Copper-impurity-free photonic integrated circuits enable deterministic soliton microcombs

Abstract

Chip-scale optical frequency combs based on microresonators (microcombs) enable GHz-THz repetition rates, broad bandwidth, compactness, and compatibility with wafer-scale manufacturing. Silicon nitride photonic integrated circuits have become a leading platform due to their low loss, broad transparency, lithographic dispersion control, and commercial 200-mm-wafer foundry access. They have enabled system-level applications in optical communications, LiDAR, frequency synthesis, low-noise microwave generation, and convolutional processing. However, real-world deployment is hindered by the challenge of deterministic soliton microcomb generation, primarily due to thermal instabilities. Although techniques like pulsed pumping, fast scanning, and auxiliary lasers help mitigate these effects, they often add complexity or reduce soliton stability. In this work, we overcome thermal limitations and demonstrate deterministic soliton generation in silicon nitride photonic circuits. We trace the thermal effects to copper impurities within waveguides, originating from residual contaminants in CMOS-grade silicon wafers that are gettered into silicon nitride during fabrication. By developing effective copper removal techniques, we significantly reduce thermal instabilities. This enables soliton generation with arbitrary or slow laser scanning, removing a key barrier to microcomb deployment. Our approach is compatible with front-end-of-line foundry processing, paving the way for broader adoption of soliton microcomb technologies.

Figures (4)

• Figure 1: Thermal absorption-induced instabilities in Dissipative Kerr soliton (DKS) generation in Si$_3$N$_4$ microresonators.a, Schematic of DKS generation and its diverse applications, including parallel LiDAR, astronomical spectrograph calibration, optical computing, optical communications, frequency calibration, and quantum optics. b, Stability chart of soliton microcomb states in Si$_3$N$_4$ microresonators with anomalous group velocity dispersion, as a function of laser detuning and pump power, with time and frequency domain waveforms predicted by the Lugiato-Lefever Equation (LLE). MI: modulation instability. CW: continuous wave. c, Resonance tilt during pump laser scanning from blue- to red-detuned regimes. Kerr effect induces self-phase modulation, causing a tilted resonance. d, Schematic of cavity resonance shifts. The cold cavity resonance $\omega_0$ thermally shifts to $\widetilde{\omega}_0$ with intracavity energy (e.g., a soliton). The effective detuning of the laser to this shifted resonance determines the soliton properties. e, Numerical simulations of DKS generation in Si$_3$N$_4$ microresonators, illustrating the destabilizing impact of thermal effects on soliton generation. f, Transmission loss spectrum of fluoride optical fibers (in dB/km/ppm atomic), showing contributions from various metal impurities, with Cu ions exhibiting the highest absorption near 1550 nm mitachi1983reduction. g, Depth profile of metal impurities in a fully SiO$_2$-cladded Si$_3$N$_4$ microresonator, measured by secondary ion mass spectroscopy (SIMS). Nitrogen is shown in grey for layer identification. The inset provides a graphical illustration of the SIMS measurement. The oxidation state in the original Si$_3$N$_4$ waveguide may differ due to the ionization process during sputtering.
• Figure 2: Copper-induced thermal absorption loss in Si$_3$N$_4$ microresonators: origins and impacts on DKS generation.a, Histogram of intrinsic linewidth $\kappa_0/2\pi$ (1260-1630 nm, TE polarization), showing optical loss variations between two concurrently fabricated samples using Si wafers from different suppliers. Inset: Simulated optical TE$_{00}$ fundamental mode. Device IDs: Sample 1: D163_11_F1_C7; Sample 2: D163_13_F8_C7. b, Kerr-normalized thermal response measurements at resonance around 1550 nm for Sample 1 and Sample 2. Colored regions represent thermal contributions, which respond at lower modulation frequencies. c, Measured soliton steps during pump laser scanning in TE polarization: Sample 1 exhibits a longer soliton step ($\sim$ 330 MHz) compared to Sample 2 ($\sim$ 8.6 MHz). d, Depth-resolved Cu concentration profiles for Sample 1 and Sample 2, measured via SIMS by monitoring the Cu$^\mathrm{+}$ peak (m/z = 62.95) during ion beam sputtering. The SiO$_2$ cladding was partially removed to focus on the Si$_3$N$_4$ waveguides. SIMS results are corrected based on the Si$_3$N$_4$ filling ratio (21.76$\%$) within the mixed Si$_3$N$_4$/SiO$_2$ waveguide layer. e, Cu profiles in Si substrates used for fabricating Sample 1 (Wafer 1: Siegert Wafer) and Sample 2 (Wafer 2: Silicon Valley Microelectronics) reveal contamination levels comparable to those in the Si$_3$N$_4$ waveguides, identifying the Si substrate as the source of Cu impurities. The measurement was done by depositing 250-nm LPCVD Si$_3$N$_4$ on Si wafers, followed by 11-hour annealing at 1200$^{\circ}$C to concentrate diluted Cu from Si into the gettering Si$_3$N$_4$ layer. f, Schematic of the copper detection method in silicon substrates, involving plasma-enhanced chemical vapor deposition (PECVD) of 200-nm Si$_3$N$_4$ films, and high-temperature annealing (800$^{\circ}$C, 10 hours) to concentrate Cu ions for detection above the sensitivity limit. g, Cu diffusion profiles during annealing (as-deposited, 4 hours, and 10 hours), showing progressive migration from the silicon substrate to the Si$_3$N$_4$ layers.
• Figure 3: Substrate preparation techniques for copper-impurity-free PICs.a, Schematic representation of two fabrication approaches. The left panel illustrates direct gettering for Si wafer purification, where a defective LPCVD or PECVD Si$_3$N$_4$ layer ($\sim$ 200 nm) is deposited on Si wafers without native SiO$_2$. Cu impurities from the Si wafer diffuse into the Si$_3$N$_4$ layer during high-temperature annealing (800$^{\circ}$C, 10 hours), and the contaminated layer is subsequently removed by dry and wet etching. The right panel depicts a Cu diffusion barrier method, where a 200-nm LPCVD Si$_3$N$_4$ layer is deposited directly on Si wafers to prevent Cu migration into the device layer. b, Cu depth profile in Si wafers before and after the direct gettering process, showing near-complete Cu removal. c, Cu depth profile in Si$_3$N$_4$ devices with a diffusion barrier, revealing a Cu peak ($\sim$ 10$^{16}$ atoms/cm$^3$) in the barrier layer, while Cu remains at the detection limit in the waveguide layer.
• Figure 4: Deterministic DKS generation in Si$_3$N$_4$ microresonators fabricated with copper-impurity-free Si substrates using the photonic Damascene process.a, Experimental setup for DKS generation. ECDL: external-cavity diode laser; AWG: arbitrary waveform generator; FPC: fiber polarization controller; EDFA: erbium-doped fiber amplifier; FBG: fiber Bragg grating; ESA: electrical spectrum analyzer; OSC: oscilloscope; OSA: optical spectrum analyzer. b, Single soliton spectra from 9 consecutive measurements in a 40-GHz FSR microresonator, with pump wavelengths filtered for clarity. Device ID: D163_03_F4_C12. c, Soliton steps observed during a continuous wavelength sweep across the microresonator resonance. The inset presents a zoomed-in view of the backward tuning regime, showing successive transitions from 5 solitons to 0 solitons. d, Soliton repetition rate recorded via photodetection and an electrical spectrum analyzer during continuous wavelength sweeping. e, Single soliton spectra from four microresonators with FSRs of 50 GHz, 100 GHz, 150 GHz, and 200 GHz, respectively, fabricated on Cu-gettered Si wafers. All microresonators have a cross-section of 2.2$\times$0.9 $\mu$m$^2$ and are pumped with TE-polarized light. Pump wavelengths are filtered for clarity. Device IDs: D163_03_F1_C2 (50 GHz), D163_03_F1_C3 (100 GHz), D163_03_F1_C6 (150 GHz), and D163_03_F1_C7 (200 GHz).