Accurate statistical circuit simulation in the presence of statistical variability

[摘要] Semiconductor device performance variation due to the granular nature of charge and matter has become a key problem in the semiconductor industry. The main sources of this ‘statistical’ variability include random discrete dopants (RDD), line edge roughness (LER) and metal gate granularity (MGG). These variability sources have been studied extensively, however a methodology has not been developed to accurately represent this variability at a circuit and system level. In order to accurately represent statistical variability in real devicesthe GSS simulation toolchain was utilised to simulate 10,000 20/22nm n- andp-channel transistors including RDD, LER and MGG variability sources. Astatistical compact modelling methodology was developed which accuratelycaptured the behaviour of the simulated transistors, and produced compactmodel parameter distributions suitable for advanced compact model generationstrategies like PCA and NPM. The resultant compact model librarieswere then utilised to evaluate the impact of statistical variability on SRAMdesign, and to quantitatively evaluate the difference between accurate compactmodel generation using NPM with the Gaussian VT methodology. Over 5 milliondynamic write simulations were performed, and showed that at advancedtechnology nodes, statistical variability cannot be accurately represented usingGaussian VT . The results also show that accurate modelling techniques canhelp reduced design margins by elimiating some of the pessimism of standardvariability modelling approaches.

[发布机构] University:University of Glasgow;Department:School of Engineering

[关键词] Statistical Variability, MOSFET, compact models, SRAM, simulations