The invention discloses a terahertz on-
chip diffraction neural network with a nonlinear function, which comprises an input end, an output end and a neural network array which are etched on a
quartz substrate, and is characterized in that the neural network array is formed by connecting a plurality of
diffraction layers and phase shift
layers at intervals along the length direction of the
quartz substrate; wherein a nonlinear layer is connected between at least one phase shift layer and a
diffraction layer behind the phase shift layer, the input end is connected with an input
signal channel of the neural network array through a
microstrip line, received terahertz
waves carrying information are transmitted into the neural network array, the diffraction layer is a
branch line coupler array, the phase shift layer is a
microstrip phase shift line, and the
microstrip phase shift line is connected with the input
signal channel of the neural network array. The nonlinear layer is composed of a plurality of
Schottky barrier diodes and provides nonlinear operation for the neural network, and the output end is connected with an output
signal channel of the neural network array. Through the on-
chip structure integrated with the
Schottky barrier diode, the nonlinear operation of the terahertz
wave band diffraction neural network is realized.