Evelyn Cristina de Oliveira Lima
◽
André Borges Cavalcante
◽
João Viana Da Fonseca Neto
One important step of the optimization of analog circuits is to properly size circuit components. Since the quantities that define specification may compete for different circuit parameter values, the optimization of analog circuits befits a hard and costly optimization problem. In this work, we propose two contributions to design automation methodologies based on machine learning. Firstly, we propose a probability annealing policy to boost early data collection and restrict electronic simulations later on in the optimization. Secondly, we employ multiple gradient boosted trees to predict design superiority, which reduces overfitting to learned designs. When compared to the state-of-the art, our approach reduces the number of electronic simulations, the number of queries made to the machine learning module required to finish the optimization.
Machine Learning-based Defect Coverage Boosting of Analog Circuits under Measurement Variations
