Non-volatile electrically programmable integrated photonics with a 5-bit operation

doi:10.21203/rs.3.rs-2430366/v1

Download PDF

Article

Non-volatile electrically programmable integrated photonics with a 5-bit operation

https://doi.org/10.21203/rs.3.rs-2430366/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 12 Jun, 2023

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

Scalable programmable photonic integrated circuits (PICs) can potentially transform the current state of classical and quantum optical information processing. However, traditional means of programming, including thermo-optic, free carrier dispersion, and Pockels effect result in either large device footprints or high static energy consumptions, significantly limiting their scalability. While chalcogenide-based non-volatile phase-change materials (PCMs) could mitigate these problems thanks to their strong index modulation and zero static power consumption, they often suffer from large absorptive loss, low cyclability, and lack of multilevel operation. Here, we report a wide-bandgap PCM antimony sulfide (Sb₂S₃)-clad silicon photonic platform simultaneously achieving low loss (< 1.0 dB), high extinction ratio (> 10 dB), high cyclability (> 1,600 switching events), and 5-bit operation. These Sb₂S₃-based devices are programmed via on-chip silicon PIN diode heaters within sub-ms timescale, with a programming energy density of ∼10fJ/nm³. Remarkably, Sb₂S₃ is programmed into fine intermediate states by applying multiple identical pulses, providing controllable multilevel operations. Through dynamic pulse control, we achieve 5-bit (32 levels) operations, rendering 0.50 ± 0.16 dB per step. Using this multilevel behavior, we further trim random phase error in a balanced Mach-Zehnder interferometer. Our work opens an attractive pathway toward large-scale energy-efficient programmable PICs with low-loss and multi-bit operations.

phase-change materials

non-volatile

programmable integrated photonics

electrical control

multilevel operation

Programmable photonic integrated circuits (PICs), usually composed of arrays of tunable beam splitters and phase shifters, can change their functionalities on demand. This flexibility has recently extended their traditional applications from optical interconnects to optical computing¹, optical programmable gate arrays², and quantum information processing³. Further scaling of the programmable PICs requires the constituent devices to have a smaller footprint and lower power consumption. However, current programmable PICs are primarily based on weak tuning mechanisms, such as thermo-optic effects^4,5, free-carrier effects⁶, and Pockels effects⁷, which are traditionally optimized for low switching power and higher speed application. Unfortunately, they provide a small change in the refractive index (Δn < 0.01), resulting in a large footprint (> 100 µm). In particular, thermo-optic effects incur severe thermal crosstalk, necessitating additional heaters and control circuits for crosstalk compensation. Most importantly, these effects are all volatile and require a constant power supply (~ 10 mW¹). Such high static power consumption is prohibitive for programmable PICs, which generally need infrequent programming⁸; example applications include post-fabrication trimming⁹, setting weights in a photonic neural network for fast inference^1,10,11, and connecting functioning devices, including quantum emitters on-chip¹². We emphasize that, the traditional metrics like switching speed and switching energy for these applications are of little importance, as the switching events occur very infrequently.

Chalcogenide-based non-volatile phase-change materials (PCMs) can provide a unique solution for these applications^13–15. PCMs have two stable, reversibly switchable micro-structural phases (amorphous and crystalline, termed here as a- and c-phase), with drastically different optical refractive indices (Δn ~ 1). Thanks to the non-volatile phase transition, no static power is needed to hold the state upon switching PCMs’ phase. The substantial refractive index change Δn and nonvolatility enable compact reconfigurable devices (~ 10 µm^10,16) with zero static energy consumption^17–22. Moreover, PCMs are compatible with large-scale integration since they can be easily deposited by sputtering^{17,20, 22–24} or thermal evaporation¹⁶ onto almost any PIC materials, including silicon and silicon nitride. Despite these advantages, archetypal PCMs such as Ge₂Sb₂Te₅ (GST) and GeTe exhibit strong absorption in both phases at relevant optical communication wavelengths. This hinders their utility in phase shifters – an essential building block for programmable PICs. Emerging wide-bandgap PCMs, such as GeSbSeTe (GSST)²⁵, antimony selenide (Sb₂Se₃)²⁶, and antimony sulfide (Sb₂S₃)²⁷ can circumvent this loss and have recently generated strong interest in the community^{16,21, 28–30}. In particular, Sb₂S₃ shows the widest bandgap among all these PCMs, allowing transparency down to ~ 600 nm in the amorphous phase²⁷. Moreover, the lack of selenium in Sb₂S₃ makes it less toxic³¹ and less prone to cause chamber contamination during sputtering or evaporation processes. Thus, Sb₂S₃ is much more amenable to be adapted in a commercial foundry. Despite these promises, high-endurance electrical control of Sb₂S₃ remained unsolved.

Here, we demonstrate electrically controlled Sb₂S₃-clad silicon photonic devices with low insertion loss (< 1 dB), high extinction ratio (> 10 dB), and high endurance (> 1,600 switching events). The phase transition is actuated by silicon PIN (p-doped-intrinsic-n-doped) diode heaters. We established the versatility of this hybrid platform with three different integrated photonic devices, including microring resonators, Mach-Zehnder interferometers (MZIs), and asymmetric directional couplers. We also observed multilevel (32 levels) operation, highest among all reported electrically controlled PCM-silicon platforms, with a resolution of 0.50 ± 0.16 dB per step. This multilevel operation is achieved by sending multiple thermally separated (~ 1 second) near-identical electrical pulses for both amorphization and crystallization. We leveraged this multilevel behavior to trim a balanced MZI to correct the random phase error caused by fabrication imperfections. The multilevel operation is crucial to avoid under or over-correction during trimming. Our work opens a new avenue for non-volatile electrically programmable PICs with low loss and on-demand multilevel operation.

We characterized the sputtered Sb₂S₃ thin films to obtain the refractive indices and verify the crystallization capability (annealed under 325°C in Nitrogen environment for 10 mins) of the as-deposited a-Sb₂S₃ (Supplementary Section S1). The silicon photonic devices (Fig. 1-Fig. 3) were designed to operate at the telecommunication O-band (1260–1360 nm) and fabricated on a standard silicon-on-insulator (SOI) wafer with 220 nm silicon and 2 µm buried oxide. The 500-nm-wide waveguides are fabricated by partially etching 120-nm silicon. We then deposited 450-nm-wide Sb₂S₃ onto the SOI chip via sputtering. The slightly smaller width than the waveguide is to compensate for electron beam lithography (E-beam) overlay tolerance. Our simulation results show a change in effective index \({{\Delta }n}_{eff}\approx 0.018\) between a- and c-Sb₂S₃ (Supplementary Section S2). The Sb₂S₃ films are electrical controlled via on-chip silicon PIN micro-heaters^20,22. The p⁺⁺ and n⁺⁺ doping regions were designed 200 nm away from the waveguide to avoid free-carrier absorption loss²², indicated in the scanning electron microscope (SEM) images with false colors in Fig. 1c, Fig. 2c, and Fig. 3c. The Sb₂S₃ stripes are encapsulated with 40 nm of Al₂O₃ grown by atomic layer deposition (ALD) under 150°C. This conformal encapsulation is critical to prevent Sb₂S₃ from oxidation and thermal reflowing, and thus is essential to attain high endurance. To show that our Sb₂S₃-clad silicon photonic platform is versatile and compatible with most PIC components, we demonstrate three widely used PIC components: (1) a microring resonator to show low-loss tuning of cavities, (2) a balanced MZI to demonstrate a full π phase shift and a broadband operation, and (3) an asymmetric directional coupler to create a compact programmable unit (see simulation results in Supplementary Section S2).

Nonvolatile microring switch integrated with Sb₂S₃ phase shifter

We deposited 10-µm-long, 20-nm-thick Sb₂S₃ on a micro-ring resonator with 30 µm radius (Figs. 1a-c). The free spectral range (FSR) is ~ 2.42 nm (Fig. 1d), and the bus-ring gap is 280 nm to achieve a near-critically coupled device. We switched the as-deposited a-Sb₂S₃ to c-Sb₂S₃ on the microring resonator by applying three 1.6 V, 200-ms-long pulses with ~ 3.4 mJ energy (or SET pulses) separated by 1 second and then re-amorphized the material via three 7.5 V, 150-ns-long pulses with ~ 56 nJ energy also separated by 1 second (RESET pulses). The unit energy density (energy/total Sb₂S₃ volume) for switching Sb₂S₃ is then estimated as 0.6 \(fJ/n{m}^{3}\) (38 \(pJ/n{m}^{3}\)) for amorphization (crystallization). We note that the 200-ms-long SET pulse is indeed significantly longer than other reported PCMs, such as GST (50 µs²² or 100 µs²¹) and Sb₂Se₃ (5 µs¹⁶ or 100 µs²¹), and causes a large crystallization energy and energy density. But we found that the SET pulse duration can be reduced to around 100 µs after the first few cycles (see Method and Supplementary Section S12). Therefore, the crystallization power is reduced to ~ 1.7 \(\mu\)J (19 \(fJ/n{m}^{3}\)) after the initial conditioning, comparable to other PCMs^16,22. With the \(100 \mu s\) crystallization pulse condition, this device could be operated at ~ kHz speed for a complete SET/RESET cycle. The slow crystallization, however, can allow amorphizing a large volume of Sb₂S₃ (Supplementary Section S13). Figure 1d shows a resonance shift of ~ 0.394 nm upon switching the 10 µm Sb₂S₃ from the a- (blue) to the c-phase (orange), corresponding to a π-phase shift length L_π of ~ 30.7 µm, significantly shorter than the 1-millimeter L_π of ferroelectric non-volatile phase shifter³². The SET and RESET processes were repeated for 10 cycles. The shift in resonance is highly repeatable, as suggested by the slight standard deviation (Fig. 1d). The excess loss from a-Sb₂S₃ is negligible²⁴, and in c-Sb₂S₃ hybrid waveguides the loss is estimated to be 0.024 dB/µm (0.72 dB/π), which is three times larger than our simulation result (0.26 dB/\(\pi\)) (Supplementary Section S2). We attribute this excess loss to the scattering from the Sb₂S₃ thin film due to non-uniform deposition/liftoff and local crystal grains in c-Sb₂S₃ (Supplementary Section S3). We verified that the loss due to mode mismatch at the transition between the bare silicon waveguide and the Sb₂S₃-loaded waveguide is small (~ 0.013 dB/facet), consistent with the fact that the thin Sb₂S₃ film should not significantly change the mode shape (Supplementary Section S2).

Nonvolatile Mach-Zehnder switch integrated with Sb₂S₃ phase shifter

Figures 2a-c show a balanced MZI operating at wavelengths between 1,320 nm and 1,360 nm with both arms covered with 30-µm-long 20-nm-thick Sb₂S₃. A multimode interferometer with a 50:50 splitting ratio was designed and fabricated, as shown in Fig. 2d (simulation results in Supplementary Section S2). Initially, the light comes out mainly from the bar port with an extinction ratio of ~ 13 dB (Fig. 2e). The light coming out from the bar port instead of the cross port in this balanced MZI can be explained by the random phase errors in two arms due to fabrication imperfection, especially that in the S-bend. One can overcome such imperfection by exploiting a wider waveguide to improve fabrication robustness³³. Alternatively, this random phase error can be corrected using Sb₂S₃ for post-fabrication trimming, as we show later in this paper. The Sb₂S₃ on one arm was switched by two 1.7 V, 200-ms SET electric pulses with an energy of 11.6 mJ (density: 42 \(pJ/n{m}^{3}\)) to provide a full π phase shift. An 8.1 V, 150-ns short RESET electric pulse with an energy of 197 nJ (density: 0.73 \(fJ/n{m}^{3}\)) switched the device back to the initial state. Figures 2e and 2f show the transmission spectra normalized to a reference waveguide when the Sb₂S₃ film is in the amorphous and crystalline phases, respectively. The c-Sb₂S₃ displays a complete spectrum flip, showing a bar state with an extinction ratio of 15 dB. We then recorded the bar port transmission at 1,330 nm for 100 switching events without device degradation (Supplementary Section S4).

Compact asymmetric directional coupler switch

We also designed and fabricated a compact asymmetric directional coupler (coupling length L_c ≈ 79 µm) (simulation in Supplementary Section S2), as shown in Figs. 3a-c. The coupler consists of two waveguides with different widths. The narrower 409-nm-wide waveguide (hybrid waveguide) was capped with 20-nm-thick Sb₂S₃ and designed to allow phase match with the wider 450-nm-wide waveguide (bare waveguide) for c-Sb₂S₃³⁴. As such, the input light could completely couple to the cross port in one coupling length. Once Sb₂S₃ is switched to the amorphous state, the effective index of the hybrid waveguide changes while the bare waveguide remains the same. The resulting phase mismatch changes the coupling strength and coupling length. Then, co-optimizing the gap and waveguide length permits a complete bar transmission. The unique c-Sb₂S₃ phase matching approach, instead of a-Sb₂S₃^18,20,34,35, allows a more symmetric performance regardless of the input port (Supplementary Section S2), crucial for a 2 × 2 device. If phase mismatch happens in the c-Sb₂S₃ state, the slight loss of c-Sb₂S₃ on one of the waveguides will result in different bar state insertion loss when the light goes from different input ports. We note that the 79-µm coupling length can be potentially reduced (~ 34 µm) by depositing a thicker (50 nm) Sb₂S₃ to provide stronger refractive index modulation.

Figures 3d and 3e show the transmission spectra for a- and c-Sb₂S₃, switched with three 9.6 V, 500 ns, 922 nJ RESET, and 2.7 V, 200 ms, 29.2 mJ SET pulses, respectively. The energy density for amorphization (crystallization) is 1.28 \(fJ/n{m}^{3}\) (40 \(pJ/n{m}^{3}\)). The insertion losses are 2 dB (0.5 dB), and the extinction ratios are around 10 dB (11 dB) for a (c)-Sb₂S₃). The unexpected high insertion loss when the Sb₂S₃ is in the amorphous state can be attributed to several factors, including gap discrepancy, Sb₂S₃ overlay deviation, or the cross-port grating coupler fabrication imperfection. To estimate the actual loss of the device, we apply the c-Sb₂S₃ loss extracted from the ring resonator to the simulation and calculate this device’s insertion loss to be ~ 0.1 dB (0.9 dB) for a(c)-Sb₂S₃ (Supplementary Section S2).

Figure 4 shows 1,600 switching events for the asymmetric directional coupler. Limited by our measurement setup, we separately measured the cross (Fig. 4a) and bar ports (Fig. 4b). The higher insertion loss (~ 1 dB) at around event 500 was due to optical fiber misalignment. We note that almost no performance degradation occurred at the end (Supplementary Section S5); hence, 1,600 switching events are not the limit of this device. The cross-port transmission shows a relatively large variation for a-Sb₂S₃ (Fig. 4a blue scatterers, from ~ -15 dB to ~ -35 dB), which was caused by incomplete amorphization or thermal reflow of the Sb₂S₃ film. Since the plot is on a logarithmic scale, such a large variation in a-Sb₂S₃ (due to higher transmission) is not visible in Fig. 4b.

Multilevel 5-bit operation with dynamic electrical control

Our Sb₂S₃-Si integrated structures further show a stepwise multilevel operation up to 32 levels with dynamic pulse control. In Fig. 5a, we show the multilevel transmission at both cross and bar ports of an asymmetric directional coupler while sending in RESET 10 V, 550 ns, 1.1 \(\mu\)J pulse every other second to amorphized the Sb₂S₃. We started the experiment with “coarse” tuning, where unoptimized, identical pulses were sent to partially amorphize the c-Sb₂S₃ device to demonstrate multiple levels. The asymmetric directional coupler was originally in the “cross state” (red region). After one partial amorphization pulse, it was reconfigured into an intermediate state (orange region), where light comes out from both cross and bar ports. After six pulses, a complete “bar state” (green region) was achieved. We repeated this experiment five times for each port and plotted the average transmission levels and the standard deviation. The variation is attributed to the stochastic phase change process using electrical controls³⁶. We also experimentally tested the partial crystallization of the device and multilevel operation (Supplementary Section S5), which is based on the growth-dominant nature of Sb₂S₃ crystallization process³⁴. In the following experiments, we mainly focused on partial amorphization because of lower energy consumption and more operation levels with finer resolution.

Such stepwise multilevel operation by applying identical pulses is distinctly different from previously reported multilevel operations in GST^20,22 and Sb₂Se₃^16,21, where different voltage amplitudes or pulse duration were used to access multiple levels during amorphization. The pulse-number-dependent behavior is quite counterintuitive: one expects that after the first amorphization pulse, the thin Sb₂S₃ film would have reached its new equilibrium phase. Moreover, since the thermal processes relax within 10 µs (Supplement Section S6), each pulse is independent due to the relatively long one-second interval. As a result, the subsequent identical and separated pulses should not further change the material phase. To understand the origin of the multi-level operation, we closely inspected four partially amorphized Sb₂S₃ devices under the microscope. We observed a few separate patches (Supplement Section S7) and a region that grew with more voltage pulses. As reported in some literature, one possibility could have been that a- and c-Sb₂S₃ have significantly different thermal conductivities and specific heat capacities. But we measured the thermal conductivities to be similar (a-Sb₂S₃: 0.2 W/m/K; c-Sb₂S₃: 0.4 W/m/K, see Methods), and hence, we ruled out this as a possible explanation. We hypothesize that this unique behavior comes from Sb₂S₃’s multiple crystalline phases. Sb₂S₃ has at least two distinct crystalline phases³⁷, which may differ in the amorphization conditions. The partial amorphization pulse can cause amorphization in the hottest region, but at the lower temperature region, it may cause phase transition to the other crystalline phase. These regions get amorphized in subsequent pulses, resulting in multilevel operation.

An even finer multilevel operation was realized by monitoring the transmission level and dynamically changing the pulse conditions slightly. Here, we demonstrate on-demand 5-bit operation in a quasi-continuously tunable directional coupler, as shown in Fig. 5b. We dynamically controlled the partial amorphization pulses to have slightly lower, near-identical voltages (ranging from 9.65 V to 9.85 V) and obtained up to 32 levels. Figure 5b demonstrates 5-bit operation (32 distinct levels) with a target resolution of 0.5 dB per level step at 1,340 nm (see the detailed pulse conditions in Supplementary Section S8). We emphasize that dynamic control is necessary to mitigate the stochastic nature of electrically controlled PCMs³⁶, hence essential for a reliable many-level operation. In Fig. 5b, a linear fit shows a slope of -0.50 dB per step and a standard deviation of 0.16 dB among five experiments, indicating a repeatable operation. While the 5-bit operation of GST was shown using laser pulses³⁸, our demonstration is thus far the highest number of operating levels reported using electrical control in PCMs-based photonics. Moreover, our multilevel operation does not require sophisticated heater geometry engineering, such as the segmented doped silicon heater design¹⁶, and solely relies on the unique phase-change dynamics of Sb₂S₃.

Random phase error correction in balanced MZIs exploiting multilevel operation

Finally, exploiting the multilevel operation, we corrected random phase errors in a balanced MZI. A perfectly balanced MZI should initially be in an all-cross state. However, random phase errors due to fabrication imperfections can easily build up to a phase error of π, making the initial state unpredictable. Therefore, balanced MZIs usually require extra calibration³³. For example, Fig. 6a shows the transmission spectra of a phase error corrupted balanced MZI, indicating a high bar transmission. Both arms of the MZI are loaded with 40-µm-long Sb₂S₃ film to guarantee a phase tuning range of more than ± π. The corrected MZI spectra by multilevel tuning of Sb₂S₃ are demonstrated in Fig. 6b, showing a pure “cross state” with a high extinction ratio of 24 dB. The trimming process is shown in Fig. 6c. We sent in a partial amorphization pulse (8.8 V, 150 ns, 232 nJ) every other second, which gradually increased the portion of amorphous Sb₂S₃, resulting in quasi-continuous changes of the bar (blue) and cross (orange) transmission (at 1,340 nm). The correction finishes once a bar transmission minimum is reached, indicated by the red arrows in Fig. 6c, and the spectra reported in Fig. 6b were then measured. Further pulses increase bar transmission because of the over-compensated phase. We performed the same experiment three times, indicated by different colored regions in Fig. 6c. Complete phase error correction was observed in all three instances, exhibiting excellent repeatability of our trimming process. Note that binary tuning cannot accomplish this task due to the random initial phase error. Even multilevel operations with limited discrete-level resolution can cause over- or under-corrected phase error, ultimately determining the trimming resolution. We highlight that this method requires zero static energy supply once the phase error is corrected, as the phase transition in Sb₂S₃ is non-volatile (> 77 days, Supplementary Section S9). Thanks to the relatively fine operation levels, slight over-tuning does not significantly affect the performance. The trimming resolution can be further improved using dynamically changed, near-identical electrical pulses, as shown in the previous Section. Moreover, if the phase error is over compensated, the device could be tuned back with partial crystallization pulses (Supplementary Section S5). In the future, our trimming process can potentially be fully automated by real-time adjusting the pulse numbers according to the measured transmission.

Among the first few cycles (~ 50 cycles), sub-millisecond pulses failed to trigger the crystallization for Sb₂S₃. We increased the voltage of a 500-µs pulse up to the material ablation voltage level without observing any crystallization. This suggests that the thermally induced Sb₂S₃ crystallization process is relatively slow, different from laser-induced crystallization²⁷, where a crystallization speed of tens of nano-second was demonstrated. We attribute this behavior to the difficulty in the initial nucleation. This hypothesis is supported by the fact that after the first few crystallization processes (conditioning), consecutive crystallization process could happen only at 100 µs (Supplementary Section S12), increasing the speed and lowering the energy by three orders of magnitudes. Hence, crystallization becomes easier after the initial nuclei/ defects are formed by the first few long thermal processes. We used 200-ms pulses in the reported experiments because the post-conditioning, shorter 100-\(\mu s\) pulse condition was identified later in our experiments.

However, this limited crystallization speed (~ kHz) should not be considered as a problem, because of PCMs’ main advantage of nonvolatility, and thus zero-static power consumption. This nonvolatile reconfiguration technology is complementary to the traditional high-speed modulators, and will find wide applications with infrequent programming needs, where high speed is not essential and the zero-static power is more important^8,39. Moreover, the slow crystallization speed could prevent unintentional recrystallization during long-pulse amorphization. Since long-pulse amorphization means a smaller thermal gradient and cooling rate, these conditions enable a thicker layer of Sb₂S₃ to switch completely. We verified that a pulse with 10 µs duration was able to trigger a large degree of amorphization (Supplementary Section S13), which is not possible in fast PCMs, such as GST. This capability of actuating phase transition in thicker films is useful in photonics, since switching a larger volume of PCMs is generally required to provide stronger modulation.

In conclusion, we demonstrated a multi-bit integrated electro-photonic platform using wide bandgap PCM Sb₂S₃ and doped silicon PIN heaters. Operating in the telecommunication O-band, the Sb₂S₃-Si hybrid ring resonators, MZIs, and directional couplers exhibit a low insertion loss (< 1.0 dB) and a high extinction ratio (> 10 dB). We report record-high electrical cyclability of Sb₂S₃ (> 1,600 switching events). Notably, the Sb₂S₃-based photonic devices could be tuned into distinct intermediate levels in a stepwise fashion by consecutively sending identical, thermally isolated partial amorphization pulses. Such pulse-number-dependent multilevel tuning allows high-resolution operation levels compared to different voltages or pulse widths. We demonstrate precise on-demand 5-bit operation with 0.50 ± 0.16 dB per level step in a tunable beam splitter. We further exploited this behavior to demonstrate random phase error correction in a balanced MZI, which will find practical applications. Our work opens a new pathway toward various integrated photonics applications, such as post-fabrication trimming, optical information processing, and optical quantum simulations, where the nonvolatile on-demand multi-bit operation is of paramount interest.

SOI Device Fabrication

The silicon photonic devices were fabricated on a commercial SOI Wafer with 220-nm-thick Silicon on 2-µm-thick SiO₂ (WaferPro). All devices were defined using electron-beam lithography (EBL, JEOL JBX-6300FS) with a positive-tone E-beam resist (ZEP-520A) and partially etched by ~ 120 nm in a fluorine-based inductively coupled plasma etcher (ICP, Oxford PlasmaLab 100 ICP-18) with mixed SF₆/C₄F₈. The etching rate was around 2.8 nm/sec. The doping regions were defined by two additional EBL rounds with 600-nm-thick poly (methyl methacrylate) (PMMA) resist and implanted by boron (phosphorus) ions for p⁺⁺ (n⁺⁺) doping regions with a dosage of 2 × 10¹⁵ ions per cm² and ion energy of 14 keV (40 keV). The chips were annealed at 950°C for 10 min (Expertech CRT200 Anneal Furnace) for dopant activation. Ideal Ohmic contact was formed after removal of the surface native oxide via immersing the chips in 10:1 buffered oxide etchant (BOE) for 10 seconds. The metal contacts were then immediately patterned by a fourth EBL step using PMMA. Metallization was done by electron-beam evaporation (CHA SEC-600) and lift-off of Ti/Pd (5 nm/180 nm). We measured a 30-\(\mu m\)-long PIN diode and obtained a threshold voltage of around 0.8 V, and a resistivity of around 50 \({\Omega }\) (Supplementary Section S4). After a fifth EBL defining the Sb₂S₃ window, a 40-nm Sb₂S₃ thin film was deposited using a GST target (AJA International) in a magnetron sputtering system (Lesker Lab 18), followed by a lift-off process. We note the actual Sb₂S₃ thickness on the waveguide was reduced to around 20nm because of the narrow PMMA trench (Supplementary Section S10). The Sb₂S₃ was then encapsulated by 40-nm-thick Al₂O₃ through thermal ALD (Oxford Plasmalab 80PLUS OpAL ALD) at 150°C. To ensure good contact between the electric probe and metal pads while applying electrical pulses, the Al₂O₃ on the metal contacts was removed by defining a window using a sixth EBL with 600 nm PMMA, then etching in a chlorine-based inductively coupled plasma etcher (ICP-RIE, Oxford PlasmaLab 100 ICP-18).

Optical simulation

The refractive index data for Sb₂S₃ were measured by an ellipsometer (Woollam M-2000)⁴⁰. The phase shifters and asymmetric directional couplers were designed (verified) by a commercial photonic simulation software package Lumerical MODE (FDTD).

Heat transfer simulation

The doped silicon PIN heater was simulated with a commercial Multiphysics simulation software COMSOL Multiphysics^41,42. In the simulation, a heat transfer in a solid model is coupled with a semiconductor model to simulate the transient time performance.

Optical transmission measurement setup

The programmable units were measured with a 25°-angled vertical fiber-coupling setup. The stage temperature was controlled at 26°C by a thermoelectric controller (TEC, TE Technology TC-720). A tunable continuous-wave laser (Santec TSL-510) sent in the input light, the polarization of which was controlled by a manual fiber polarization controller (Thorlabs FPC526) to achieve a maximum fiber-to-chip coupling efficiency. A low-noise power meter (Keysight 81634B) measured the static optical transmission. The transmission spectra of all Sb₂S₃ devices were normalized to the spectra of the nearest reference waveguide. For the on-chip electrical switching, electrical pulses were applied to the on-chip metal contacts via a pair of electrical probes on two probe positioners (Cascade Microtech DPP105-M-AI-S). The crystallization and amorphization pulses were generated from a pulse function arbitrary generator (Keysight 81160A). The tunable laser, power meter, thermal controller, source meter, and pulse function arbitrary generator were controlled by a LabView program⁴¹.

When electrically switching Sb₂S₃, we found that a single amorphization or crystallization voltage pulse switches the device to transmission levels with large state-to-state variations (Supplementary Section S11). This could be attributed to the switching of Sb₂S₃ into random intermediate structural phases since it has two (or more) crystalline phases³⁷. This issue was tackled by using three identical pulses to switch the Sb₂S₃ phase completely.

Thermal parameter measurement

The thermal conductivity of our films was measured with time-domain thermoreflectance (TDTR)^43,44. TDTR uses ultrafast modulated laser heating through the absorption of a thin metallic transducer layer (70 nm Pt). An unmodulated probe laser then measures the surface temperature through a proportional change in the transducer reflectivity. Measurements were taken at a pump beam modulation frequency of 4 MHz and 10 MHz to ensure the thermal penetration depth exceeded the thickness of the films. Knife edge measurements provided 1/e² beam radii of 5.5 and 3.1 ± 0.05 µm for the pump and probe, respectively. The resulting thermoreflectance data were then fit to the solution of a 3D heat diffusion model for a multi-layer stack of materials (Pt-film-substrate) and the effective thermal resistance was determined as a function of film thickness (Supplementary Section S14). A linear regression of the foregoing data provided the thermal boundary resistance of the material stack, which was then used to determine the intrinsic thermal conductivities for the films. The properties of the Pt layer, the films, and the Si substrate were determined from independent measurements or adopted from the literature^45,46.

Acknowledgments: The authors thank Asir Intisar Khan, Kathryn M. Neilson, Prof. Eric Pop at Stanford University, and Prof. Juejun Hu at MIT for the insightful discussions.

Funding: The research is funded by National Science Foundation (NSF-1640986, NSF-2003509), ONR-YIP Award, DARPA-YFA Award, NASA-STTR Award 80NSSC22PA980, and Intel. F.M. is supported by a Draper Scholars Program. Part of this work was conducted at the Washington Nanofabrication Facility/ Molecular Analysis Facility, a National Nanotechnology Coordinated Infrastructure (NNCI) site at the University of Washington, with partial support from the National Science Foundation via awards NNCI-1542101 and NNCI-2025489. Part of this work was performed at the Stanford Nano Shared Facilities (SNSF), supported by the National Science Foundation under award ECCS-1542152).

Author contributions: R.C. and A.M. conceived the project. R.C. simulated and fabricated the Sb₂S₃ silicon photonic devices, performed optical characterizations and data analysis. Z.F. and J.Z. helped with the device simulation, fabrication, characterization, and data analysis. F.M., S.G., K.K. and A.S. helped with the optical measurements. C.P. and K.E.G. measured thermal conductivity of Sb₂S₃ thin films. A.M. supervised and planned the project. R.C. wrote the manuscript with input from all the authors.

Competing interests: Authors declare that they have no competing interests.

Data availability: The data that support the findings of this study are available from the corresponding author upon reasonable request.

Shen, Y. et al. Deep learning with coherent nanophotonic circuits. Nature Photonics 11, 441–446 (2017).
Capmany, J., Gasulla, I. & Pérez, D. Microwave photonics: The programmable processor. Nature Photonics 10, 6–8 (2016).
Harris, N. C. et al. Quantum transport simulations in a programmable nanophotonic processor. Nature Photonics 11, 447–452 (2017).
Pérez, D. et al. Multipurpose silicon photonics signal processor core. Nature Communications 8, 1–9 (2017).
Zhang, W. & Yao, J. Photonic integrated field-programmable disk array signal processor. Nature Communications 11, 1–9 (2020).
Reed, G. T., Mashanovich, G., Gardes, F. Y. & Thomson, D. J. Silicon optical modulators. Nature Photon 4, 518–526 (2010).
Wang, C. et al. Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages. Nature 562, 101–104 (2018).
Chen, R. et al. Opportunities and Challenges for Large-Scale Phase-Change Material Integrated Electro-Photonics. ACS Photonics (2022) doi:10.1021/acsphotonics.2c00976.
Schrauwen, J., Thourhout, D. V. & Baets, R. Trimming of silicon ring resonator by electron beam induced compaction and strain. Opt. Express, OE 16, 3738–3743 (2008).
Feldmann, J. et al. Parallel convolution processing using an integrated photonic tensor core [S]. 589, (2021).
Wu, C. et al. Programmable phase-change metasurfaces on waveguides for multimode photonic convolutional neural network. Nature Communications 12, 96 (2021).
Wan, N. H. et al. Large-scale integration of artificial atoms in hybrid photonic circuits. Nature 583, 226–231 (2020).
Wuttig, M., Bhaskaran, H. & Taubner, T. Phase-change materials for non-volatile photonic applications. Nature Photonics 11, 465–476 (2017).
Abdollahramezani, S. et al. Tunable nanophotonics enabled by chalcogenide phase-change materials. Nanophotonics 9, 1189–1241 (2020).
Fang, Z., Chen, R., Zheng, J. & Majumdar, A. Non-Volatile Reconfigurable Silicon Photonics Based on Phase-Change Materials. IEEE Journal of Selected Topics in Quantum Electronics 28, 1–17 (2022).
Ríos, C. et al. Ultra-compact nonvolatile phase shifter based on electrically reprogrammable transparent phase change materials. PhotoniX 3, 26 (2022).
Zheng, J. et al. GST-on-silicon hybrid nanophotonic integrated circuits: a non-volatile quasi-continuously reprogrammable platform. Optical Materials Express 8, 1551 (2018).
Xu, P., Zheng, J., Doylend, J. K. & Majumdar, A. Low-Loss and Broadband Nonvolatile Phase-Change Directional Coupler Switches. ACS Photonics 6, 553–557 (2019).
Fang, Z. et al. Non-Volatile Reconfigurable Integrated Photonics Enabled by Broadband Low-Loss Phase Change Material. Advanced Optical Materials 9, (2021).
Chen, R. et al. Broadband Nonvolatile Electrically Controlled Programmable Units in Silicon Photonics. ACS Photonics (2022) doi:10.1021/acsphotonics.2c00452.
Fang, Z. et al. Ultra-low-energy programmable non-volatile silicon photonics based on phase-change materials with graphene heaters. Nat. Nanotechnol. 842–848 (2022) doi:10.1038/s41565-022-01153-w.
Zheng, J. et al. Nonvolatile Electrically Reconfigurable Integrated Photonic Switch Enabled by a Silicon PIN Diode Heater. Advanced Materials 32, 2001218 (2020).
Rios, C. et al. Integrated all-photonic non-volatile multi-level memory. Nature Photonics 9, 725–732 (2015).
Fang, Z. et al. Non-Volatile Reconfigurable Integrated Photonics Enabled by Broadband Low-Loss Phase Change Material. Advanced Optical Materials 9, 2002049 (2021).
Zhang, Y. et al. Broadband transparent optical phase change materials for high-performance nonvolatile photonics. Nature Communications 10, 1–9 (2019).
Delaney, M., Zeimpekis, I., Lawson, D., Hewak, D. W. & Muskens, O. L. A New Family of Ultralow Loss Reversible Phase-Change Materials for Photonic Integrated Circuits: Sb2S3 and Sb2Se3. Advanced Functional Materials 30, 2002447 (2020).
Dong, W. et al. Wide Bandgap Phase Change Material Tuned Visible Photonics. Advanced Functional Materials 29, 1806181 (2019).
Shalaginov, M. Y. et al. Reconfigurable all-dielectric metalens with diffraction-limited performance. Nature Communications 12, 1–8 (2021).
Lu, L. et al. Reversible Tuning of Mie Resonances in the Visible Spectrum. ACS Nano 15, 19722–19732 (2021).
Moitra, P. et al. Programmable Wavefront Control in the Visible Spectrum Using Low-Loss Chalcogenide Phase-Change Metasurfaces. Advanced Materials n/a, 2205367.
Spallholz, J. E. On the nature of selenium toxicity and carcinostatic activity. Free Radical Biology and Medicine 17, 45–64 (1994).
Geler-Kremer, J. et al. A ferroelectric multilevel non-volatile photonic phase shifter. Nat. Photon. 16, 491–497 (2022).
Song, L., Li, H., Dai, D. & Dai, D. Mach–Zehnder silicon-photonic switch with low random phase errors. Opt. Lett., OL 46, 78–81 (2021).
Teo, T. Y. et al. Comparison and analysis of phase change materials-based reconfigurable silicon photonic directional couplers. Opt. Mater. Express, OME 12, 606–621 (2022).
Zhang, Q. et al. Broadband nonvolatile photonic switching based on optical phase change materials: beyond the classical figure-of-merit. Optics Letters 43, 94 (2018).
Tuma, T., Pantazi, A., Le Gallo, M., Sebastian, A. & Eleftheriou, E. Stochastic phase-change neurons. Nature Nanotechnology 11, 693–699 (2016).
Gutierrez, Y. et al. Interlaboratory Study on Sb2S3 Interplay between Structure, Dielectric Function and Amorphous-to-Crystalline Phase Change for Photonics. iScience 104377 (2022) doi:10.1016/j.isci.2022.104377.
Li, X. et al. Fast and reliable storage using a 5 bit, nonvolatile photonic memory cell. Optica 6, 1 (2019).
Zhang, Y. et al. Myths and truths about optical phase change materials: A perspective. Applied Physics Letters 118, 210501 (2021).
Xu, P., Zheng, J., Doylend, J. K. & Majumdar, A. Low-Loss and Broadband Nonvolatile Phase-Change Directional Coupler Switches. ACS Photonics 6, 553–557 (2019).
Zheng, J. et al. Nonvolatile Electrically Reconfigurable Integrated Photonic Switch Enabled by a Silicon PIN Diode Heater. Advanced Materials 32, 2001218 (2020).
Zheng, J., Zhu, S., Xu, P., Dunham, S. & Majumdar, A. Modeling Electrical Switching of Nonvolatile Phase-Change Integrated Nanophotonic Structures with Graphene Heaters. ACS Appl. Mater. Interfaces 12, 21827–21836 (2020).
Kwon, H., Perez, C., Park, W., Asheghi, M. & Goodson, K. E. Thermal Characterization of Metal–Oxide Interfaces Using Time-Domain Thermoreflectance with Nanograting Transducers. ACS Appl. Mater. Interfaces 13, 58059–58065 (2021).
Perez, C. et al. Dominant Energy Carrier Transitions and Thermal Anisotropy in Epitaxial Iridium Thin Films. Advanced Functional Materials n/a, 2207781.
Černošková, E., Todorov, R., Černošek, Z., Holubová, J. & Beneš, L. Thermal properties and the structure of amorphous Sb2Se3 thin film. J Therm Anal Calorim 118, 105–110 (2014).
Ben Nasr, T., Maghraoui-Meherzi, H. & Kamoun-Turki, N. First-principles study of electronic, thermoelectric and thermal properties of Sb2S3. Journal of Alloys and Compounds 663, 123–127 (2016).

There is NO Competing Interest.

supplementv5sub.docx

Download PDF

Journal Publication

published 12 Jun, 2023

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

Non-volatile electrically programmable integrated photonics with a 5-bit operation

Status:

Journal Publication

Version 1

Abstract

Figures

Main

Results

Nonvolatile microring switch integrated with Sb₂S₃ phase shifter

Nonvolatile Mach-Zehnder switch integrated with Sb₂S₃ phase shifter

Compact asymmetric directional coupler switch

Multilevel 5-bit operation with dynamic electrical control

Random phase error correction in balanced MZIs exploiting multilevel operation

Discussion

Methods

SOI Device Fabrication

Optical simulation

Heat transfer simulation

Optical transmission measurement setup

Thermal parameter measurement

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1

Non-volatile electrically programmable integrated photonics with a 5-bit operation

Status:

Journal Publication

Version 1

Abstract

Figures

Main

Results

Nonvolatile microring switch integrated with Sb2S3 phase shifter

Nonvolatile Mach-Zehnder switch integrated with Sb2S3 phase shifter

Compact asymmetric directional coupler switch

Multilevel 5-bit operation with dynamic electrical control

Random phase error correction in balanced MZIs exploiting multilevel operation

Discussion

Methods

SOI Device Fabrication

Optical simulation

Heat transfer simulation

Optical transmission measurement setup

Thermal parameter measurement

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1

Nonvolatile microring switch integrated with Sb₂S₃ phase shifter

Nonvolatile Mach-Zehnder switch integrated with Sb₂S₃ phase shifter