Optical computing

Optical computing offers the potential to build high-throughput, low-energy, and faster computing systems. The recent success of large language models has highlighted a future where massive tensor computation power is essential. Efforts across various disciplines are converging to develop low-latency tensor processor platforms with minimal energy consumption. By leveraging multiplexing in time, space, and wavelength, we developed a Hypermultiplexed Integrated Tensor Optical Processor (HITOP) using Vertical Cavity Surface Emitting Lasers (VCSELs) and integrated Thin-film Lithium Niobate (TFLN) modulators to perform tensor multiplications[1]. Moreover, we are actively developing photonic tensor processor with larger scale based on balanced photodetector arrays. By co-packaging with high bandwidth memory, we aim to realize tensor processors with ultra-low energy consumption.

Phase-Change Material (PCM) optical memory, a novel data storage solution, is now gaining traction as a critical component in the optical computing landscape. PCM relies on materials that can switch between amorphous and crystalline states when exposed to certain light patterns, allowing data to be stored and read using light pulses rather than electricity. This shift not only offers speed and efficiency gains but also enables memory systems that consume far less energy. We envision PCM optical memory being integrated into future optical processors, complementing advancements in speed with a durable and efficient storage solution essential for next-generation computing.

Our work in this area is supported by DARPA NanoWatt Platforms for Sensing, Analysis and Computation (NaPSAC).

[1]Ou, S., “Hypermultiplexed Integrated-Photonics-based Tensor Optical Processor”, Art. no. arXiv:2401.18050, 2024. doi:10.48550/arXiv.2401.18050.