Arbitrary waveform generator based on machine learning and design method thereof

The invention relates to an arbitrary waveform generator based on machine learning and a design method thereof. The arbitrary waveform generator comprises a silicon substrate layer, a silicon dioxide layer and a silicon design layer which are sequentially arranged from bottom to top, a pixel topological pattern of a middle pixel topological pattern design area of the silicon design layer is obtained by adopting an arbitrary waveform generator reverse design method based on machine learning; the method comprises the following steps: acquiring an expected waveform generator transmitted spectrum as the input of a reverse prediction multilayer sensor, and acquiring a pixel topological pattern of a waveform generator design area; wherein when the reverse prediction multilayer perceptron is trained, a forward simulation multilayer perceptron is trained, then parameters of the forward simulation multilayer perceptron are fixed, the reverse prediction multilayer perceptron and the forward simulation multilayer percep