Dissipation in ultrahigh quality factor SiN membrane resonators

S. Chakram, Y. S. Patil, L. Chang and M. Vengalattore

Laboratory of Atomic and Solid State Physics, Cornell University, Ithaca, NY 14853
(Dated: November 7, 2013)
We study the optomechanical properties of stoichiometric SiN resonators through a combination
of spectroscopic and interferometric imaging techniques. At room temperature, we demonstrate
ultrahigh quality factors of 5 107 and a f Q product of 1 1014 Hz that, to our knowledge,
correspond to the largest values yet reported for mesoscopic flexural resonators. Through a com-
arXiv:1311.1234v1 [cond-mat.mes-hall] 5 Nov 2013

prehensive study of the limiting dissipation mechanisms as a function of resonator and substrate
geometry, we identify radiation loss through the supporting substrate as the dominant loss process.
In addition to pointing the way towards higher quality factors through optimized substrate designs,
our work realizes an enabling platform for the observation and control of quantum behavior in a
macroscopic mechanical system coupled to a room temperature bath.

PACS numbers: 85.85.+j,42.50.-p,62.25.-j,37.90.+j

Mesoscopic mechanical resonators with ultrahigh qual- material choice and design of the supporting substrate.
ity factors are ubiquitous ingredients in diverse applica- In addition, our resonators are a promising platform for
tions of sensing, inertial navigation and communications the observation and control of quantum behavior in a
[1, 2] as well as in foundational tests of quantum me- mesoscopic mechanical system at room temperature.
chanics in macroscopic systems [36]. The quantum co- The mechanical oscillators in our study are fabri-
herent control of these resonators and the realization of cated by NORCADA Inc., and consist of LPCVD Sili-
quantum-limited sensors require the cooling of these res- con nitride square membranes under high tensile stress
onators to low phonon occupancies. However, to date, of around 0.8 - 0.9 GPa. The membranes range in thick-
existing schemes of optomechanical cooling place strin- ness from h 30 200 nm with lateral dimensions in the
gent constraints on the mechanical and optical proper- range L 0.5 5 mm. The membranes are deposited on
ties of these resonators. Crucially, overcoming the ther- single crystal Silicon wafers. The membranes constitute
mal coupling to the environment and the attainment of one arm of a Michelson interferometer while the other
long coherence times require low mechanical resonance (reference) arm is actively stabilized against ambient vi-
frequencies, a large frequency - quality factor product brations. This realizes a precise measurement of the in-
(f Q > kB Tambient /h) and low optical absorption stantaneous position of the membrane with a sensitivity
[7, 8]. This combination has hitherto been difficult to of 0.1 pm/Hz1/2 for typical powers of 200 W incident on
achieve despite the exploration of a wide range of micro- the membrane. The optical measurements are performed
and nanoscale systems. Thus, recent demonstrations of with a external-cavity diode laser (ECDL) operating at
optomechanical cooling to the mechanical ground state a wavelength of 795 nm. At this wavelength, real and
[9, 10] require cryogenic cooling of the mechanical sys- imaginary parts of the refractive index of the membranes
tem to reduce the thermal coupling to the environment. are measured to be R(n) = 1.95 0.02 and I(n) < 105
Stoichiometric silicon nitride membrane resonators respectively, leading to a peak reflectivity of 0.34 for the
show great promise for optomechanics due to their high 100nm thick resonators (see Supplementary Information
quality factors and low optical absorption [11, 12]. Re- for details).
cent work has demonstrated f Q products at room From thin plate theory, the mechanical eigenfrequen-
temperature as high as 2 1013 Hz in these resonators
p p
cies of the membranes are jk = 2 /4L2 j 2 + k 2
[13]. However, the dominant loss mechanisms and poten- where is the intrinsic tensile stress, = 2.7 g/cm3 is
tial routes to further enhance the quality factors of these the mass density and L the lateral dimension of the mem-
resonators remain poorly understood. brane. We have confirmed this relation is accurate at the
In this Letter, we demonstrate membrane resonators of 0.1% level by spectroscopy of the various modes and es-
stoichiometric Silicon nitride with quality factors in ex- timate the tension in the range of 0.8 - 0.9 GPa for the
cess of 5107 and a frequency-Q product f Q 11014 various samples studied.
Hz that is more than an order of magnitude greater than The mechanical quality factors for various modes are
the requirement for ground state cooling and coherent measured through ringdown measurements of the mem-
quantum control of a room temperature optomechanical brane oscillation. For this, the membrane is piezo-
system. Further, we identify radiation loss through the actuated at the various membrane resonances up to am-
supporting substrate as the dominant loss mechanism. plitudes of around 200 pm for durations of 10 ms be-
This finding points to further enhancements of the perfor- fore switching off the drive. The amplitude of membrane
mance of such membrane resonators through appropriate oscillation is then monitored through a lock-in ampli-
 2

8 8
10 10
(L/h)2

Peak Quality Factor

7 (L/h)
Quality Factor

10
100 7
10

Amplitude [pm]
10
6
10

1
6
20 60 100 140 10
5 Time [s] 4 5
10 10 10
0.0 0.5 1.0 1.5 2.0 2.5 3.0 L/h
Frequency [MHz]
FIG. 2. Peak mechanical quality factors versus film geometry
FIG. 1. Peak mechanical quality factors of a L = 5 mm, parametrized by the ratio of membrane width (L) to mem-
h = 100 nm SiN membrane versus frequency (). The solid brane thickness (h). For L/h < 105 , we observe a scaling
line corresponds to f Q = kB /h 300K. For low mode fre- consistent with Q (L/h)2 . A linear scaling is also shown
quencies (jk < 300 kHz) , the quality factors can be improved for reference.
by an order of magnitude () simply by reducing the contact
region between the substrate and the in-vacuum mount. The
weak frequency dependence of the measured Qs at high fre- factors.
quencies is further reduced for h = 30 nm (), (see text for
For our measurements, we distinguish between two
discussion). Inset: Characteristic mechanical ringdown of the
(6,6) mode at 66 = 678 kHz. regimes of behavior : (i) For mode indices (j 2 + k 2 )1/2 <
4, the measured Qs exhibit highly non-monotonic behav-
ior with increasing frequency. We also observe a large
fier to measure the (1/e) decay time of the oscilla- variation and a sensitive dependence of the quality fac-
tion amplitude. The quality factor is then estimated as tors on the clamping mechanism as well as the geom-
Qjk = jk /2 where jk is the eigenfrequency of the etry of the modal structure. Our observations in this
mode under study. regime are quantitatively consistent with the dominant
Various checks of systematic effects on these measure- loss mechanism being anchor losses from the membrane
ments were made to ensure the validity of our interpre- into the supporting mount [14]. The sensitive dependence
tation. These include the negligible effect of radiation of the Qs on the clamping mechanism can be greatly re-
pressure or photothermal heating due to the laser field duced (,Fig. 1) by ensuring minimal contact between
on the mechanical motion, the linearity of the drive as the supporting silicon wafer and the in-vacuum mount,
well as the negligible influence of viscous damping at our reinforcing the above interpretation. (ii) For mode in-
operating background pressure of p 2 107 Torr dices (j 2 +k 2 )1/2 & 4, the peak Qs reach a plateau around
(see Supplementary Information for details). Changes 5107 . The peak quality factors in this regime are a weak
in the peak amplitude by up to a factor of five in ei- function of the resonant frequencies with a scaling esti-
(0.70.15)
ther direction does not change the measured value of the mated as Qjk jk for the 100 nm membranes.
quality factor indicating that we are operating far from This scaling becomes much weaker as the thickness of the
any intrinsic nonlinearities of the mechanical resonator. membrane is reduced (, Fig.1), resulting in Qs that are
For the typical amplitudes of mechanical motion during almost independent of frequency. In this latter regime,
the ring-down measurements, self-stiffening nonlineari- the measured Qs are more robust to variations in the
ties were measured to be below the 1 ppm level. Finally, clamp making it less obvious that direct anchor loss is
the mechanical linewidth inferred from thermal Brownian the dominant loss mechanism.
motion is consistent with that derived from the ringdown We have measured the peak Qs in the plateau regime
measurements. for a range of membrane geometries ranging in width
Fig. 1 shows the typical trend observed for the peak from L = 0.5 5 mm and in thickness from h = 30 200
quality factors of various mechanical modes exhibited by nm (Fig.2). For L/h < 105 , we find the following scaling
the membrane. We typically extend our measurements relations Q (L/h)2 and f Q (L/h). The noticeable
up to mode indices (j 2 + k 2 )1/2 40. For higher mode discrepancy in this scaling for the thinnest membranes is,
indices, the rapidly increasing density of modes renders it at present, unexplained [15].
challenging to accurately resolve individual modes of the We have performed a series of experiments to elucidate
resonator. More importantly, we also observe substantial the limiting damping process of the mechanical excita-
intermodal coupling between proximal eigenmodes which tions in the plateau regime. In the context of our system,
complicates the interpretation of the measured quality the possible range of mechanisms include thermoelastic
 3

damping (TED) [1618], damping due to localized de- modeled as being due to two level systems (TLS) whose
fects on the surface or within the bulk of the resonator energy splitting is modulated by the oscillating strain
[19, 20], anharmonic processes such as Akhiezer damp- field [26]. The subsequent re-equilibration of these two
ing [21] and anchor (radiation) losses from the resonator level systems results in attenuation of the mechanical en-
to the supporting structure [14, 2224]. Based on mod- ergy. At the elevated temperatures of our experiments,
els and experimental measurements described below, we these two-level systems can be regarded as being ther-
find quantitative evidence that our current quality fac- mally activated over a wide range of energy scales. For
tors are limited by anchor losses from the membrane into the mechanical frequencies in this work, the TLS model
the substrate. [27] predicts a scaling of Qjk 1/jk and a dissipa-
tion that scales very weakly with the modal structures,
(a) (c) (d) dimensions of the resonator and details of the support
200
structure. These predictions are inconsistent with our
150
observations (see Fig. 2,4).
The quality factors for a given sample remain stable
100
pm

within 10% of the measured values even after exposure

50
to air for several days. We have also annealed the mem-
0
branes at temperatures up to 650 C under vacuum to re-
duce surface contamination without a significant change
in quality factors. These observations rule out surface-
induced losses as a contributing influence.
In addition to the above observations that rule out
specific intrinsic processes, a more mechanism-agnostic
(b) (e) (f)
argument can be formulated for a large class of intrin-
sic dissipation mechanisms. For any such process that
FIG. 3. Interferometric imaging of the mechanical modes : In couples mechanical motion to a source of dissipation, the
situ images of the (a) (1,10) mode and (b) (9,9) mode (Color
leading symmetry-allowed term in the equation of mo-
scale for displacement shown). Substrate-induced coupling
between proximal asymmetric eigenmodes result in hybridiza- tion must be proportional to the local curvature of the
tion into more symmetric structures. (c),(d) show the modal displacement field (see, for example [28]). Thus, the mea-
structures corresponding to 10,6 6,10 hybridized modes. sured quality factors for the various modes should cor-
These can be compared to the calculated mode profiles for relate with the local modal curvature or higher powers
this hybridization (e,f). thereof. In order to better quantify this reasoning, we
have developed an interferometric imaging technique ca-
Firstly, we have developed a model of thermoelastic pable of spatially resolving the modal structure of the res-
dissipation in our membrane resonators that takes into onator [29]. These images (Fig.3) yield a wealth of spatial
account their large intrinsic tensile stress of 1GPa. information complementing our spectroscopic measure-
Within this model, the mechanical motion of the mem- ments. In the plateau regime, we observe that the quality
brane couples to a local temperature field associated with factors can vary by almost two orders of magnitude for
microscopic changes in the volume of the resonator. This a corresponding variation in the integrated curvature of
results in local irreversible heat flows and dissipation (see less than 20% pointing to an extremely weak correlation
Supplementary Information for more details). Our model between the two quantities.
is built upon the formalism introduced in [16, 25] and We also note that we have measured Qs up to 2.7 107
accurately reproduces the eigenfrequencies of our res- in low stress SiN membranes ( = 0.25 GPa) of similar
onators for the entire range of geometries studied. Im- geometry, i.e. within a factor of two of those measured
portantly, for established material parameters of stoichio- in the high stress membranes. Further, we observe a
metric SiN, our model predicts a room temperature limit substantial influence of the substrate on the membrane
of QT ED 1012 for the frequency ranges of our study. modes (see Fig.3) with nominally degenerate eigenmodes
Further, our model also predicts that the quality factor hybridizing into more symmetric structures. At larger
2 drive amplitudes than those used in this study, we ob-
scales as Q 1/jk in the frequency regime jk /11 > 10
in distinct contrast to the weak dependence experimen- serve multimode bistability as a result of such coupling
tally observed (Fig.1). Based on these model predictions, to the substrate [30]. Based on the preceding arguments,
we discount TED in the membrane as a limiting influence we conclude that there is little or no influence of intrinsic
of our observed quality factors. material processes on our measured Qs.
Another source of dissipation is the coupling between Finally, we discuss the role of radiation loss from the
the mechanical motion and intrinsic, localized defects membrane into the supporting substrate. This loss of
within the membrane. While the microscopic origins of mechanical energy arising from the coupling to the ex-
these defects remain unknown, these losses are typically ternal supporting substrate represents one of the funda-
 4

30
mental restrictions to a high-Q device that is only weakly
dependent on the material parameters of the resonator. 25 60
50
Accordingly, various treatments and models of this ra- 20 10
7
40
diation loss [22, 31, 32] have been developed to address

x10
15 30

Q
j
design methodologies that can alleviate this loss. A par-

6
20
10
ticularly intuitive picture of phonon tunneling [14, 33] 10
6
has recently emerged, wherein the resonator can be re- 5 0 10
(a) (b)
garded as a phononic cavity coupled to the external sub-
5 10 15 20 25 30 10 20 30 40 50 60 70 80
strate through a weak coupling parameter. k
Guided by our Q-factor measurements and modal im-
ages, we have extended this model to the higher mode 12 (c) (d) 7
10
indices of the plateau regime and obtain excellent agree- 10

ment (see Supplementary Information for details). In 8

20

Q
j
order to decouple the geometrical dependence of the 6 15 6
10
quality factor from other frequency-dependent factors,

x10 6
4 10
we compare the model to Qs measured over a narrow
2 5
range
p of frequency corresponding to an arc of radius 5
10
j 2 + k 2 12 in the (j, k) space. In particular, we see 20 4 6
k
8 10 12 10 20 30 40 50 60 70 80

that the substrate-induced mode coupling and ensuing
hybridization (see Fig. 3) suppresses the large Q varia-
FIG. 4. (a) Predicted quality factors versus mode indices
tion between modes of even and odd parity that is seen based on asymptotic limits of our anchor loss model (see Sup-
at lower frequencies. Instead, we observe a more gradual plementary Information for details), (b) Predicted quality fac-
variation with lower Qs measured for modes with either tors versus arctan(j/k) for mode indices (j 2 +k2 )1/2 = 12
j k or j k, and higher Qs measured for j k with (without) substrate-mediated hybridization are shown as
(see Fig. 4). The close agreement between our observa-  (). Also shown is the asymptotic expression for the qual-
tions and the tunneling model (suitably modified to in- ity factors from our model (dashed line). (c) Measured Qs for
clude interference effects arising from substrate-induced mode indices indicated by the green arc in (a). (d) Compari-
son between the measured Qs in this arc and our predictions
hybridization), further clarifies the dominant role of an- from (b).
chor losses in determining our peak quality factors.
Our modified treatment of the phonon tunneling model
accurately captures the increased robustness and en- the optomechanical cooling and quantum control of a
hanced quality factors arising from the interference be- mesoscopic resonator at room temperature. Further, the
tween nominally degenerate resonator modes. One can low mechanical frequencies and ultrahigh quality factors
extend this logic to the complementary scenario, i.e. demonstrated in this work are ideally suited to various
where interference between distinct substrate modes lead schemes to interface atomic gases or solid state spin sys-
to a similar enhancement of the quality factors. The ex- tems to the mechanical degree of freedom , thereby real-
treme limit of this latter scenario is when the substrate izing hybrid quantum devices for sensor and transduction
exhibits an acoustic bandgap in the vicinity of the rele- applications as well as for fundamental studies [34].
vant resonator modes. This should lead to a vastly en- We acknowledge S. Vengallatore for valuable discus-
hanced quality factor and f Q product that is limited, sions, A. K. Bhat for experimental assistance and E. J.
in principle, only by the material properties of the mem- Mueller for a critical reading of the manuscript. This
brane. work was supported by the DARPA QuASAR program
In summary, we demonstrate stoichiometric SiN mem- through a grant from the ARO and the Cornell Center for
brane resonators with ultrahigh quality factors up to Materials Research with funding from the NSF MRSEC
5 107 and f Q 1014 Hz that is, to our knowledge, program (DMR-1120296). M. V. acknowledges support
the largest reported in a room temperature mechanical from the Alfred P. Sloan Foundation.
system. We have developed models of our system that
identify anchor loss as the dominant decay mechanism
that limits the current quality factors. Remarkably, the
material properties of stoichiometric Silicon nitride do
not limit the performance of these membrane resonators

mukundv@cornell.edu
even at such high quality factors. The true intrinsic ma- [1] M. Li, H. X. Tang, and M. L. Roukes, Nat. Nanotechnol.
2, 114 (2007).
terial limitations of stoichiometric SiN on the quality fac-
[2] M. Poot and H. S. J. van der Zant, Phys. Rep. 511, 273
tor and f Q products of our resonators remain a topic (2012).
for further study. [3] K. C. Schwab and M. L. Roukes, Phys. Today 58, 36
Our current system realizes an enabling platform for (2005).
 5

[4] M. Aspelmeyer and K. C. Schwab, New. J. Phys. 10, and J. Jarzynski, J. Appl. Phys. 101, 013521 (2007).
095001 (2008). [33] I. Wilson-Rae, R. A. Barton, S. S. Verbridge, D. R.
[5] T. J. Kippenberg and K. J. Vahala, Science 321, 1172 Southworth, B. Ilic, H. G. Craighead, and J. M. Parpia,
(2008). Phys. Rev. Lett. 106, 047205 (2011).
[6] M. Aspelmeyer, T. J. Kippenberg, and F. Marquadt, [34] S. K. Steinke, S. Singh, M. E. Tasgin, P. Meystre, K. C.
arXiv:1303.0733 (2013). Schwab, and M. Vengalattore, Phys. Rev. A 84, 023841
[7] F. Marquadt, J. P. Chen, A. A. Clerk, and S. M. Girvin, (2011).
Phys. Rev. Lett. 99, 093902 (2007). [35] M. Bao, H. Yang, H. Yin, and Y. Sun, J. Micromech.
[8] I. Wilson-Rae, N. Nooshi, W. Zwerger, and T. J. Kip- Microeng. 12, 341 (2002).
penberg, Phys. Rev. Lett. 99, 093901 (2007). [36] B. Zwickl, Progress toward observation of radiation pres-
[9] J. D. Teufel, T. Donner, D. Li, J. W. Harlow, M. S. sure shot noise, Ph.D. thesis, Yale (2011).
Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, K. W.
Lehnert, and R. W. Simmonds, Nature 475, 359 (2011).
[10] J. Chan, T. P. Mayer Alegre, A. H. Safavi-Naeini, J. T.
SUPPLEMENTARY INFORMATION
Hill, A. Krause, S. Groblacher, M. Aspelmeyer, and
O. Painter, Nature 478, 89 (2011).
[11] S. S. Verbridge, J. M. Parpia, R. B. Reichenbach, L. M. Sample preparation
Bellan, and H. G. Craighead, J. Appl. Phys. 99, 124304
(2006). The membrane resonators are placed on an aluminum
[12] J. D. Thompson, B. M. Zwickl, A. M. Jayich, F. Mar-
ring approximately 1.5 cm in diameter and glued to this
quadt, S. M. Girvin, and J. G. E. Harris, Nature 452,
72 (2008). ring along the edge of the supporting silicon substrate
[13] D. J. Wilson, C. A. Regal, S. B. Papp, and H. J. Kimble, using an epoxy such as Torr-seal. We take care to use
Phys. Rev. Lett. 103, 207204 (2009). very small amounts of the epoxy and have found that
[14] I. Wilson-Rae, Phys. Rev. B 77, 245418 (2008). excessive use of glue along the edges can reduce the mea-
[15] Through X-ray photoemission spectroscopy, we have ver- sured quality factor. For most of the studies described in
ified that the material composition of the membranes of this work, the silicon substrate makes contact with the
various thicknesses are identical. High resolution images
Aluminum ring only along two corners. The aluminum
of the 30 nm membranes reveal no visible sign of damage.
Further, the measured eigenfreqencies are consistently ring is then glued to the front of a ring-piezo actuator
within 2% of those measured for the thicker membranes and placed within an UHV chamber for our ringdown
of the same width indicating that the tensile stress is of studies.
similar magnitude.
[16] R. Lifshitz and M. I. Roukes, Phys. Rev. B 61, 5600
(2000). Viscous damping limits on Quality factor
[17] D. M. Photiadis, B. H. Houston, X. Liu, J. A. Bucaro,
and M. H. Marcus, Physica B 316, 408 (2002).
[18] A. N. Norris and D. M. Photiadis, Quart. J. Mech. Appl. For the operating background pressures in our studies,
Math. 58, 145 (2005). the limits on the mechanical quality factor due to viscous
[19] C. Seoanez, F. Guinea, and A. H. Castro Neto, Phys. damping from background gas can be estimated as [35]
Rev. B 77, 125107 (2008). s
[20] Q. P. Unterreithmeier, T. Faust, and J. P. Kotthaus,  3/2 kB T 1
Phys. Rev. Lett. 105, 027205 (2010). Qjk (p) = hjk (1)
[21] A. Akhiezer, J. Phys. (Moscow) 1, 277 (1939). 2 mg p
[22] Y. H. Park and K. C. Park, J. Microelectromech. Syst.
13, 238 (2004). where p is the background pressure and m is the molecu-
[23] A. Jockel, M. T. Rakher, M. Korppi, S. Camerer, lar mass of the gas molecules in the background. For our
D. Hunger, M. Mader, and P. Treutlein, Appl. Phys. operating base pressure of p 2 107 , this limit eval-
Lett. 99, 143109 (2011). uates to around Q 109 for modal frequencies around
[24] G. D. Cole, I. Wilson-Rae, K. Werbach, M. R. Vanner,
= 2 500 kHz.
and M. Aspelmeyer, Nat. Commun. 2:231 (2011).
[25] A. H. Nayfeh and M. I. Younis, J. Micromech. MicroEng.
14, 170 (2004).
[26] J. Jackle, Z. Physik 257, 212 (1972). Measurement of optical properties of the SiN
[27] D. Tielburger, R. Merz, R. Ehrenfels, and S. Hunklinger, membranes
Phys. Rev. B 45, 2750 (1992).
[28] P. L. Yu, T. P. Purdy, and C. A. Regal, Phys. Rev. Lett. The imaginary part of the refractive index was esti-
108, 083603 (2012). mated by placing the membrane resonators within a high
[29] S. Chakram and M. Vengalattore, (to be published).
finesse Fabry-Perot cavity and measuring the correspond-
[30] S. Chakram, L. Chang, and M. Vengalattore, (to be
published). ing decrease in the finesse due to absorptive losses in the
[31] D. M. Photiadis and J. A. Judge, Appl. Phys. Lett. 85, membrane. For these measurements, we made use of a
482 (2004). Fabry-Perot cavity characterized by a finesse of 115, 000
[32] J. Judge, D. M. Photiadis, J. F. Vignola, B. H. Houston, corresponding to a cavity linewidth of = 2 70 kHz.
 6

5 Themoelastic damping limits

5
4

4 3
In this section, we describe our model for thermoelas-
tic damping in the high stress membranes. Our model

x10
follows the formalism outlined in [16, 25]. The equation

6
2
3
j

1 of motion for a membrane under tension is

 4
w 4w
 2
w 2w 2w
 
2 0
D + h + = h (2)
x4 y 4 x2 y 2 t2
1
where w(x, y) is the displacement of the membrane in the
1 2 3 4 5
z direction and is assumed to be much smaller than the
k
membrane thickness h 100 nm. The lateral dimension
of the membrane is given by the width L, the membrane
S 1. Parity dependence of the measured quality factors for
low mode indices (j, k 5). tension is denoted by and = 2.7 g/cm3 is the density
of stoichiometric silicon nitride. The flexural rigidity D
is related to the Youngs modulus (E), the thickness h
and the Poissons ration P through the expression
A stoichiometric SiN membrane of thickness h = 200 nm
was placed within this cavity. This choice of membrane Eh3
D= (3)
thickness resulted in a significant modification of the 1 P2
bare cavity parameters due to the optical losses within
the membrane. Perhaps more importantly, the mem- In the presence of thermoelastic damping, the local
brane reflectivity for this particular thickness is close to temperature variations within the membrane and the
zero at our operating wavelength of = 795 nm, (i.e. resulting thermal expansion lead to additional stresses.
2R(n)h/ r 0). This ensures a minimal de- The equation of motion in the presence of thermal strain
pendence of the modified cavity finesse on the alignment is given by [25]
and position of the membrane with respect to the intra-
2w
cavity field. D4 w h2 w + h = 2 M T N T 2 w (4)
t2
Over a range of positions of the membrane within the
Fabry-Perot cavity, we measured a modified finesse of where the thermal axial force and the thermal bending
70, 000 corresponding to a cavity linewidth of = 2 moment are respectively given by
105 kHz. Under the assumption that this modification E
Z
is entirely due to absorptive losses within the membrane, NT = dz (5)
1 P
our measurements are consistent with I(n) < 9 106 .
E
Z
MT = zdz (6)
1 P
where (x, y, z, t) = T (x, y, z, t)T0 is the local deviation
in temperature, T0 is the equilibrium temperature and
Parity dependence of Quality factors for low mode
is the coefficient of thermal expansion. The local tem-
indices
perature field itself satisfies the diffusion equation given
by
For low mode indices, we observe that the measured
quality factors can show large variations depending on ET0
2 = CP (z2 w) (7)
the parity of the modes (see Fig. S1). Also, in contrast t 1 P t
to our observations of the modes in the plateau regime,
where is the thermal conductivity and CP is the specific
we find that the modal structures for these low mode in-
heat. The coupled equations (1),(6) can be solved for the
dices are negligibly influenced by the substrate. Thus,
normal modes of vibration in the presence of thermoelas-
the phonon tunneling model as described in [14, 33] ac-
tic damping. The quality factor can then be extracted
curately predicts our measured quality factors in this
from the corresponding eigenfrequencies as
regime. In contrast, as the mode indices grow larger, in-
terferences between degenerate eigenmodes arising from I()
coupling to the substrate result in a suppression of the Q1 () = 2 (8)
R()
large parity-dependent variation of the quality factors
and a more robust occurence of a large number of high-Q We solve the diffusion equation for the local tempera-
modes. ture field with boundary conditions corresponding to no
 7

heat flow at the top and bottom surfaces. This yields Modal image analysis and estimate of local curvature
!
ET0 sin(kz)
(x, y) = 2 w(x, y) z Here, we describe our method of estimating the local
(1 P )CP k cos(kh/2) curvature from the interferometric images of the various
q (9) resonator modes. To image a particular mode, the mem-
where k = (1 i) C
2 . Substituting this solution into
P brane is driven on resonance to an amplitude of around
the expressions for the thermal strain and bending mo- 200 pm. An imaging beam of 3.5 mm waist and a power
ment results in a modified equation for the membrane of 10 W is incident on the membrane. The DC compo-
motion nent of the reflected light is filtered by means of a 200
m spot and the resulting dark field image is captured
(D + Dt )4 w0 h2 w0 = h 2 w0 (10) on a CCD camera with typical exposure times of 1 ms.
where we have written w(x, y, t) = w0 (x, y)eit and The amplitude of motion is calibrated with a Michelson
! interferometer with a smaller spot size focused at the
E 2 2 T0 h3 h 2 tan(kh/2) center of the membrane. In addition to each modal im-
Dt = + (11)
(1 P )2 CP 12 k 2 k 3 age (I(x, y)), images are also obtained of the background,
i.e. of the undriven membrane (Ibkgd ) and the laser beam
is the correction due to thermoelasticity. The damping of profile (Ibeam ).
the membrane is encapsulated in the non-zero imaginary The image proportional to the modulus of the displace-
part of this term. Solving the equation of motion for the ment is given by
normal modes yields the eigenfrequencies
s
I(x, y) Ibkgd (x, y)
s
2 2 (D + Dt ) 2 |w(x, y)| = (18)
mn = 2
(m + n2 ) + (m + n2 )2 4 Ibeam (x, y)
L hL4
(12)
The local curvature C(x, y) is obtained by taking the
In the case of large tensile stress, i.e. in the regime where
D finite differences from the 266266 pixel image according
hL 2 1, the expression for the eigenfrequencies
to the expression
reduces to
2 (D + Dt ) 2
 
0 2 w[i + d, j] + w[i d, j] 2w[i, j]
mn = mn 1 + (m + n ) (13) C(x, y) C[i, j] =
2hL2 d2
q w[i, j + d] + w[i, j d] 2w[i, j]
where mn 0
= 2 4L 2 2
2 (m + n ). The quality factor is
+ (19)
d2
then given by
where an optimal value of d = 5 pix was used as a com-
2 (m2 + n2 ) promise between signal to noise and spatial resolution.
Q1
mn I[Dt ] (14)
hL2 As noted earlier, an intrinsic loss mechanism would
where contribute terms in the equation of motion that, to lead-
E 2 2 T0 h3 ing order, are proportional to the local curvature of the
I[Dt ] = g() (15) displacement field. Accordingly, the energy loss in the
12(1 P )2 CP
membrane due to such mechanisms is given by
6 sinh + sin 6
g() = 3 2 (16) 2
cosh + cos Z 
2w 2w
E dxdy + (20)
and = h(CP /2)1/2 . Written in terms of the un- x2 y 2
0
damped natural eigenfrequency mn of the resonator, the
thermoelastic damping limit on the quality factor finally while the stored energy in the membrane can be writ-
ten as E = 21 hjk
2
dxdy w(x, y)2 . The quality factor
R
reduces to
associated with such an intrinsic loss mechanism is then
(1 P )2 CP 2 1
Q(mn ) = 12 0 )2 E 2 2 T h2 g()
(17) given by
(mn 0
dxdy w(x, y)2
R
Using bulk material parameters for stoichiometric Silicon 2E
Q= R 2
nitride and for a typical membrane thickness of h = 100 E

2 2
dxdy xw2 + yw2
nm, this limit evaluates to QT ED 1012 for modal fre-
2
P
quencies around = 2 500 kHz. This estimate is i,j w[i, j]
(21)
consistent with that derived by Zwickl et al [36], but sig- 2
P
i,j C[i, j]
nificantly higher than the estimate in Wilson et al [13]
which neglects the large intrinsic stress of the SiN mem- By construction, the above quantity does not depend on
branes. the magnitude of the membrane displacement.
 8

onator modes that share the same symmetry, i.e. res-

onator modes with even parity only couple to substrate
200 modes with even parity etc. In the case of the single-
mode calculation, this consideration results in distin-
150
guishing among three distinct cases, (i) Symmetric modes
(j, k are odd), (ii) Antisymmetric modes (j, k are even)

pm
100
and (iii) Mixed symmetry modes (j even, k odd or vice
50 versa). Accordingly, this results in separate expressions
of the anchor-limited quality factors with different angu-
0
lar dependences.
In contrast, the hybridized modes exhibit highly sym-
S 2. Interferometric images of the modal structures resulting
from substrate-mediated coupling. Left: Hybridized mode
metric modal structures that is less dependent on the
corresponding to 3,11 11,3 , Right: Hybridized mode cor- parity of the mode indices (see, for example, Fig. S2).
responding to 10,6 + 6,10 . This results in a suppression of the radiated substrate
modes and a concomitant enhancement of the quality
factor for a larger class of resonator modes.
Anchor loss model for hybridized resonator modes The full calculation that takes this enhanced symme-
try into account, while laborious, is fairly straightforward
p
and rapidly convergent in the limit j, k > 2 2
In this section, we outline our analysis of the anchor p j + k .
2
The results of this calculation for modes j + k = 12 2
loss model that takes into account the interference effects
due to substrate-mediated hybridization. The phonon are shown () in Fig. 4(b). Also shown alongside is the
tunneling model introduced in [14] permits a general calculation that does not take into account any modal hy-
expression for the clamping-limited quality factor Q in bridization. The contrast between the two predictions is
terms of the modal overlap between the resonator and especially stark for modes of low symmetry. The single-
the free substrate mode calculation predicts a large variation of Q with
Z Z 2 alternating parity while the modified calculation sup-
1 
~ 0
 presses this variation resulting in uniformly higher Qs.
Q (R ) = 3

dS R ~uq  (R (q))
2s R R q S  For
p large mode indices j, k 1 with j, k >
(22) j 2 + k 2 , the suppressed parity dependence can be
where s (R ) are the mass densities of the substrate (res- well approximated by the asymptotic expression
onator), R is the resonant frequency, R is the stress
16j 2 k 2 1
 
field of the resonator and ~u0q is the displacement field of 1 1 R h X 3 (0) 2
Qkj,asympt p + 4 |u |
the free substrate modes. j 2 + k2 j 4 k s L
The large intrinsic stress of the stoichiometric SiN and (24)
the extremely small thickness of the membrane allow us where the summation is now carried out over the var-
to relate the stress field at the periphery to the slope of ious modes (surface acoustic, transverse, longitudinal
the resonator mode evaluated at the clamp [33], i.e. and flexural waves) supported by the substrate. Here,
= cR /c 1 is the ratio of phase velocities in the
Z R
R xdz = h
z (23) membrane and the substrate. Note that, in this limit,
x the summation evaluates to a geometry-independent con-
stant and the entire modal dependence is contained in the
and a similar relation for y. Here, is the intrinsic stress
pre-factor. For comparison, this approximate expression
of the silicon nitride membrane and R (x, y) is the res-
is plotted alongside the exact calculation in Fig. 4(b).
onator eigenmode. For the square membranes, this ap-
Intuitively, hybridization between degenerate res-
proximation results in
onator modes results in modal structures that are more
Z Z R symmetric than either of the constituent modes (see
S dS~
~u0q = 4 h dx ~u(x, y = L/2) |y=L/2
S y Fig. S2). This enhanced symmetry results in two obvi-
Z R ous trends that are captured in Fig. 4(b), i.e. (i) An in-
+ 4 h dy ~u(x = L/2, y) |x=L/2 creased quality factor due to destructive interference of
x
substrate modes radiated from the evenly spaced antin-
Symmetry considerations imply that contributions to odal segments around the clamp and (ii) A suppressed
the surface integral only arise from substrate and res- dependence of Q on the parity of the resonator mode.