In the post Moore's era, conventional electronic digital computers have encountered escalating challenges in supporting massively parallel and energy-hungry artificial intelligence (AI) workloads, which raises a high demand for a revolutionary AI computing solution. Optical neural network (ONN) is a promising hardware platform that could represent a paradigm shift in efficient AI with its ultra-fast speed, high parallelism, and low energy consumption. In recent years, efforts have been made to facilitate the ONN design stack and push forward the practical application of optical neural accelerators. In this paper, we present a holistic solution with state-of-the-art cross-layer co-design methodologies towards scalable, robust, and self-learnable integrated photonic neural accelerator designs across the circuit, architecture, and algorithm levels. We will introduce (1) an area-efficient butterfly-style ONN architecture design beyond traditional general tensor units, (2) model-circuit co-optimization that boosts variation-tolerance and endurance of photonic in-memory computing, (3) efficient ONN on-chip training algorithms that enable self-learnable photonic AI engines, and (4) AI-assisted automated photonic integrated circuit (PIC) design methodology beyond manual PIC designs in footprint, expressivity, and noise-tolerance. Our proposed ONN design stack is integrated into our open-source PyTorch-centric ONN library TorchONN to construct customized photonic AI engine designs and perform high-performance ONN training and optimization.

Optical neural network (ONN) is a promising platform for implementing deep learning tasks thanks to the critical features of light, such as high parallelism, low latency, and low power consumption. Previous ONN architectures are mainly composed of arrays of single-operand photonic devices such as Mach-Zehnder Interferometer (MZI) or microring resonator arrays. However, as the size of deep neural networks (DNNs) continues to grow, these ONNs will encounter unnecessary hardware costs, such as large chip areas and high power consumption. In this work, we devise several compact customized multi-operand active photonic components for tensor operations, for example, multi-operand ring modulators, to reduce the hardware cost of optical AI accelerators. Furthermore, we propose ONN architectures based on these multi-operand active photonic components. Compared to previous ONNs based on single-operand MZI or microring arrays, our work uses fewer optical and electrical components to implement matrix multiplications with comparable task performance. Finally, we experimentally demonstrate the utility of our proposed ONN architectures based on multi-operand photonic devices in several deep learning tasks.

The exponential growth in data center traffic which creates a need for a paradigm shift in I/O technology that meets future connectivity requirements within data centers. Silicon Photonics (SiPh) based optical interfaces significantly improve I/O density by optimizing solutions along three technology vectors independently: Packaging density, speed per lane, and number of wavelengths per channel. This paper shows the necessary technologies for a SiPh based optical I/O solution that merges mature silicon chiplet packaging and fiber connectivity to achieve the highest I/O efficiency for networking applications. An early Broadcom prototype system is demonstrated.

Photonic Crystal Surface Emitting Lasers (PCSELs) have emerged as a third-generation diode laser technology for high power high brightness high speed lasers with applications in LiDAR and 3D sensing, datacom and lasercom, photonic and quantum integrated circuits. In this talk, we report PCSEL performance trade-offs and challenges in power scaling, modal competition, and charge injection control. We will report the impact of electron block layers in the heterostructure design, the symmetry breaking strategy in the optical cavity design, and the regrowth/hybrid integration trade-offs. Linewidth and watts level experimental results will be presented and discussed.

We develop a 2-D non-periodic optical phased array for beam steering in the mid-infrared spectral region in an InGaAs-InP platform. The device consists of an MMI light-splitting tree, thermally isolated thermo-optic phase shifters, and total internal reflection mirror emitters. The device achieves X resolvable points across a steering range of ±14° in both the longitudinal and azimuthal directions.

We have proposed a novel graded-index plastic optical fiber (GI POF) to achieve stable data transmission in a multimode fiber link, which is capable of error-free data transmission (bit error rate < 10^-12) employing four-level pulse-amplitude modulation at data rate of 53 Gb/s without the use of forward error correction. The novel GI POF has strong mode coupling due to the microscopic heterogeneities in the fiber core material, which decreases noise and stabilizes data transmission. Here, we demonstrate that the novel GI POF significantly improves fiber-misalignment tolerance for stable data transmission. The novel GI POF paves the way for an unprecedented optical interconnect that enables reliable data transmission without the requirement of precise fiber alignment and high-precision fiber connectors.

Co-packaged optical modules aim to meet the increasing bandwidth and power reduction requirements in next-generation datacenter switches. Meeting the cost per capacity targets requires new and innovative wafer-scale manufacturing solutions. In this work, glass with low-loss (<0.1 dB/cm) single-mode ion-exchanged waveguides is proposed as an optoelectronic substrate for co-packaged optics. High-speed ultrafast laser processes are developed to fabricate through glass vias for electrical connections, ablated features to enable passive alignment of MPO-connector to chip coupling, and for the singulation of glass wafers into individual optical circuits with optical quality end-facets for low-loss edge coupling without subsequent post-polishing or finishing steps.

This presentation in on automatic assembly solutions for optical interconnecting integrated waveguides.

Micro-resonator modulators working in the critical coupling mode is usually sensitive to fabrication variation. We employ two modulated racetrack-resonators symmetrically coupled to a waveguide on two sides on a silicon-on-insulator wafer. The structure in the strong-coupling regime can perform modulation function stably through the interference of two racetracks. Fabrication variation (30~50%) of coupling constant or quality factor can be compensated by static voltage bias applied on the diode-embedded resonators. Detailed working principle and fabrication-variation compensation will be presented. Experimentally, we demonstrate 50 ~ 56 Gb/s modulation with high extinction ratio of 7.9 ~ 9.4 dB and high signal-to-noise ratio of 6~7, while maintaining low driving voltage (<2.5 Vpp) and small size (tens of microns).

Photonic technologies are being investigated to realize various next-generation devices, including quantum emitters, transducers and detectors, as well as classical devices, such as lasers, modulators, and detectors. A packaging approach that maintains high optical coupling efficiency over a wide temperature and broad wavelength range is needed for these devices and others. Photonic wire bonds (PWBs) may provide greater design flexibility, increased manufacturability, and higher tolerances to thermally driven misalignment at cryogenic temperatures. Freedom Photonics (a Luminar company) is developing high bandwidth optical interconnects for chip-to-chip or chip-to-fiber applications, utilizing PWBs to reduce optical loss at cryogenic temperatures.

Silicon microring resonator (Si-MRR) plays an essential role in the scalable wavelength division multiplexing (WDM) device due to its ultra-compact size and low energy consumption. Typically, the resonant wavelength of the Si-MRR is controlled via the thermal heater with high power dissipation. Recently, a tunable Si-MRR heterogeneously integrated with high mobility transparent conducting oxide (HMTCO)/hafnium oxide insulator/Si metal-oxide-semiconductor (MOS) capacitor has demonstrated an alternative method to electrically tune the working wavelength with negligible power consumption. This talk will demonstrate a WDM cascaded with four HMTCO gated MOS Si-MRRs and show the large wavelength tuning range with high power efficiency.