Photon two-dimensional convolution acceleration method and system based on time-wavelength interleaving

Abstract

The invention discloses a photon two-dimensional convolution acceleration method and system based on time-wavelength interleaving, and the method comprises the steps: firstly, loading a to-be-convolved signal to a multi-wavelength optical signal containing M * N wavelengths, and dividing the multi-wavelength optical signal into M sub-optical signals each containing N wavelengths through a wavelength division demultiplexer; the M sub optical signals are sent to the delay weighted micro-ring array after being sequentially increased with equal interval delays, and coupling coefficients of M * N micro-rings in M delay weighted micro-ring units in the delay weighted micro-ring array are controlled to achieve amplitude weighting of M * N wavelength signals respectively; the M sub-optical signals after amplitude weighting are subjected to wavelength division multiplexing to form a composite optical signal, and then photoelectric conversion is completed to obtain a corresponding characteristic signal. According to the method, light is used as an information carrier, two-dimensional convolution kernel convolution acceleration operation of two-dimensional data can be realized in a single signal period based on two-stage delay and two-dimensional micro-ring array weighting, and the speed and the energy efficiency ratio of convolution operation are greatly improved.

Description

technical field [0001] The invention relates to an artificial intelligence-oriented photonic neural network convolution acceleration method, in particular to a photonic neural network convolution acceleration system, which belongs to the field of photonic computing. Background technique [0002] Artificial intelligence is now widely used in the fields of machine vision, natural language processing, and autonomous driving. Among them, artificial neural networks, one of the important models of artificial intelligence technology, are widely used because of their excellent generalization ability and stability. Because the current electronic chip adopts a classical computer structure that separates the program space and the data space, the data load between the storage unit and the computing unit is unstable and the power consumption is high, which limits the efficiency of network model training. The common solution is to improve the computing efficiency by improving the integrat...

AI Technical Summary

Problems solved by technology

[0003] The technical problem to be solved by the present invention is: to overcome the deficiencies of the existing technology, use the microring resonator array including the delay waveguide combined with the delay array to realize the weighting of the two-dimensional convolution kernel matrix coefficients of the signal to be convoluted, and solve the traditional method of data redundancy, The problem of partial data loss, and the convolution kernel matrix can be flexibly expanded, suitable for multi-dimensional data convolution operations

Images