Light speeds up matrix multiplication f

0

Image: a, concept of a photonic accelerator with photonic matrix multiplication. b, Photonic matrix multiplication methods. c, Schematic diagram of optoelectronic-hybrid AI computing chip framework.
outlook more

Photo credits: by Hailong Zhou, Jianji Dong Junwei Cheng, Wenchan Dong, Chaoran Huang, Yichen Shen, Qiming Zhang, Min Gu, Chao Qian, Hongsheng Chen, Zhichao Ruan, and Xinliang Zhang

There is an ever-growing demand for fifth-generation artificial intelligence and communications worldwide, resulting in very large computing power and storage requirements. The slowdown or even failure of Moore’s Law makes it increasingly difficult to improve their performance and energy efficiency by relying on advanced semiconductor technology. Optical devices can have super wide bandwidth and low power consumption. And light has an ultra-high frequency of up to 100 THz and multiple degrees of freedom in its quantum state, making optical computing one of the most competitive candidates for high-capacity, low-latency matrix information processing in the “More than Moore” era. In recent years, photonic matrix multiplication has been rapidly developed and widely used in photonic acceleration fields such as optical signal processing, artificial intelligence, and photonic neural networks. These applications based on matrix multiplication show the great potential and possibilities in photonic accelerators.

In a new review published in Light Science & Application, a team of scientists led by Professor Jianji Dong of Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology in China and collaborators presented photonic matrix multiplication methods and summarize the milestones in the development of photonic matrix multiplication and related applications. Then, their detailed advances in applications in optical signal processing and artificial neural networks in recent years were reviewed. Comments on the challenges and prospects of photonic matrix multiplication and photonic acceleration were also discussed

The paper reviewed and discussed the progress of photonic accelerators from a unique perspective of photonic matrix multiplication. These scientists summarize the main content of this review:

“The methods for photonic matrix-vector multiplication (MVMs) mainly fall into three categories: the planar light conversion (PLC) method, the Mach-Zehnder interferometer (MZI) method and the wavelength division multiplexing (WDM) method.”

“The photonic matrix multiplication network itself can be used as a general linear photonic loop for photonic signal processing. In recent years, MVM has evolved into a powerful tool for a variety of photonic signal processing methods.”

“AI technology is widely used in various electronics industries, such as deep learning-based speech recognition and image processing. MVM as the basic building block of ANNs occupies most of the computing tasks, e.g. B. Over 80% for GoogleNet and OverFeat models. Improving MVM performance is one of the most effective means of ANN acceleration. Compared to electrical computing, optical computing is poor at data storage and flow control, and the low efficiency of optical nonlinearities limits applications in nonlinear computing, such as B. activation functions. While it offers significant advantages in massively parallel computing through wavelength, mode and polarization multiplexing strategies, it offers extremely high data modulation speeds of up to 100 GHz. So photonic meshes are pretty good at MVM. By combining optical computing and AI, intelligent photonic processors and photonic accelerators are to be realized. AI technology has also developed rapidly in the field of optics in recent years.”

“In general, photonic computing has obvious advantages in terms of signal rate, latency, power consumption, and computational density, and its accuracy is generally lower than that of electrical computing.”

“Before all-optical ANNs mature, especially in optical nonlinear effects and optical cascades, hybrid optoelectronic AI is a more practical and competitive candidate for deep ANNs. Therefore, the development of a highly efficient and dedicated optoelectronic hybrid AI hardware chip system is one of the core research paths of photonic AI.”


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of the press releases published on EurekAlert! by contributing institutions or for the use of information about the EurekAlert system.

Share.

Comments are closed.