Statistical modeling techniques: FPV vs. BPV

A comparison of forward and backward propagation of variance methods is presented, applied to generation of SPICE models that capture the statistical variation of microelectronic devices. MOSFET examples taken directly from a CMOS process are used to highlight the challenges in accurately capturing statistical process variation during process development. Ring oscillator simulations are used to quantify the performance of worst and best case models generated by each method.