Photonic and optoelectronic neuromorphic computing

Abstract

Recent advances in neuromorphic computing have established a computational framework that removes the processor-memory bottleneck evident in traditional von Neumann computing. Moreover, contemporary photonic circuits have addressed the limitations of electrical computational platforms to offer energy-efficient and parallel interconnects independently of the distance. When employed as synaptic interconnects with reconfigurable photonic elements, they can offer an analog platform capable of arbitrary linear matrix operations, including multiply–accumulate operation and convolution at extremely high speed and energy efficiency. Both all-optical and optoelectronic nonlinear transfer functions have been investigated for realizing neurons with photonic signals. A number of research efforts have reported orders of magnitude improvements estimated for computational throughput and energy efficiency. Compared to biological neural systems, achieving high scalability and density is challenging for such photonic neuromorphic systems. Recently developed tensor-train-decomposition methods and three-dimensional photonic integration technologies can potentially address both algorithmic and architectural scalability. This tutorial covers architectures, technologies, learning algorithms, and benchmarking for photonic and optoelectronic neuromorphic computers.

Atomically Thin Synaptic Devices for Optoelectronic Neuromorphic Vision

“…A CNN model has features of segregate layers, modularized kernels and combine weight sharing and sparse connectivity to perform fast and accurate feature recognition with far fewer weights than conventional networks. [50,51] Herein, the adopted CNN architecture is composed of input images of 10 different classes with the size of 32 × 32 × 3 and three convolutional layers (Figure 4a). [52] Each convolutional layer with kernel of 3 × 3 weights is followed by ReLu activation function and maxpooling layer to avoid overfitting.…”

Scalable Nanophotonic-Electronic Spiking Neural Networks

“…MZI meshes can perform unitary matrix transformations that correspond to lossless multiplication and are thus particularly suitable for low-power neuromorphic computing. See [36] for a longer discussion on the design trade-offs between each of these devices. Section III-B describes our MZI mesh architecture, while Section III-C details algorithms for training SNNs using MZI meshes.…”

Implementation of input correlation learning with an optoelectronic dendritic unit

“…Some proposals aimed at addressing challenges in analog computing [20,21,29], while others aimed at implementing spiking networks with timedependent plasticity (STDP) [22,23,30,31]. While some of these works were mainly based on proof-of-concept photonic architectures, many works lately focus on the potential of these designs towards scalable, energy-efficient, and robust computing accelerators [32][33][34][35][36].…”