In this work, we aim to optimize the all-optical phase-controlled phase modulators that incorporate lasers, optical directional couplers, optical amplifiers, and phase shifters. Two types of the systems are studied: linear type and sinusoidal type in which the phase of the output light beam of the systems follows a linear and a sinusoidal dependence on the phase difference between the input light and the control light. Using the inverse design approach based on the 2-layer multilayer perceptron through the pseudoinverse of matrix, the coupling ratio of directional couplers and the phase delay of the phase shifters of the systems can be obtained. The standard errors of phase and power between the output modulated light and the target output light are less than those of our previous reports. The non-linear phase-controlled phase modulator could be applied in optical neural networks as an activation function.
Time delay reservoir computing (TDRC) based on microring resonators (MRRs) offers a compact and energy efficient route to photonic neuromorphic processing. However, its computational capability is bottlenecked by traditional schemes that fail to exploit the synergy between the wavelength degree of freedom and spatial depth, thereby capping the accessible nonlinear dimensionality. Here, we propose a highly expressive and variation-tolerant TDRC architecture built upon multi-wavelength cascaded MRRs. In the shallow, single-layer configuration, the multi-resonance of MRRs enables multiple longitudinal modes to jointly generate an enlarged set of virtual nodes with enriched dynamics, significantly expanding processing capacity. Furthermore, a cascaded architecture is introduced where the optical response of a preceding four wavelength MRR is merged with an input bypass and injected into the subsequent one. Comprehensive evaluations are conducted on benchmark tasks, including Santa Fe time-series prediction and nonlinear channel equalization, alongside systematic parameter analyses. Results demonstrate that this configuration yields a multi-fold reduction in normalized mean square error (NMSE) relative to the initial single-wavelength shallow reservoir, and suppresses the symbol error rate (SER) by nearly two orders of magnitude in unstable equalization regimes, offering a scalable, hardware-friendly pathway toward high-performance integrated photonic RC.
Wafer-level optical alignment is a major source of time overhead and uncertainty in silicon photonics testing because of mechanical drift, environmental disturbances, and brute-force scanning. This study presents a prediction-guided framework for reducing the X/Y/Z coupling-search range of a six-axis optical coupling arm under a fixed-angle wafer-level measurement condition. Nineteen repeated full-wafer swept-wavelength measurement records were collected from the same 64-die silicon photonics wafer using an 80 μm × 80 μm scan window. Coupling-validity screening was performed at the optical-path level. Unsuccessful coupling trajectories, missed peaks, and physically inconsistent optical-power responses were removed, while all 19 full-wafer records were retained. Missing entries caused by path removal were imputed using training-derived feature-wise mean values when necessary. The records were partitioned at the record level into 13 training, 4 validation, and 2 testing records. A gated recurrent unit network with Simplified Swarm Optimization-based hyperparameter tuning was used to learn the temporal relationship between X/Y/Z movements and optical-power responses. On the two held-out test records, SSO-GRU achieved R2 values of 0.9019, 0.8453, and 0.9699 for DX, DY, and DZ, respectively (averaged over six coupling regimes), and reduced the mean MAE by 23.8%, 34.0%, and 34.0% compared with the non-optimized GRU model. The results support operation with an 80 μm × 80 μm local scan window relative to a 100 μm × 100 μm reference window, corresponding to a 20% reduction in scan width and a 36% reduction in scan area. Closed-loop throughput validation across additional wafers and platforms remains future work.
Driven by the explosive growth of global data traffic and the rapid development of AI data centers, there is an increasingly urgent demand for electro-absorption modulators (EAMs) with high modulation depth (MD), ultra-wide bandwidth, and compact footprint. However, conventional graphene silicon-on-insulator (SOI) EAMs suffer from intrinsically low MD, and the RC time constant remains a bandwidth bottleneck for most reported devices. In this work, we propose a plasmonics-enhanced graphene EAM built on a standard SOI platform, which achieves an MD of 1.26 dB/μm at 1550 nm, more than 11 times that of conventional graphene SOI EAMs. Moreover, the bandwidth calculated via a lumped-parameter RC circuit model reaches 1.67 THz, indicating that the RC time constant is no longer the bandwidth limitation. The proposed device retains a simple structure with high fabrication feasibility and compatibility with mainstream silicon photonic manufacturing platforms. Benefiting from plasmonic enhancement, the silicon waveguide cross-section is reduced from 520 nm × 340 nm down to only 40 nm × 40 nm. This design provides a promising candidate for high-density, ultra-high-speed photonic integrated circuits.
We demonstrate gain-switched picosecond pulse generation from electrically pumped vertical external cavity surface emitting lasers (EP-VECSELs) emitting around 1030 nm in the low-MHz regime. The devices combine two gain stages joined via a tunnel junction (TJ). A second proton-implanted buried TJ connects the upper gain stage to a top n-type current spreading layer. This approach produced Gaussian electroluminescence near-field profiles for apertures up to 20 μ, and uniform flat-top profiles for larger apertures up to 100 μm. At 5 MHz repetition rate, the gain-switched pulses were single-transverse mode with durations of 73.4–435 ps, and energies of 7.6–403 pJ. To our knowledge, this is the first gain-switched operation of an EP-VECSEL architecture. The presented results establish a path toward compact, electrically driven picosecond sources for LiDAR, sensing, and seeding applications.
