Fig. 1. Process flow and sample images of the 6-inch Si3N4 foundry fabrication process. (a) Photograph of dozens of Si3N4 chips on a 6-inch wafer, which contains microresonators of different FSR and meter-long spirals. (b) Optical micrograph showing a curved bus waveguide slowly approaching a 100-GHz-FSR microring resonator. (c) SEM image showing the Si3N4 waveguide core with SiO2 cladding. The TE00 mode is plotted, showing tight confinement in the Si3N4 waveguide core. (d) DUV subtractive process flow. WOX, wet oxide (SiO2).

Fig. 2. Statistical loss characterization and yield analysis. (a) Typical TE00 resonance profile with a Lorentzian fit, showing κ0/2π=13.8  MHz and negligible mode split. (b) Histogram of 11,741 TE00 resonances from sixty 100-GHz-FSR microresonators of 2.40 μm waveguide width, showing the most probable value of κ0/2π=14  MHz and Q0=1.4×107. (c) Characterization of waveguide-width-dependent loss. Microresonators of 2.40, 2.20, 2.00, and 1.80 μm waveguide widths are characterized and compared. A trend of lower κ0 with a larger width is shown. The size and color tone of the circles indicate the probability of occurrence. (d) Uniformity and yield analysis over the 6-inch wafer scale. Right, the DUV stepper reticle layout contains 16 chips and is uniformly exposed in discrete fields over the 6-inch wafer. Left, the most probable values κ0/2π of the C11 chips, as well as the measured GVD parameters D2/2π, are marked in each field over the wafer. NA: not applicable, due to visible defects or missing C11 chips near wafer edge.

Fig. 3. Characterization of microresonator dispersion and coupling ideality. (a) Measured integrated dispersion of the microresonator that is fitted with Dint(μ)=D2μ2/2+D3μ3/6+D4μ4/24 (top), the resonance frequency deviations from D3μ3/6+D4μ4/24 (middle), and the deviations from D4μ4/24 (bottom). Avoided mode crossings are revealed in the bottom panel, however weak for the later soliton generation experiment. (b) Characterization of coupling ideality of the TE00 (top) and TM00 (bottom) modes. For the TE00/TM00 modes, in total thirty-four/seventeen 100-GHz-FSR microresonators are characterized, providing 3269/959 data points in each plot. A clear trend from under-coupling to critical coupling and then to strong over-coupling is observed. The calculated curves of I=0.9,0.95,1 with κ0/2π=14 or 17 MHz are plotted for comparison, showing near-unity coupling ideality.

Fig. 4. Single-soliton generation in silicon nitride. (a) Experimental setup. AFG, arbitrary function generator; ECDL, external-cavity diode laser; EDFA, erbium-doped fiber amplifier; BPF, bandpass filter; EOM, electro-optic modulator; DUT, device under test; FBG, fiber Bragg grating; PD, photodiode; OSA, optical spectrum analyser; OSC, oscilloscope; VNA, vector network analyzer. (b) When the laser frequency is scanned from the blue-detuned to the red-detuned side of a resonance, a soliton step of sub-millisecond length appears, enabling direct access to soliton states via simple piezo tuning of laser frequency. (c) Cavity response measurement using the EOM and VNA. The appearance of S-resonance verifies soliton generation. (d) Single-soliton spectra of 100.17 GHz mode spacing, with 19 mW (red) and 126 mW (blue) CW pump power on the chip, and their spectral fit (green). With 19/126 mW power, the arrows mark the 3 dB bandwidth of 17.99/35.68 nm. A prominent Raman self-frequency shift of 10.4 nm is observed with 126 mW power. (e) Single-soliton spectrum of 19.975 GHz mode spacing with 518 mW (blue) CW pump power on the chip and its spectral fit (green). The arrows mark the 3 dB bandwidth of 21.96 nm, containing 137 comb lines. In both (d) and (e), the EDFA’s amplified spontaneous emission (ASE) noise is filtered out by the BPF, and the pump laser in the soliton spectra is filtered out by the FBG.

Fig. 5. Characterization of the microresonator TM00 mode. (a) Typical TM00 resonance profile with a Lorentzian fit, showing κ0/2π=15.6  MHz. (b) Histogram of 7944 TM00 resonances from forty 100-GHz-FSR microresonators of 2.40 μm waveguide width, showing the most probable value of κ0/2π=17  MHz and Q0=1.1×107. (c) Uniformity and yield analysis over the 6-inch wafer scale. Right, the DUV stepper reticle layout contains 16 chips and is uniformly exposed in discrete fields over the 6-inch wafer. Left, the most probable values κ0/2π of the C11 chips, as well as the measured GVD parameters D2/2π, are marked in each field over the wafer. NA: not applicable, due to visible defects or missing C11 chips near wafer edge.

Fig. 6. Broadband measurement of resonance linewidth. Measured and fitted κ0/2π, κex/2π, and κ/2π=(κ0+κex)/2π of each resonance from 1480 to 1640 nm. The alignment of κex/2π values on a line indicates correct resonance fit with reasonable precision. Local κ0/2π increase at multiple wavelengths is likely caused by AMXs. No prominent hydrogen-related absorption around 1520–1540 nm is observed.

Fig. 7. Illustration of the CMP dishing effect observed during fabrication. (a) SEM images showing smooth top surface after CMP for waveguides with smaller critical dimension (e.g., below 3 μm). (b) SEM images showing rough top surface due to the CMP dishing effect for waveguides with larger critical dimension (above 3 μm).