Cryogenic Optical Packaging Using Photonic Wire Bonds

We present the required techniques for the successful low loss packaging of integrated photonic devices capable of operating down to 970 mK utilizing photonic wire bonds. This scalable technique is shown to have an insertion loss of less than 2 dB per connection between a SMF-28 single mode fibre and a silicon photonic chip at these temperatures. This technique has shown robustness to thermal cycling and is ultra-high vacuum compatible without the need for any active alignment.

It is desirable to measure the optical absorption of the Vancore A resist used in the PWB process.This information will allow for the determination of the suitability of PWB for coupling to various quantum emitters.Fourier-transform infrared spectroscopy (FTIR) was used for this purpose.The use of FTIR allows for the entire wavelength range between 1000nm and 5500nm to be studied with the detectors we have available.This range covers all the emitters of interest in a silicon on insulator platform.[1][2][3][4].
Our Bruker Vertex 80 FTIR has been customized with a liquid nitrogen cooled germanium photodetector.This detector allows for the detection of extremely weak signals in the 800nm to 1800nm wavelength range.Unfortunately, this detector occupies the FTIR's sample compartment, making it impossible to place samples in this location.As an alternative, the samples were placed in the detector compartment as shown in Fig. S1.

Fig. S1.
A resist sample mounted in the FTIR.The sample is mounted before the detector.A 3D printed holder allows for the samples to be quickly exchanged.Cardboard blocks stray light from the internal light source, maximizing the signal to noise ratio obtained.
A diagram of the optical path with this measurement configuration is shown in Fig. S2.
To avoid any interference from the choice of substrate a sample of Vancore A resist was prepared on a metal grid by placing the resist on the grid and hardening it with the DynaCure DUV LED for 180 seconds on both sides.The metal grid can be seen in Fig. S3.
The high absorption for wavelengths longer then 2000nm means that PWB is not suitable for MIR defect centres such as selenium at 2902nm [4].The high absorption in the 2870 nm to 3435 Fig. S2.A simplified diagram of the optical path of the FTIR.Light from the internal source passes through the interferometer then the resist sample indicated by the green line before reaching the detector.nm region is from OH and CH bonds.It is unlikely this is able to be overcome using a different resist as all organic photoresists will have these bonds [5][6][7].

A. Choice of Glue
Samples that were assembled using vacuum compatible UV epoxy (Masterbond® UV10TKLO-2 [8]) did not survive cooldown.Although it is desirable to use UV epoxy that can cure within seconds using UV light during the assembly process, it is observed that even with a relatively slow cooling process (around 4 hours), this epoxy still suffers macroscopic cracking at low temperatures [Fig 4(d)].This could be due to internal thermal stress within the epoxy and from shear or compression stress between the different materials.This damage made it prone to detach completely after being cooled.As a result, a cryo-and vacuum-rated two-part epoxy (Masterbond® EP29LPSPAO-1 Black [9]) is used instead for high mechanical and dimensional stability.The epoxy is also thermally conductive which helps the sample maintain good thermal contact with the sub-mount and consequently the cold stage.Additionally, reducing the amount of epoxy used and maintaining an even spread of epoxy on both sides of the FA and chip was found to help prevent mechanical failure.To ensure the sample remains aligned during the epoxy's multi-hour curing process, small amounts of UV glue (Bondic® [10]) is first used to secure the aligned components onto the sub-mount.Once the epoxy has fully cured, this UV glue is easily and cleanly removed using a scalpel.

B. Thermal Contraction Matching and Stress Management
The cryogenic PWB assembly technique is modified from the room temperature technique to achieve successful operation down to cryogenic temperatures.Normally, at room temperature, it is ideal to aim for shorter photonic wire bonds with a minimal number of bends and minimum bending radius (> 80 µm) for low insertion loss (demonstrated by solid bond in Fig. S5(a)).An etched silicon shim is used to raise the height of the FA such that the fiber cores are roughly 20 µm above the chip surface.The FA face is also butted against the chip edge to ease the assembly process and prevent any trapping of unexposed resist in the crevice between the FA and chip edge.However, this configuration is not suitable when the sample, which includes the silicon photonic Fig. S3.A photo of the mesh grid with a sample of photonic wirebond resist cured inside.chip and shim, copper sub-mount, and polymer-based PWB, undergoes thermal contraction during cooldown.Polymer experiences the highest amount of thermal contraction (∆L/L = 1.22 %) from 293 K to 4 K followed by copper (∆L/L = 0.324 %) and silicon (∆L/L = 0.022 %) [11,12].Optical images of the failed samples show that PWBs with this configuration experience high tensile stress that resulted in the bonds being peeled off from the surface taper and thus failing.This suggests there is both a vertical and horizontal stress is present during cool down.Subsequently, a series bonds with slack added (demonstrated by the dashed bond in Fig. S5(a)) was written to compensate for the thermal contraction.Here, slack is added to the bond by increasing the "height" of the middle bond section and the amount of slack is quantified by the percentage of length difference compared to a straight line.The slack was varied from 0.63 % to 3.65 %.However, the bonds continue to peel away and detach from the surface taper with no noticeable improvements even for the bonds with the highest amount of slack [Fig.S5(b)].

C. Choice of Sub-mount and Shim
As described in section B the samples that utilize silicon shims between the copper sub-mount, the FA and the chip to achieve optimal height matching did not survive cooling.Both configurations, with and without a gap between the FA and the chip, experienced failed bonds.This suggests that silicon is not a suitable shim material as its relatively low coefficient of thermal expansion (CTE) does not match the high CTE of the polymer bonds.It is also not a good thermal conductor.As a result, the FA and chip is glued directly to the copper sub-mount and no shim is used.

EXPERIMENTAL SETUP
The gold plated copper sub-mount of the sample is bolted to the gold-plated copper cold stage by two stainless steel screws.Any excess fiber is carefully wrapped and secured to the radiation shield using Kapton™ tape and separated from touching the wall of the outer vacuum chamber.Any bending of the fibers is minimized and the targeted minimum bend radius is larger then 30 mm.
Coiling and anchoring the excess fiber to the radiation shield can create torsion and bending that introduces unwanted birefringence [13,14].In addition, the fiber is subject to refractive index change during cooling as the radiation shield reaches down to 40 K.This can induce polarization dependent loss in the measurement as the polarization of the laser light becomes random when travelled through the fiber and which causes a mode mismatch between the PWB and the polarization sensitive surface taper.Although a polarization controller can be used to optimize the polarization during the whole process, it is limited to only one wavelength.
As a result, a polarization scrambler is used to minimize the degree of polarization across the measuring spectrum such that the random polarization dependent loss is decoupled from the insertion loss (IL) calculation of the PWB.The detector's averaging time is set to 2 ms (500 Hz), which is higher than the scrambler's recommended cut-off frequency of 10 kHz, to achieve a low degree of polarization of < 5 %.The laser was sweept at a speed of 5nm/s between 1490-1640nm.Measurement of multiple samples using a polarization paddle controller optimized for TE coupling versus a scrambler show a consistent excess measurement loss of 0.7 dB ± 0.2 dB per PWB when measured using the depolarized method.S6 shows the measured transmission through the whole optical pathway (red), the loss from the scrambler and fibers (purple) and UHV feedthrough (green), and the calibrated transmission through only the sample which consists of the two PWBs and ring device (yellow).The data shown in blue is the loss through the scrambler, fibers, and UHV feedthrough (purple + green).The micro ring resonator shown in Fig. 5(a) of the main paper was cooled several times to test the repeatability of the PWBs.Fig. S7 (a)-(c) shows the calibrated measurement through the two PWBs and ring resonator for the second cool down following the first cool down shown in Fig. 5 (b)-(d).We show the fourth cool down in Fig. S7 (d)-(e) (another device was measured for the third cool down).There is no data for warm up because the bond broke during high power testing.The data is not calibrated and the extra loss compared to Fig. S6 is from added attenuation for the high power testing.The measurement shows the transmission of the sample slightly degrades from 300 K to 5 K, remains the same for many hours at 4 K, and returns back to its original value after warming up.Fig. S8 shows the measured resonance shift of a ring resonator inside an unbaked cryogenic chamber with pressure of ∼ 9 × 10 −5 mbar over several cooling and warming cycles compared to the expected theoretical resonance shift assuming in perfect vacuum.We expect a 12.5 nm blue shift around 1550 nm (green) due to the decrease in effective refractive index (n e f f ) of the waveguide.However, the first cool down measurement (black arrow) shows a resonance shift of only 5.7 nm (light blue) which suggests there is a counter effect on the n e f f of the waveguide during cooldown.This is likely due to the deposition of gas molecules that are inside the chamber, such as water and nitrogen, onto the surface of the chip.Moreover, the resonance undergoes a steep red shift (dashed arrow) during warm up starting around 185 K which is likely due to the deposition of water molecules that evaporated from the warmed chamber on to the still cold sample.Water has a vapour pressure of ∼ 9.4 × 10 −5 mbar at 183 K [15].Once all gas molecules have evaporated off the sample at 300K, the resonance returns to the same wavelength prior to cooling.The resonance shift matches closer to the simulation with each new cool down cycle as gas particles are removed from the chamber by the turbo pump over time.

MODELING OF RING RESONATOR COOL DOWN
The change of refractive index of silicon as a function of temperature is the main effect in causing the resonance shift of the ring resonator during cool down.The m th order resonance wavelength of a given ring with radius R is n e f f is the effective index of the waveguide mode and is calculated using Lumerical MODE.For a fixed bent waveguide at temperature, T, the n e f f can be fitted to a Taylor expansion approximation: Using Eq.S1 and Eq.S2, the resonance wavelength of the m th order is derived.
The waveguide height and width in the MODE simulation is adjusted to match the fabricated dimension measured by the AFM and SEM.The width is finely tuned until the FSR and resonance wavelengths match close to the measured transmission spectrum.To model λ m at different temperatures, the refractive index data from [16][17][18] at 1500 nm is fed into the MODE simulation as a perturbation to the refractive index of silicon and silicon dioxide to extract the new n e f f (T, λ).
Here, the effect due to thermal contraction is neglected as it is a 100 factor smaller than the thermo-optic effect [19].Additionally, the refractive index perturbation is assumed to be the same for the range of the wavelength shift and for resonances near 1500 ± 50 nm .Fig. 6(a) in the main paper shows the resonance wavelength versus temperature over multiple cool down and warm up cycles and the data shows good agreement to the model.

POWER LIMIT
An optical microscope image of the photonic wirebonds that evaporated during high power testing is shown in Fig. S9.Based on the evaporated section visible on the bottom waveguide loop back and the MRR labled A it was conclude that the lack of convection cooling in the vacuum environment was responsible for the much lower power limit.

Fig. S4 .Fig. S5 .
Fig. S4.A normalized transmission spectrum for Vancore a prepared on a grid.The transmission values were scaled to match the spectrums between the different detectors and sources in their overlap regions.In blue is the InGaAs detector in combination with the NIR source.In red is the InSb detector in combination with NIR source.Finally in yellow is the InSb detector in combination with the MIR source.The sharp spikes present around 1400nm, 1800nm and 4500nm are from absorption caused by gases present in the atmosphere.The transmission through this sample is too low between 2800nm and 3600nm to see the absorption peaks from OH and CH bonds.

Fig. S6 .
Fig. S6.Plot showing the sources of loss that was calibrated out of the measurement to extracted the transmission through only the two PWBs and ring resonator.

Fig.
Fig.S6shows the measured transmission through the whole optical pathway (red), the loss from the scrambler and fibers (purple) and UHV feedthrough (green), and the calibrated transmission through only the sample which consists of the two PWBs and ring device (yellow).The data shown in blue is the loss through the scrambler, fibers, and UHV feedthrough (purple + green).
Fig. S7.Plot (a)-(c) shows the calibrated transmission through the two PWBs and ring resonator device for the second cool down cycle.Plot (d) and (e) shows the uncalibrated transimssion through the whole optical path for the fourth cool down cycle.

Fig. S8 .
Fig. S8.Plot showing the shift of the resonance wavelength around 1550 nm as a function of temperature without baking.The resonances in blue are measured during a cool down cycle while the resonances in yellow are measured during a warm up cycle.
Fig. S9.A focus stacked optical microscope image of the failed photonic wire bonds.One can see the evaporated bond from the high power testing in both the ring resonator structure labeled A and the waveguide loop back below A. The ring resonator structure labeled B is present with two intact bonds; this device was not powered during the experiment and is present as a reference.