On-chip photonic neural network device based on cascaded optical coupler, chip and application method thereof

Abstract

The invention discloses an on-chip photonic neural network device based on a cascaded optical coupler, a chip and an application method of the on-chip photonic neural network device. The photonic neural network device comprises an all-optical photonic neural network. The all-optical photon neural network comprises an amplitude modulation unit, multiple layers of neurons and a detection unit. The amplitude modulation unit comprises N0 amplitude modulators; the multi-layer neuron comprises M repetitive units, wherein each repetition unit comprises a plurality of amplitude modulation phase modulators and an MIMO optical coupler, the plurality of amplitude modulation phase modulators of any mth repetition unit are used for performing amplitude modulation and phase modulation on Nm-1 paths of input optical signals respectively and then obtaining Nm paths of output through the MIMO optical couplers, and the detection unit comprises NM paths of detectors. According to the invention, an all-optical photon neural network deep learning system with large bit width input can be realized in an on-chip integrated system on the basis of the cascaded optical coupler, the calculation overhead of aforward reasoning process after training learning is completed is extremely low, the power consumption is low, the on-chip system is easy to integrate, and large bit width data input is supported.

Description

technical field [0001] The present invention relates to neural network deep learning technology, in particular to an on-chip photonic neural network device, chip and application method based on cascaded optical couplers, which are used to realize large An all-optical photonic neural network deep learning system with bit-wide input. Background technique [0002] Artificial intelligence is a new technology science that studies and develops the theory, method technology and application system for simulating and expanding human intelligence. On the one hand, the development of algorithms, computing power, and big data drives artificial intelligence to make breakthroughs in computing intelligence, perceptual intelligence, and cognitive intelligence from three dimensions; on the other hand, artificial intelligence is rapidly integrating with various industries, helping traditional The upgrading and transformation of the industry, the improvement of quality and efficiency, have tr...

AI Technical Summary

Problems solved by technology

However, because its realization principle is a natural two-dimensional convolution calculation, it cannot be applied to non-two-dimensional convolutional neural network algorithms. In practical applications, convolutional neural networks only account for less than 10% of all neural network applications.

On the other hand, this solution is based on the principle of space optics, and it is difficult to integrate it into a system-on-chip with dimensionality reduction.

Images