Efficient Synthesis Methods for High-Frequency Integrated Passive Components and Amplifiers

Abstract

Existing design automation methods for RF ICs and microwave passive components often rely on parasitic-aware lumped equivalent circuit models. That framework is difficult to apply to synthesis tasks at high frequencies (e.g. 40GHz and above) due to the distributed effect. When directly embedding the computationally expensive electromagnetic (EM) simulations in the optimization loop, a too long synthesis time results. This chapter presents a new method for highfrequency integrated passive component synthesis, called Memetic Machine Learning-based Differential Evolution (MMLDE), and the first method for mm-wave integrated circuit synthesis, called Efficient Machine Learning-based Differential Evolution (EMLDE), both addressing the problem of obtaining highly optimized design solutions in a very practical time. The common idea of these two methods is the on-line surrogate model assisted evolutionary algorithm (SAEA), where a computationally cheap surrogate model is constructed adaptively in the optimization process to replace expensive EM simulations. The differences between the two algorithms are that a memetic SAEA is built to enhance the optimization ability and efficiency in MMLDE, while a decomposition method is used to address the ”curse of dimensionality” of SAEA in EMLDE. Experimental results show the effectiveness and the high efficiency obtainable with MMLDE and EMLDE.