It is well known the impact of variability is growing dramatically with each new nanometer node. Critical applications such as automotive and health require failure modes in the unit per billions range. Statistical analyses have emerged and been largely accepted for advanced and accurate verification of circuit performance across process, voltage, and temperature variations. Brute Force Monte Carlo (MC) works but results converge too slowly for high-sigma analysis. Depending on the application. engineers wish to verify their design to 3 sigma, to 6 sigma or even more. 6-sigma MC analysis needs Billions of runs, and 7-sigma would need Trillions! A conflicting set of requirements for high-sigma analyses include managing non-Gaussian statistical distributions, a wide range of digital measurements, managing a huge number of circuit parameters, and a very limited budget of simulations while keeping precision at a high level. How can high-sigma analysis be deployed in the design flow?
