Study on silicon photonic devices for photonic neural network

Abstract

With the rapid development of artificial intelligence, the traditional computer architecture can no longer meet the growing computing performance requirements. At the same time, with the chip manufacturing process approaching the physical limit, a single electronic technology cannot adapt the rapid development of artificial intelligence chips, so there is an urgent need for new computing chips. Photonic neural network chip, which combines artificial intelligence, silicon photonic, integrated circuit and other technologies, will get unprecedented opportunities for the development. Silicon based opto-electronic integration is a large-scale integration technology with optical signal as the main information carrier. It can integrate micro-nano-size optical and electrical devices on the silicon substrate, to form a new large-scale integration chip with comprehensive functions. At present, the development of silicon photonic devices is mainly focused on the field of optical communication and data center, while silicon photonic devices for photonic neural networks are still in the initial stage. Starting from the underlying unit devices, silicon-based photonic devices were studied deeply by combining the artificial neural network with the silicon photonic technology in this paper. Based on 200 mm CMOS process, a lot of process modules for photonic neural network were developed. According to the characteristics of photonic neural network architecture and the performance requirements for the basic unit devices, a series of silicon photonic devices, such as waveguides, grating couplers, MMI, thermal modulators, and other unit devices, were designed and developed. These devices provide important basic conditions for the implementation of high performance photonic neural network chips.