Application Of Deep Learning In Analog Circuit Sizing

Operational amplifiers (opamp) are the most common structures in analog circuit. However, opamp sizing problems rely heavily on the designer's personal experience. One problem that hampers opamp's automatic sizing is that the search space for the opamp size is too large, making the search process time consuming. In recent years, with the rapid development of GPU-based deep learning technology, it has become possible to automatically explore the sizing problem of analog circuits through the machine; that is, given the performance index, the machine is required to automatically output the length and width of each transistor and the output circuit performance should be close to the input performance specifications. In this paper, we use two neural network structures, Recurrent Neural Network (RNN) and Deep Neural Network (DNN), to construct two deep learning models that can automatically solve the sizing problem. The final simulation results show that our model can accurately predict the size of the transistor, where the accuracy of the RNN model is 92.6%, and the accuracy of the DNN is 95.6%.