Analog circuit performance bound estimation based on extreme value theory

Estimation of performance bound with respect to design parameters play an important role in analog circuit design. The challenge is that circuit performance among design parameter space is normally non-parametric distributed, which is difficult to model or fit with known probability distributions. This paper present a novel performance bound estimation method based on extreme value theory. The method first generates generalized extreme value (GEV) distribution of performance parameter through sampling. Then the performance bound is estimated on the GEV distribution. Experimental results show that this method achieves less than 5% estimation error within 1,000 circuit simulations.