I-Corps: Translation Potential of a Photonic Optical Processing Unit for Quantum-Enhanced Machine Learning and Sensing

This I-Corps project is based on the development of a computer chip that performs artificial-intelligence calculations with light instead of electricity. Artificial intelligence consumes large and rapidly growing amounts of energy. This technology addresses that problem with a chip that encodes information onto many colors of laser light traveling in a single optical channel that lets the optical material itself carry out the mathematical operations, all at speeds and energy efficiencies that electronics cannot match. In addition, the chip is manufactured with standard photonic foundry processes, and it can be produced at scale and integrated into existing datacenters, telecommunications, and sensing infrastructure. Potential commercial applications include energy-efficient data centers, real-time signal processing for wireless communications and radar, and high-speed pattern recognition for medical and industrial sensing. This may reduce energy consumption and infrastructure cost while expanding access to high-performance computing. This I-Corps project utilizes experiential learning coupled with first-hand investigation of the industry ecosystem to assess the translation potential of a monolithic photonic optical processing unit implemented on thin-film lithium niobate. This technology encodes complex-valued weight vectors onto the lines of an electro-optically generated frequency comb. It programs their amplitudes and phases with an on-chip waveshaper composed of add-drop microring selectors, Pockels phase shifters, and push-pull electro-optic Mach-Zehnder attenuators, and performs computation through quasi-phase-matched second-harmonic and sum-frequency generation in a periodically poled waveguide. The output natively realizes the convolution of the programmed weight vector in a single optical pass. Preliminary results in a proof-of-principle nonlinear photonic neural network achieved competitive classification accuracy using frequency-comb encoding and second-order nonlinear mixing. In addition, this work has been extended to a fully integrated chip-scale design with verified device models spanning comb generation, spectral shaping, nonlinear conversion, and output demultiplexing. Compared with interferometer-mesh photonic processors, the footprint and calibration complexity scale linearly with vector size, and every actuator is field-linear and gigahertz-capable. Early target customers are quantum computing companies and defense/aerospace sensing programs seeking quantum-enhanced inference and detection capabilities that classical hardware cannot replicate. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.