Large-scale neuromorphic systems enabled by integrated photonics

In this chapter, we first introduce the photonic implementations of synapses and neurons for weight loading and nonlinear activation. On this basis, photonic integrated neuromorphic systems are reported with different architectures. Since the advantages on latency and efficiency are much more significant in large-scale photonic systems, we mainly focus on the large-scale and fully integrated photonic architectures with both the linear and nonlinear units integrated on chip. Besides, training method with real-time, in situ, and photonic implemented capabilities is another critical issue for large-scale photonic neuromorphic systems. We present various solutions for photonic training and finally analyze the opto-electronic cointegration and copackaging technologies for large-scale neuromorphic photonic processors toward ultrahigh density and energy efficiency.