions of the finite element model in uncertainties on the order of a few hundred ppm [6], [19] –
were hence varied ± 50 nm in addition to the 0.1 μm already small compared to the other uncertainties. Though the ther-
accounted for. Re-running the finite element model with these mal expansion coefficient was assumed to be constant, its
dimensional variations resulted in changes to the resonant effect on the temperature coefficients has been accounted for
frequencies as well as the coefficients γ . The resonant fre- (Section II.A).
quency changes were propagated as uncertainties in the elastic 7) Electrostatic Spring Softening: The electrostatic actu-
constants ci , while changes in γ resulted in uncertainties of ation force is known to cause a reduction of the spring
the temperature coefficients T ci using the relation (15). It is stiffness proportional to the square of the applied bias voltage,
observed that the uncertainties due to lateral etch offsets were resulting in a lowering of the resonant frequency. This effect
larger for devices with release-etch holes compared to devices was characterized by measuring the resonant frequency at
without release-etch holes, where the higher values of the several bias voltages. For the bulk-mode resonators used in
specified ranges in Table V correspond to the devices with this extraction, the frequency shift due to the spring softening
etch holes. For devices with etch holes, the lateral dimension effect was under 20 ppm, negligible compared to the other
offset is seen to be one of largest causes of uncertainty in ci . uncertainties detailed above.
The thicknesses of the device layers were measured using
SEM imaging at locations close to the tested devices. However, V. P REDICTED F REQUENCY-T EMPERATURE D EPENDENCE
there could also be offset errors due to within-wafer thickness AND V ERIFICATION OF E XTRACTED VALUES
variations: ± 1 μ m variations in thickness were simulated
The importance of the extracted temperature-dependent
due to wafer specifications. The effect of thickness was
elastic constants is that they allow for the frequency-
simulated by varying the finite element model and performing
temperature dependence of doped silicon resonators to be
a similar procedure to the lateral dimension offset analysis
predicted. The temperature dependence of the elastic constants
above. The uncertainties due to thickness variation were found
can be accounted for in the material properties of finite element
to be relatively small compared to the other error sources
analysis software by substituting in the relevant elastic prop-
as Lamé modes are in-plane modes that are theoretically
erties for each temperature step and performing an eigenmode
insensitive to thickness variations, while length extensional
analysis. The fractional frequency change is then noted for
modes have a low sensitivity to thickness variations for a

each step, giving the frequency-temperature dependence due
device length much greater than the device thickness. 
3) Rotational Misalignment: The device-to-crystal align- to the elastic properties, i.e. the f0f  component in (2).
α=0
ment from wafer specifications could be off by as much Finally, the thermal expansion effect is then added back as
as 1° in plane, and 0.5° out of plane. The effect on the per (2), resulting the total frequency-temperature dependence.
elastic constants and their temperature dependences can be The code for performing this simulation in COMSOL with
estimated using (7), (8), and (11) to analytically obtain the MATLAB LiveLink) is available on our group’s website [33].
uncertainties due to the crystal misalignment. It is seen that Resonant modes not used in the elastic constant
the misalignment does not significantly affect the results at extraction are then used for verification. Since only a few
these small angles. modes (as shaded in Table III) were used for the extraction,
NG et al.: TEMPERATURE DEPENDENCE OF THE ELASTIC CONSTANTS OF DOPED SILICON 739

Fig. 5. Predicted (lines) vs. measured (markers) frequency-temperature curves for the resonant modes not used in the extraction. Predictions were made
using the extracted elastic constants.

TABLE VI  
 pred meas / f meas ,
D ISCREPANCY B ETWEEN P REDICTIONS AND M EASUREMENTS F IRST ROW:  f 25 °C − f 25 °C 25 °C
 pred
 f meas  
S ECOND ROW: M AX   fpred − f meas  − 40 °C to 85 °C
f 25 °C 25 °C

predictions were performed for the other available modes, and fabrication process variation is believed to have a more pro-
these predictions are compared to experimental data (Fig. 5). nounced effect, leading to larger discrepancies. Another cause
The discrepancies between the predicted and measured values could be the dopant out-diffusion that occurs from the oxide-
are listed in Table VI, for the resonant frequencies at 25 °C free silicon surface [34] during the high temperature sealing
as well as the fractional frequency variations across the process. The experimental frequency-temperature plots for the
−40 °C to 85 °C temperature range. For bulk-mode resonators, DRG correspond to a slightly lower doping concentration than
the mean resonant frequency discrepancy is about 0.5%, while modeled, indicating some dopant loss may have occurred.
the maximum discrepancy for the fractional frequency varia-
tion across the temperature range is on average around 20 ppm.
VI. C ONCLUSION
Larger discrepancies of about 2% and 80 ppm respectively
are noted for the flexural mode resonators (DETF and DRG). Silicon resonators have differing frequency-temperature
Since the DETF beams are 6 μm in width, and the DRG characteristics that are dependent on doping and the
consists of numerous thin concentric rings of 3 μm width, resonant mode. Using a suite of different types of resonators,
740 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 24, NO. 3, JUNE 2015

the first and second order temperature dependences of the [19] A. Jaakkola, M. Prunnila, T. Pensala, J. Dekker, and P. Pekko, “Determi-
elastic constants were extracted for p-doped silicon up to nation of doping and temperature-dependent elastic constants of degen-
erately doped silicon from MEMS resonators,” IEEE Trans. Ultrason.,
1.7e20 cm−3 , and n-doped silicon up to 6.6e19 cm−3 . Ferroelectr., Freq. Control, vol. 61, no. 7, pp. 1063–1074, Jul. 2014.
Encapsulated Lamé resonators at 10 MHz with quality factors [20] J. F. Nye, Physical Properties of Crystals: Their Representation by
of 1.7 million have been temperature compensated to first Tensors and Matrices. Oxford, U.K.: Oxford Univ. Press, 1985.
[21] K. F. Graff, Wave Motion in Elastic Solids. New York, NY, USA: Dover,
order using p-type doping, and are additionally seen to have 1975.
a much reduced second order temperature dependence, as [22] V. Kaajakari et al., “Square-extensional mode single-crystal silicon
low as −21 ppb/°C2 , compared to n-doped silicon. With micromechanical RF-resonator,” in Proc. Solid-State Sens., Actuators,
Microsyst. Conf. (TRANSDUCERS), Jun. 2003, pp. 951–954.
the extracted elastic constants, it is seen that the fractional [23] L. J. Walpole, “The elastic shear moduli of a cubic crystal,” J. Phys. D,
frequency-temperature dependence of arbitrary linear bulk Appl. Phys., vol. 19, no. 3, pp. 457–462, Mar. 1986.
resonant modes with similar doping concentrations can be [24] R. N. Candler et al., “Long-term and accelerated life testing of a novel
single-wafer vacuum encapsulation for MEMS resonators,” J. Micro-
predicted to about 20 ppm. electromech. Syst., vol. 15, no. 6, pp. 1446–1456, Dec. 2006.
[25] A. Partridge and M. Lutz, “Episeal pressure sensor and method for mak-
R EFERENCES ing an episeal pressure sensor,” U.S. Patent 6 928 879, Aug. 16, 2005.
[26] E. J. Ng, Y. Yang, Y. Chen, and T. W. Kenny, “An etch hole-free process
[1] V. Kaajakari, J. Kiihamäki, A. Oja, S. Pietikäinen, V. Kokkala, and for temperature-compensated, high Q, encapsulated resonators,” in Proc.
H. Kuisma, “Stability of wafer level vacuum encapsulated single-crystal Solid-State Sens., Actuators, Microsyst. Workshop, Hilton Head, SC,
silicon resonators,” Sens. Actuators A, Phys., vols. 130–131, pp. 42–47, USA, Jun. 2014, pp. 99–100.
Aug. 2006. [27] S. Wang et al., “Nonlinearity of hermetically encapsulated high-Q
[2] B. Kim, R. N. Candler, M. A. Hopcroft, M. Agarwal, W.-T. Park, double balanced breathe-mode ring resonator,” in Proc. IEEE Int. Conf.
and T. W. Kenny, “Frequency stability of wafer-scale film encapsulated Micro Electro Mech. Syst. (MEMS), Jan. 2010, pp. 715–718.
silicon based MEMS resonators,” Sens. Actuators A, Phys., vol. 136, [28] C. H. Ahn et al., “Geometric compensation of (100) single crystal
no. 1, pp. 125–131, May 2007. silicon disk resonating gyroscope for mode-matching,” in Proc. Solid-
[3] V. A. Hong et al., “High-stress fatigue experiments on single crys- State Sens., Actuators, Microsyst. Conf. (TRANSDUCERS), Jun. 2013,
tal silicon in an oxygen-free environment,” in Proc. Solid-State pp. 1723–1726.
Sens., Actuators, Microsyst. Workshop, Hilton Head, SC, USA, 2012, [29] R. N. Candler et al., “Impact of geometry on thermoelastic dissipation in
pp. 271–274. micromechanical resonant beams,” J. Microelectromech. Syst., vol. 15,
[4] E. J. Ng, H. K. Lee, C. H. Ahn, R. Melamud, and T. W. Kenny, “Stabil- no. 4, pp. 927–934, Aug. 2006.
ity of silicon microelectromechanical systems resonant thermometers,” [30] C. Tu and J. E.-Y. Lee, “Study on thermoelastic dissipation in bulk mode
IEEE Sensors J., vol. 13, no. 3, pp. 987–993, Mar. 2013. resonators with etch holes,” in Proc. 7th IEEE Int. Conf. Nano/Micro
[5] M. A. Hopcroft, W. D. Nix, and T. W. Kenny, “What is the Young’s Eng. Molecular Syst. (NEMS), Mar. 2012, pp. 478–482.
modulus of silicon?” J. Microelectromech. Syst., vol. 19, no. 2, [31] Y. Yang, E. J. Ng, C. H. Ahn, V. A. Hong, E. Ahadi, and
pp. 229–238, Apr 2010. T. W. Kenny, “Mechanical coupling of dual breathe-mode ring
[6] J. J. Hall, “Electronic effects in elastic constants of n-type silicon,” Phys. resonator,” in Proc. Solid-State Sens., Actuators, Microsyst. Conf.
Rev., vol. 161, no. 3, pp. 756–761, 1967. (TRANSDUCERS), Jun. 2013, pp. 502–505.
[7] R. Melamud et al., “Temperature-insensitive composite micromechanical [32] S. A. Chandorkar, M. Agarwal, R. Melamud, R. N. Candler,
resonators,” J. Microelectromech. Syst., vol. 18, no. 6, pp. 1409–1419, K. E. Goodson, and T. W. Kenny, “Limits of quality factor in bulk-mode
Dec. 2009. micromechanical resonators,” in Proc. IEEE Int. Conf. Micro Electro
[8] R. Tabrizian, G. Casinovi, and F. Ayazi, “Temperature-stable sili- Mech. Syst. (MEMS), Jan. 2008, pp. 74–77.
con oxide (SilO x ) micromechanical resonators,” IEEE Trans. Electron [33] E. J. Ng. (2014). Predicting f-T Characteristics of Resonators in
Devices, vol. 60, no. 8, pp. 2656–2663, Aug. 2013. Doped Silicon. [Online]. Available: http://micromachine.stanford.edu/
[9] D. M. Chen, J. H. Kuypers, A. Gaidarzhy, and G. Zolfagharkhani, projects/doping/
“Mechanical resonating structures including a temperature compensation [34] M. S. Lee et al., “Effect of ashing, strip and annealing process on
structure,” U.S. Patent 8 362 675, Nov. 15, 2011. the dopant concentration of silicon,” ECS Trans., vol. 11, pp. 211–217,
[10] J. C. Salvia, R. Melamud, S. A. Chandorkar, S. F. Lord, and T. W. Kenny, Sep. 2007.
“Real-time temperature compensation of MEMS oscillators using an
integrated micro-oven and a phase-locked loop,” J. Microelectromech.
Syst., vol. 19, no. 1, pp. 192–201, Feb. 2010.
[11] R. Melamud et al., “MEMS enables oscillators with sub-PPM fre- Eldwin J. Ng received the B.S. degree in mechan-
quency stability and sub-ps jitter,” in Proc. Solid-State Sens., Actuators, ical engineering from the University of California
Microsyst. Workshop, Hilton Head, SC, USA, 2012, pp. 66–69. at Berkeley, Berkeley, CA, in 2009, and the M.S.
[12] K. J. Schoepf, “TCMO: A versatile MEMS oscillator timing platform,” degree in mechanical engineering from Stanford
in Proc. Precise Time Time Interval (PTTI), Nov. 2009, pp. 481–492. University, Stanford, CA, USA, in 2012, where he is
[13] A. K. Samarao and F. Ayazi, “Temperature compensation of silicon res- currently pursuing the Ph.D. degree with the Depart-
onators via degenerate doping,” IEEE Trans. Electron Devices, vol. 59, ment of Mechanical Engineering under a scholar-
no. 1, pp. 87–93, Jan. 2012. ship from the Agency for Science, Technology, and
[14] M. Shahmohammadi, B. P. Harrington, and R. Abdolvand, “Turnover Research, Singapore. His research interests include
temperature point in extensional-mode highly doped silicon microres- microfabrication technologies, RF resonators, and
onators,” IEEE Trans. Electron Devices, vol. 60, no. 3, pp. 1213–1220, MEMS sensors and switches.
Mar. 2013.
[15] T. Pensala, A. Jaakkola, M. Prunnila, and J. Dekker, “Temperature
compensation of silicon MEMS resonators by heavy doping,” in Proc.
IEEE Int. Ultrason. Symp., Oct. 2011, pp. 1952–1955. Vu A. Hong received the B.S. degree in mechani-
[16] R. W. Keyes, “Electronic effects in the elastic properties of semi- cal engineering from the Massachusetts Institute of
conductors,” Solid State Phys., vol. 20, pp. 37–90, Oct. 1967, doi: Technology, Cambridge, MA, USA, in 2010, and
10.1016/S0081-1947(08)60217-9. the M.S. degree in mechanical engineering from
[17] F. S. Khan and P. B. Allen, “Temperature-dependence of the elastic- Stanford University, Stanford, CA, USA, in 2012,
constants of p+ silicon,” Phys. Status Solidi B, vol. 128, pp. 31–38, where he is currently pursuing the Ph.D. degree with
Mar. 1985. the Department of Mechanical Engineering.
[18] C. Bourgeois, E. Steinsland, N. Blanc, and N. F. de Rooij, “Design of Mr. Hong was a recipient of the National Science
resonators for the determination of the temperature coefficients of elastic Foundation Graduate Research Fellowship.
constants of monocrystalline silicon,” in Proc. IEEE Int. Freq. Control
Symp., May 1997, pp. 791–799.
NG et al.: TEMPERATURE DEPENDENCE OF THE ELASTIC CONSTANTS OF DOPED SILICON 741

Yushi Yang received the dual bachelor’s degree Thomas W. Kenny received the B.S. degree from
in mechanical engineering from Purdue Univer- the University of Minnesota, Minneapolis, MN,
sity, West Lafayette, IN, USA, and Shanghai Jiao USA, in 1983, and the M.S. and Ph.D. degrees
Tong University, Shanghai, China, in 2011, and from the University of California at Berkeley, Berke-
the M.S. degree in mechanical engineering from ley, CA, USA, in 1987 and 1989, respectively,
Stanford University, Stanford, CA, USA, in 2013, all in physics. From 1989 to 1993, he was with
where she is currently pursuing the Ph.D. degree the Jet Propulsion Laboratory, National Aeronautics
with the Department of Mechanical Engineering. and Space Administration, Pasadena, CA, where his
Her research interests include studying the nonlin- research focused on the development of electron-
ear behavior of bulk-mode MEMS resonators, and tunneling high-resolution microsensors. In 1994, he
analyzing the phase noise performance of MEMS joined the Department of Mechanical Engineering,
oscillators under large driving conditions. Stanford University, Stanford, CA, where he directs microsensor-based
research in a variety of areas, including resonators, wafer-scale packaging,
cantilever beam force sensors, microfluidics, and novel fabrication techniques
for micromechanical structures. He is the Founder and CTO of Cooligy Inc.,
Mountain View, CA, (currently, a Division of Emerson), a microfluidics chip
Chae Hyuck Ahn received the B.S. degree in cooling component manufacturer, and is the Founder and a Board Member of
mechanical engineering from Seoul National SiTime Corporation, Sunnyvale, CA, a developer of timing references using
University, Seoul, Korea, in 2010, and the M.S. MEMS resonators. He is currently a Bosch Faculty Development Scholar
degree in mechanical engineering from Stanford and was the General Chairman of the 2006 Hilton Head Solid-State Sensor,
University, Stanford, CA, USA, in 2012, where Actuator, and Microsystems Workshop, and will be the General Chair of the
he is currently pursuing the Ph.D. degree with upcoming Transducers 2015 meeting in Anchorage, AK, USA. From 2006 to
the Department of Mechanical Engineering under 2010, he was on leave to serve as a Program Manager with the Microsystems
a Kwanjeong Scholarship. His research interests Technology Office, Defense Advanced Research Projects Agency, Arlington,
include resonant thermometers, micromachined VA, USA, starting and managing programs in thermal management, nanoman-
disk resonating gyroscopes, silicon anisotropy ufacturing, manipulation of Casimir forces, and received the Young Faculty
compensation by geometric design, and the Award. He has authored and co-authored over 250 scientific papers, and holds
integration of inertial measurement units and timing references. 50 issued patents.

Camille L. M. Everhart received the B.S. degree


in mechanical engineering with a minor in microsys-
tems from the Massachusetts Institute of Technol-
ogy, Cambridge, MA, USA, in 2013. She is currently
pursuing the master’s degree with the Department of
Mechanical Engineering, Stanford University, Stan-
ford, CA, USA, supported by a Stanford Graduate
Fellowship as a Chambers Fellow. She is currently
a Research Assistant with the Micro Structures
and Sensors Laboratory, Stanford University. Her
research interests include microstructures and sys-
tems, with a focus on energy-efficient sensors.

You might also like