Artificial neural networks for accurate high frequency CAD applications

A unique approach for applying neurocomputing technology for accurate high-frequency CAD of circuits is described. In our proposed method, a full-wave electromagnetic (EM) analysis is employed to rigorously characterize monolithic IC passive elements. Equivalent circuit parameters (ECPs) are extracted from these EM results and are used to train a multilayer perceptron neural network (MLPNN). To demonstrate this technique, the π-network for 32 different spiral inductors is modeled by a single neural network. The MLPNN computed ECP values in excellent agreement with the extracted ECPs. The neural networks ability to generalize and predict accurate ECPs for inductors outside the training set is also demonstrated