Cryogenic Silicon Cavities: Metrology & Photonics
- Cryogenic silicon cavities are optical resonators operated at cryogenic temperatures, using silicon as a spacer or integrated medium to minimize thermal expansion and noise.
- They encompass diverse architectures such as bulk single-crystal Fabry–Pérot, hybrid silicon–sapphire, and integrated silicon photonic devices, each optimized for high finesse and stable frequency response.
- Advancements address challenges like coating-related excess noise, bonding reliability, and thermal transport, enabling improved laser stabilization, gravitational-wave metrology, and quantum photonics.
Cryogenic silicon cavities are optical resonators in which silicon serves as the spacer, mirror-substrate material, or integrated photonic cavity medium and the device is operated at cryogenic temperature to reduce thermal expansion, Brownian noise, thermo-optic shifts, and absorption-induced heating. The topic spans single-crystal Fabry–Pérot reference cavities for laser frequency stabilization, hybrid bonded cavities in which silicon is the structural reference element, and integrated silicon photonic resonators operated in enclosed helium or superfluid helium environments. Across these implementations, the central design objective is the same: to make cavity frequency or optical response less sensitive to thermal, mechanical, and materials-dissipation processes than is possible at room temperature (Wiens et al., 2014, Kedar et al., 2022, Wasserman et al., 2022, Sun et al., 2013).
1. Platform classes and cavity architectures
The field includes at least three experimentally distinct cavity classes. First are bulk single-crystal Fabry–Pérot reference cavities, where silicon defines the resonator length directly. A representative example is a horizontally mounted cavity made from dislocation-free float-zone single-crystal silicon, with a cylindrical spacer $250$ mm long and $70$ mm in diameter, a $15$ mm central bore, one flat mirror, one $1$ m-radius mirror, and mirror substrates optically contacted to the spacer with the same crystal orientation. At $1.5$ K, that cavity exhibited finesse and linewidth $3$ kHz (Wiens et al., 2014).
A second class uses silicon as the bulk cryogenic reference structure but combines it with non-silicon optical substrates. In a hybrid silicon–sapphire Fabry–Perot cavity, the spacer is a $60$ mm long bulk silicon cuboid with cross section $25.4$ mm $70$0 mm, axial bore $70$1 mm, and vent hole $70$2 mm, while the end mirrors are sapphire substrates carrying multilayer $70$3 coatings. That system is relevant because it shows that silicon can remain the mechanically dominant cavity element even when optical or bonding constraints motivate a hybrid assembly (Sun et al., 2018).
A third class consists of integrated silicon photonic cavities on SOI. Two representative devices are especially important. One is a wavelength-scale silicon disk nanocavity embedded in a sunflower-type circular photonic crystal on 220 nm silicon / 3 $70$4m buried oxide SOI, with measured $70$5 in the $70$6 to $70$7 range depending on temperature and helium environment. The other is a silicon-on-insulator whispering-gallery-mode slot resonator on a $70$8 mm $70$9 $15$0 mm chip with layer stack 220 nm Si / 2 $15$1m $15$2 / 500 $15$3m Si; its resonator consists of two 180 nm-wide annuli separated by a 100 nm slot, outer radius $15$4m, and measured optical $15$5. In the latter, the optical mode is predominantly confined in the slot, which is precisely the geometry needed for strong interaction with condensed helium in the optical near field (Sun et al., 2013, Wasserman et al., 2022).
A further experimental branch combines silicon spacers and substrates with substrate-transferred semiconductor crystalline coatings. In the 6 cm Si6 cavity and 21 cm Si5 cavity, both spacer and mirror substrates are silicon, while the coatings are $15$6. Si6 operated mainly at $15$7 K and $15$8 K with finesse $15$9, linewidth $1$0 kHz, and birefringent polarization splitting $1$1 kHz; Si5 operated at $1$2 K with finesse $1$3 and birefringent splitting $1$4 kHz (Kedar et al., 2022).
A compact comparison is useful because “cryogenic silicon cavity” refers to a family rather than a single architecture.
| Platform | Silicon role | Representative cryogenic result |
|---|---|---|
| Single-crystal Fabry–Pérot | Spacer and mirror substrates | $1$5 finesse at $1$6 K (Wiens et al., 2014) |
| Hybrid silicon–sapphire Fabry–Pérot | Silicon spacer | Repeated cycling to $1$7 K with finesse $1$8 (Sun et al., 2018) |
| Crystalline-coated silicon Fabry–Pérot | Spacer, substrates, crystalline coatings | Dual-polarization residual $1$9 for Si6 (Kedar et al., 2022) |
| Integrated SOI helium-coupled cavity | Silicon photonic resonator | $1.5$0 nm resonance shift from superfluid loading (Wasserman et al., 2022) |
| Integrated SOI nanocavity in LHe | Silicon disk nanocavity | $1.5$1 intracavity photons at $1.5$2 K (Sun et al., 2013) |
2. Operating temperatures, thermal expansion, and cryogenic setpoints
The thermodynamic operating point is fundamental because cavity frequency obeys the inverse-length relation
$1.5$3
so the linear thermal expansion coefficient
$1.5$4
sets first-order temperature sensitivity directly (Wiens et al., 2014).
For bulk single-crystal silicon cavities, two experimentally important temperature regions are emphasized. One is the low-temperature zero crossing near $1.5$5 K. In the 250 mm silicon resonator, a frequency minimum occurred at $1.5$6 K, with measured $1.5$7 at $1.5$8 K and
$1.5$9
at 0 K. The small derivative near this low-temperature zero crossing is important because it suppresses residual second-order temperature sensitivity. The same work reported resonator temperature stabilization below 1K for averaging times longer than about 2 s and concluded that the system should enable laser stabilization in the low-3 range (Wiens et al., 2014).
A second widely used setpoint is near 4–5 K. In cryogenic gravitational-wave instrumentation, 6 K is selected because silicon’s coefficient of linear thermal expansion crosses zero there and because the temperature is accessible with comparatively simple cryogenic infrastructure. A low-vibration suspended optical test cavity containing a crystalline silicon cantilever was radiatively cooled to 7 K in 8 hours and held there over many months with 9 mK relative stability (Kapasi et al., 2024). In the Si5 crystalline-coated silicon reference cavity, operation at $3$0 K provided the higher-temperature comparison point for cryogenic coating-noise studies (Kedar et al., 2022).
Below $3$1 K, the sub-kelvin regime becomes attractive because silicon’s CTE becomes extremely small and the thermal-noise floor is projected to fall further. A 180 mm single-crystal silicon Fabry–Perot cavity being developed for dilution-refrigerator operation uses silicon mirror substrates with crystalline $3$2 coatings and projects thermal-noise contributions of $3$3 at $3$4 K, $3$5 at $3$6 K, and $3$7 at $3$8 K. The same study states that, assuming peak-to-peak temperature fluctuations of $3$9 mK, the CTE must be below $60$0 to reach the $60$1 level (Barbarat et al., 2024).
The low-temperature CTE description is not fully settled below $60$2 K. The sub-kelvin study quotes two different low-temperature CTE models. The Lyon model gives
$60$3
while a Wiens-based fit yields $60$4 at $60$5 K, $60$6 at $60$7 K, and $60$8 at $60$9 K. This establishes a genuine materials-modeling uncertainty for sub-kelvin cavity optimization rather than a settled design constant (Barbarat et al., 2024).
A more speculative extension places the operating point near $25.4$0 K in a lunar permanently shadowed region. That proposal uses a cryogenic monolithic silicon cavity behind nested shields, with the cavity chamber at about $25.4$1–$25.4$2 K, an active shield at about $25.4$3 K, and the cavity itself stabilized at approximately $25.4$4 K. The underlying logic is the same as in terrestrial cryogenic silicon metrology, but the environmental implementation is distinct (Ye et al., 6 Feb 2026).
3. Frequency stability, thermal noise, and excess noise processes
In precision metrology, the principal attraction of cryogenic silicon cavities is reduced Brownian thermal noise together with reduced temperature sensitivity. In the 250 mm single-crystal silicon resonator, absolute frequency drift was less than $25.4$5 Hz over $25.4$6 hour, and absolute frequency instability reached about $25.4$7 Hz for integration times longer than $25.4$8 s. Under the assumption of coating loss angle $25.4$9 mrad, the expected Brownian-noise-induced fractional frequency instability at 0 K was 1, and the system was projected to support laser stabilization below 2 for averaging times longer than 3 s (Wiens et al., 2014).
That same study also showed that cryogenic performance is not determined by Brownian noise alone. The measured upper bound on cavity frequency sensitivity to laser power was 4 Hz for a 5 power change at 6 K and 7 K, corresponding to 8W, whereas finite-element modeling predicted 9W of total dissipated power and a thermal equilibration time of about $70$00 s. This difference did not contradict the model; rather, it indicated that the experiment did not resolve the smaller expected effect directly (Wiens et al., 2014).
Crystalline-coated cryogenic silicon cavities altered this picture by reducing the expected coating loss but revealing a previously unmeasured excess-noise channel. In Si6 and Si5, the measured total noise was decomposed phenomenologically as
$70$01
The intended advantage came from the low room-temperature measured mechanical loss angle of $70$02,
$70$03
which is stated to be more than an order of magnitude lower than that of amorphous $70$04 coatings at cryogenic temperature. Assuming that room-temperature loss angle persisted at cryogenic temperature, the projected thermal-noise-limited fractional instability for Si6 at $70$05 K was about $70$06. Experimentally, however, single-polarization interrogation was dominated by birefringent noise rather than by the expected Brownian floor (Kedar et al., 2022).
The crucial observation was that fluctuations of the two polarization eigenmodes were anti-correlated. Dual-polarization interrogation therefore suppressed the birefringent term. In the $70$07–$70$08 Hz band, the residual-plus-Brownian noise measured for Si6 was
$70$09
which was slightly below the previously measured thermal-noise-limited instability of the equivalent dielectric-coated 6 cm cryogenic silicon cavity, quoted as
$70$10
The long-term residual instability was $70$11 for Si6 and $70$12 for Si5 (Kedar et al., 2022).
The birefringent excess noise was not a small perturbation. It showed a phenomenological empirical dependence
$70$13
where $70$14 is intracavity circulating power, and a mode-area scaling consistent with the $70$15 Vinet factor $70$16; experimentally,
$70$17
Equally important, the single-polarization birefringent spectrum was essentially unchanged between $70$18 K and $70$19 K. This argues against a simple thermal mechanism with strong temperature dependence across that interval (Kedar et al., 2022).
Sub-kelvin development work makes clear that further cooling does not automatically translate into better realized instability. Although the projected thermal-noise contribution drops to $70$20 at $70$21 mK, the same analysis assigns a temperature-independent “Global noise” term of $70$22 at $70$23 K, $70$24 K, and $70$25 K, based on prior observations in cryogenic AlGaAs-coated cavities. A central misconception is therefore corrected by the experimental record: lower temperature lowers the Brownian floor, but coating-related birefringent and residual excess-noise mechanisms can remain dominant unless they are separately mitigated (Barbarat et al., 2024).
4. Thermal transport, thermalization, and helium-coupled regimes
For cryogenic silicon structures below $70$26 K, heat flow through narrow supports is neither generically diffusive nor generically ballistic. Measurements on $70$27m-thick single-crystal silicon beams showed that the relevant control parameter is $70$28, where $70$29 is beam length and $70$30 is phonon mean free path set by boundary scattering. In long beams with $70$31, conductance scales approximately as $70$32 and becomes highly sensitive to roughness and residue; in short beams with $70$33, conductance is ballistic-dominated and mainly set by beam cross-sectional area. The paper states a simple interpolation,
$70$34
and reports $70$35 fractional deviation for ballistic-beam-dominated devices, while nominally similar long beams could differ in conductance by about a factor of $70$36 depending on fabrication route and surface state (Rostem et al., 2014).
This result matters because cryogenic cavity thermalization is often limited not by the bulk silicon spacer but by the support geometry. The same study showed that even $70$37–$70$38 nm rms roughness can matter below $70$39 K, despite a thermal wavelength of $70$40m at $70$41 mK, and reported resonant phonon scattering in beams with pitted surfaces of about $70$42 nm depth and $70$43 nm lateral scale. It also found no evidence for excess specific heat in single-crystal silicon membranes, allowing thermal mass to be engineered with deposited metals such as a 400 nm Pd layer giving about $70$44 (Rostem et al., 2014).
At the system level, thermalization and vibration isolation are strongly coupled. A cryogenic low-vibration test cavity designed for gravitational-wave instrumentation cooled a suspended optical cavity inside approximately $70$45 litres of radiation-shield volume to $70$46 K in $70$47 hours and held it at that temperature over many months with $70$48 mK stability. The thermal design used a pulse-tube cryocooler, flexible OFHC copper braid, a large copper thermal reservoir, and nested radiation shields; conductive heat transport was described by
$70$49
while radiative exchange was described with the Stefan–Boltzmann form
$70$50
A notable systems result was that after compressor switch-off, the coldhead took about $70$51 hours to warm back to $70$52 K, permitting cryocooler-off low-noise measurement windows (Kapasi et al., 2024).
Integrated silicon nanocavities show a different thermal regime when immersed directly in liquid helium. In a pedestal-supported silicon disk nanocavity, operation in superfluid helium suppressed the silicon thermo-optic coefficient from $70$53 at $70$54 K to less than $70$55 around $70$56 K, while reducing the free-carrier lifetime from $70$57 ns at $70$58 K to $70$59 ns at $70$60 K and $70$61 ns at $70$62 K. Near the helium lambda point, the transmission spectra exhibited blue-shifted bistability because free-carrier dispersion dominated once thermo-optic red shifting had collapsed. At $70$63 K, the spectra returned to nearly symmetric Lorentzian lineshapes, and the cavity supported about $70$64 stored photons, ultimately limited by a local helium phase transition (Sun et al., 2013).
The helium-assisted thermalization concept also appears at the packaging level. A hermetically sealed cryogenic photonic package was explicitly designed so that enclosed helium could be “used as a heat-exchange gas to thermalize photonic devices, or condensed into a superfluid covering the device.” The paper did not report a direct heat-transport benchmark, but it established a framework for engineering superfluid film thickness through enclosed-gas loading and device height above a liquid reservoir (Wasserman et al., 2022).
5. Packaging, bonding, optical coupling, and cryogenic assembly
Cryogenic silicon cavities are often limited by packaging and bonding rather than by nominal cavity design. One route is hydroxide catalysis bonding with controlled thermal oxide. In silicon–sapphire bonds, natural oxide of about $70$65 nm was inadequate: two natural-oxide pairs cooled from room temperature to $70$66 K both broke during one thermal cycle, and surviving natural-oxide samples exhibited shear strength $70$67 MPa. By contrast, artificial thermal oxide layers from $70$68 nm to $70$69 nm survived the reported cryogenic cycles, and bonds with $70$70 nm oxide showed $70$71 MPa shear strength after cooling. The hybrid cavity built with about $70$72 nm oxide on each spacer end survived three thermal cycles, including cooling to $70$73 K, while maintaining good optical finesse (Sun et al., 2018).
The optical data from that hybrid cavity are useful because they show that a cryogenic bond process need not catastrophically degrade cavity optics. Cavity ringdown measurements gave finesse $70$74 at $70$75 K immediately after assembly, $70$76 at $70$77 K, $70$78 at $70$79 K, and $70$80 again after further cycling. The ringdown relations were given as
$70$81
This established repeated cryogenic survivability for a silicon-centered bonded cavity rather than merely for bonded witness samples (Sun et al., 2018).
For integrated photonic cavities, the packaging problem is broader: fiber coupling, sealing, repeated thermocycling, and helium retention must all be solved together. A brass “puck” package was developed as a hermetic, gas-tight, and superfluid-tight enclosure. After fiber attachment, the chip was mounted to the brass base with Apiezon N cryogenic vacuum grease for both retention and thermal contact; the two brass halves were then sealed with Stycast 2850FT epoxy and catalyst type 9. Gas filling used an oxygen-free copper pinch-off tube of copper CDA 101, and the measured leak rate was below $70$82. The filling sequence was evacuation, room-temperature backfill to the desired helium pressure, and cold-weld pinch-off with a hand-operated hydraulic crimp tool (Wasserman et al., 2022).
That work also provided unusually detailed quantitative guidance on fiber attachment. Across 690 discrete bond stress tests involving silicon, thermal oxide, and HSQ substrates and the glues NOA 61, NOA 68, NOA 88, NOA 86H, and cyanoacrylate superglue, NOA 86H gave the best survival performance on all substrate types. The authors attributed its advantage to near-complete polymerization enabled by thermal post-cure at $70$83C for $70$84 minutes. After switching to NOA 86H, they state that they had not observed failures in later devices cooled to $70$85 mK (Wasserman et al., 2022).
Optical coupling in that package was implemented either with angle-polished fibers and grating couplers or with a silica tapered-fiber adiabatic coupler. At cryogenic temperature, grating couplers yielded about $70$86 coupling efficiency and $70$87 nm bandwidth; the adiabatic coupler yielded about $70$88 coupling efficiency and $70$89 nm bandwidth. The paper explicitly states that in both cases the efficiency did not degrade upon cooling to cryogenic temperatures. In the actual superfluid demonstration, filling the package with $70$90 mbar of $70$91He at room temperature produced a $70$92 nm whispering-gallery resonance shift on cooling, interpreted as slot filling with superfluid and surrounding film thickness of $70$93 nm (Wasserman et al., 2022).
The same paper also gave the main working formulas for helium-film design. In the unsaturated vapor-pressure regime,
$70$94
while in the saturated-vapor-pressure regime,
$70$95
Using $70$96 for silica and $70$97, the authors identified practical film-thickness ranges of about $70$98–$70$99 nm for downward pinch-off geometry with $15$00–10 cm and up to about $15$01 nm for millimeter-scale $15$02 in the opposite orientation (Wasserman et al., 2022).
Sub-kelvin Fabry–Perot development reinforces the same systems lesson from a different direction. In a dilution-refrigerator cavity intended for operation below $15$03 K, the initial three-point support implied negligible conductive cooling, and radiation-only cooldown from $15$04 K to the final low temperature was estimated to take a few hundred days. The mitigation was a copper thermal support with flexible copper braids and a copper foil wrap contacting the cavity, explicitly balancing high thermal conductance against low vibration coupling (Barbarat et al., 2024).
6. Applications, limits, and future directions
Cryogenic silicon cavities already serve several distinct research agendas. In precision frequency metrology, the single-crystal silicon Fabry–Perot platform was developed as an optical frequency reference whose dimensional stability transfers directly to a stabilized laser. The applications explicitly identified include optical clocks, optical frequency combs, low-phase-noise microwave generation, precision spectroscopy, and tests of Lorentz invariance and possible spacetime effects (Wiens et al., 2014).
In integrated photonics and optomechanics, the emphasis is different. Silicon cavities operated in enclosed helium or immersed in liquid helium couple strongly to refractive loading, acoustic modes of the fluid, and enhanced thermalization pathways. The hermetically sealed SOI slot-resonator package was positioned as relevant to superfluid optomechanics, quantum photonics, and quantum optomechanics, and it demonstrated MHz-frequency superfluid acoustic modes driven by a blue-detuned laser via radiation pressure and fountain pressure and read out optically by heterodyne detection (Wasserman et al., 2022). The ultrahigh-$15$05 silicon disk nanocavity in superfluid helium showed that the dominant room-temperature limitation of silicon nanophotonics—thermo-optic and carrier-mediated nonlinear distortion—can be displaced by a different limit set by local helium phase transition (Sun et al., 2013).
In gravitational-wave instrumentation and cryogenic mechanics, silicon cavity technology is as much about the surrounding infrastructure as about the optical resonator itself. A low-vibration cryogenic facility measured displacement noise from a gram-scale silicon cantilever at the level of $15$06 at $15$07 kHz and achieved a facility noise floor of $15$08 above $15$09 Hz. The stated purpose was to guide the design of suspensions in future cryogenic detectors such as LIGO Voyager and the Einstein Telescope, and the facility was also described as suitable for testing new mirror coatings at cryogenic temperatures (Kapasi et al., 2024).
Several limitations recur across the literature. Cooling alone does not guarantee state-of-the-art realized stability. Hybrid bonded cavities show that survivability and finesse retention are possible, but they do not directly establish thermal-noise-limited performance (Sun et al., 2018). Crystalline coatings promise lower Brownian noise, yet birefringent and residual excess noise can dominate unless dual-polarization interrogation and further materials understanding are brought to bear (Kedar et al., 2022). Sub-kelvin cavities promise projected thermal-noise floors in the $15$10 range, but present analyses still assign dominant total instability to unresolved excess noise and to vibration and thermalization challenges (Barbarat et al., 2024). In integrated helium-coupled systems, packaging and adhesive survival can be a more immediate bottleneck than cavity $15$11 itself (Wasserman et al., 2022).
A final direction is extrapolative rather than demonstrated. A proposed lunar silicon cavity places a cryogenic monolithic silicon cavity in a permanently shadowed lunar region, using passive radiative cooling and the local low-vibration, ultra-high-vacuum environment to target thermal-noise floors of $15$12 for a 21 cm cavity with conventional coatings, $15$13 for a 21 cm cavity with crystalline coatings, $15$14 for a 50 cm cavity with conventional coatings, and $15$15 for a 50 cm cavity with crystalline coatings. The proposal explicitly identifies dust, ionizing radiation, and micrometeorites as risk factors requiring mitigation, so it is best understood as a systems-level extension of cryogenic silicon cavity logic rather than an experimental result (Ye et al., 6 Feb 2026).
Taken together, these results establish cryogenic silicon cavities as a broad technological category rather than a single device type. Their defining technical themes are operation near favorable silicon CTE setpoints, control of coating and interface dissipation, packaging that preserves alignment and thermalization through deep cooldown, and continual negotiation between thermal conductance and vibration isolation. The literature also shows that the limiting physics has shifted: in many mature platforms, the decisive obstacles are no longer the basic availability of cryogenic silicon or high finesse, but excess coating noise, helium-interface effects, bonding reliability, and the thermal-mechanical behavior of the complete cryogenic assembly.