Multi-objective Optimization of Rotary Travelling Wave Oscillator (RTWO) With Neuro-Genetic Nondominated Sorting Algorithm

Abstract

Rotary travelling wave oscillator (RTWO) represents a transmission line based technology for multi-gigahertz clock generation. RTWO design is a multiparameter-multi-objective problem with tradeoffs of performance measures, power and phase noise. In this paper, non-dominated based genetic algorithm for multi-objective optimization is used to determine the Pareto optimal front of solutions for low power and phase noise with emphasis on variation of transmission line width and spacing. Optimization is followed by sensitivity assessment wherein Monte Carlo simulations and corner analysis are performed on the Pareto points with respect to process variations. The algorithm is validated in the design of RTWO whose frequency varies between 3 to 5GHz due to varying dimensions of coupled transmission line. The optimization is a two step process. A neural network is developed from experimental data to estimate phase noise and power with transmission line width and spacing as inputs. The neural network is then coupled with genetic algorithm for subsequent design optimization. Our results show a set of solutions for width and spacing with objective functions less sensitive to process variations.