Photonic Multiply-Accumulate Operations for Neural Networks

Abstract

It has long been known that photonic communication can alleviate the data movement bottlenecks that plague conventional microelectronic processors. More recently, there has also been interest in its capabilities to implement low precision linear operations, such as matrix multiplications, fast and efficiently. We characterize the performance of photonic and electronic hardware underlying neural network models using multiply-accumulate operations. First, we investigate the limits of analog electronic crossbar arrays and on-chip photonic linear computing systems. Photonic processors are shown to have advantages in the limit of large processor sizes (>100 µm), large vector sizes (N > 500), and low noise precision (≤4 bits). We discuss several proposed tunable photonic MAC systems, and provide a concrete comparison between deep learning and photonic hardware using several empiricallyvalidated device and system models. We show significant potential improvements over digital electronics in energy (>10 2), speed (>10 3), and compute density (>10 2). Index Terms-Artificial intelligence, neural networks, analog computers, analog processing circuits, optical computing. I. INTRODUCTION P HOTONICS has been well studied for its role in communication systems. Fiber optic links currently form the backbone of the world's telecommunications infrastructure, vastly overshadowing the best electronic communication standards in information capacity. Light signals have many advantageous properties for the transfer of information. For one, a photonic waveguide, with diameters ranging from those in fiber (∼80 μm) to those fabricated on-chip (∼500 nm), can carry information at enormous bandwidth densities-i.e., terabits per secondwith an energy efficiency that scales nearly independent of distance. This density is possible thanks to signal parallelization in photonic waveguides, in which hundreds of high speed, multiplexed channels can be independently modulated and detected. Photonic channels also experience less distortion, jitter,

Scalable reservoir computing on coherent linear photonic processor

“…Here, we discuss the computational efficiency of the proposed circuit. Although there are many indices for expressing the performance of a computing device, the multiply-accumulate per second speed (MAC•s −1 ) is now widely considered to be a milestone in the photonic neuromorphic computation region 9,53,54 . Thus, we discuss this index for the reservoir computer.…”

Overview of Phase-Change Materials Based Photonic Devices

“…As the calculation is performed by measuring the optical transmission of reconfigurable and non-resonant (Figure 16(b)), i.e., broadband, passive components operation at a bandwidth exceeding 14 GHz (Figure 16(c)), the designed photonic core exhibits the computational potential at the speed of light at very low power, thus providing an effective method to remove the computing bottleneck in machine learning hardware for applications ranging from live video processing to autonomous driving and Ai-aided life saving applications. In addition to above findings, Prucnal et al [102][103][104] also proposed a photonic network, i.e., digital electronics and analogue photonics (DEAP), suited for convolutional neural networks based on silicon photonics technologies ( Figure 16(d)). DEAP was estimated to perform convolutions between 2.8 and 14  faster than a GPU while roughly using 25% less energy.…”

Programmable phase-change metasurfaces on waveguides for multimode photonic convolutional neural network

“…We harness the PMMC’s high-precision programmability and in-memory computing capability to demonstrate an optical convolutional neural network (OCNN) 28 – 30 , 48 . A typical CNN consists of an input layer and an output layer, which are connected by multiple hidden layers in between.…”