Two-dimensional photon neural network convolution acceleration chip based on series structure

The invention discloses a two-dimensional photon neural network convolution acceleration chip based on a series structure. The chip is integrated by a modulator, M micro-ring delay weighting units, M-1 secondary delay waveguides, a wavelength division multiplexer and a photoelectric detector. According to the method, any convolution kernel matrix coefficient weighting of the to-be-convolved signal is realized based on the micro-ring resonator array, the adjustment speed is high, and the optimal convolution kernel matrix is fed back, trained and extracted in real time; the micro-ring resonator is cascaded with the integrated waveguide to realize first-stage wavelength-time interleaving of sub-weighted modulation signals with different wavelengths, and the second-stage delay waveguide is serially connected with the micro-ring delay weighting unit to realize second-stage wavelength-time interleaving of the sub-weighted modulation signals with different wavelengths, so that delay multiplexing of the waveguide is