Silicon Photonic Programmable Optical Processors for Machine Learning and Optical Quantum Computing

Abstract. Programmable optical processors are promising structures for ultrafast and energy efficient computation in classic and quantum photonics. These processors efficiently perform the vector-matrix multiplication extensively used in artificial intelligence and machine learning tasks. Due to the inherent parallelism presents in optics in contrast with sequential operations in electronics, optical processors offer better energy efficiency compared to their electronic counterparts. Today, deep learning is facing growing computational demand limiting its development if we continue using conventional electronic processors. Energy efficient computational accelerators fabricated in silicon photonic are candidates to meet the computational demands of future machine learning and deep learning tasks. Programmable optical processors also pave the way for integrated optical quantum computing. Single photons are excellent candidates for quantum computing due to their noise and decoherence-free nature. One can generate optical qubits by encoding single photons in one degree of freedom such as polarization or path. Optical integrated quantum computing requires quantum logic gates to manipulate these qubits. Quantum logic gates are represented by unitary matrices, such that a 2n × 2n unitary matrix multiplication is identical to an n-qubit gate. Therefore, a programmable optical processor capable of performing unitary matrix multiplication on single photons works as an arbitrary optical integrated quantum gate. Speaker(s): Dr. Kaveh Mojaver, Room: TR0070, Bldg: Lorne M. Trottier Engineering Building, McGill University, 3630 University Street, Montreal, Quebec, Canada, Virtual: https://events.vtools.ieee.org/m/310651