This thesis describes the development and application of statistical circuit simulation methodologies to analyse digital circuits subject to intrinsic parameter fluctuations. The specific nature of intrinsic parameter fluctuations are discussed, and we explain the crucial importance to the semiconductor industry of developing design tools which accurately account for their effects. Current work in the area is reviewed, and three important factors are made clear: any statistical circuit simulation methodology must be based on physically correct, predictive models of device variability; the statistical compact models describing device operation must be characterised for accurate transient analysis of circuits; analysis must be carried out on realistic circuit components. Improving on previous efforts in the field, we posit a statistical circuit simulation methodology which accounts for all three of these factors. The established 3-D Glasgow atomistic simulator is employed to predict electrical characteristics for devices aimed at digital circuit applications, with gate lengths from 35 nm to 13 nm. Using these electrical characteristics, extraction of BSIM4 compact models is carried out and their accuracy in performing transient analysis using SPICE is validated against well characterised mixed-mode TCAD simulation results for 35 nm devices. Static d.c. simulations are performed to test the methodology, and a useful analytic model to predict hard logic fault limitations on CMOS supply voltage scaling is derived as part of this work. Using our toolset, the effect of statistical variability introduced by random discrete dopants on the dynamic behaviour of inverters is studied in detail. As devices scaled, dynamic noise margin variation of an inverter is increased and higher output load or input slew rate improves the noise margins and its variation. Intrinsic delay variation based on CV/I delay metric is also compared using ION and IEFF definitions where the best estimate is obtained when considering ION and input transition time variations. Critical delay distribution of a path is also investigated where it is shown non-Gaussian. Finally, the impact of the cell input slew rate definition on the accuracy of the inverter cell timing characterisation in NLDM format is investigated.
展开▼
