“State-of-the-Art”
A rather disturbing version of the “Moore’s law” ? the number of compact model parameters doubles every decade
BSIM3v3 ? no “challenge” because it is the de facto industry standard
- Local/group-device optimization ? unphysical parameter values
- Global optimization ? takes weeks to generate a model (or a day with the Genetic Algorithm [IBM])
Our challenge ? use experimental/TCAD data to aid model formulation in finding the “global functionality” to cover a wide variable range, leaving “insensitive” process-dependent parameters for empirical fitting at the “extreme” conditions
- TCAD synthesis + interpolation
- Empirical formulation + nonlinear regression
- Analytical derivation + experimental correlation
Guidance ? “Everything should be made as simple as possible, but not any simpler.” (Albert Einstein)