Fast Monte-Carlo analysis method of ring oscillators with neural networks

Abstract

Because of the slow speed of SPICE Monte-Carlo (MC) simulation, the limited number of MC samples causes inaccuracy for statistical analysis of ring oscillators. In this paper, we propose the MC simulation method of ring oscillators with artificial neural networks, which shows 5 order faster than SPICE. It is shown that the figure of merits of the ring oscillator can be accurately estimated through simple neural networks training with random samples of SPICE data.