A Fully Automated Method to Create Monte-Carlo MOSFET Model Libraries for Statistical Circuit Simulations

J. Wang, H. Trombley, J. Watts, M. Randall, R. Wachnik
IBM Semiconductor Research and Development Center, US

Keywords: MOSFET, modeling, Monte-Carlo, variation, statistical simulation


As the MOSFET continues to shrink, device-to-device variations become increasingly important. Therefore, it is critical to develop accurate, Monte-Carlo (MC) models that capture various device variations to allow statistical circuit simulations. To model the global (or ‘chip-mean’) variations in a MOSFET, a Gaussian distribution needs to be enabled for each selected model parameter and its sigma-value (tolerance) is adjusted so that the output of the MC model matches the upper/lower bounds of a number of critical device metrics. To speed up the model parameter tolerance extraction, a Backward Variation Propagation (BVP) method was introduced in the literature. However, with potential inconsistencies in device variation metrics and/or model parameter sensitivities, BVP may not always return a valid solution. Therefore, manual optimization is normally needed when creating a MC library using BVP. In this paper, we introduce a Sequential Variation Determination (SVD) technique that extends BVP to guarantee a best solution of the MC model for a given set of device variation specs. With SVD, we developed a fully automated method for creating MC MOSFET model libraries. Our results clearly show that this approach significantly shortens the development time of a MC library and improves the model accuracy at the same time.