Structure-aware high-dimensional performance modeling for analog and mixed-signal circuits

Abstract

Efficient high-dimensional performance modeling of nanoscale analog and mixed signal (AMS) circuits is extremely challenging. In this paper, we propose a novel structure-aware modeling (SAM) technique. The key idea of SAM is to accurately solve the model coefficients by applying an efficient statistical algorithm to exploit the underlying structure of AMS circuits. As a result, SAM dramatically reduces the required number of sampling points and, hence, the computational cost for performance modeling. Several circuit examples designed in a commercial 32nm CMOS process demonstrate that SAM achieves more than 2× runtime speedup over the traditional sparse regression technique without surrendering any accuracy.