Static and dynamic numerical device modeling for transient circuit simulation (Q1193206)

Aiming to develop numerical semiconductor models with short computing time but preserving essential features like nonnegativity and monotonicity of data, the author considers multivariate Bernstein polynomials. Generalizing classical theorems to the multivariate case, he proves convergence to the given function and its partial derivatives, and nonnegativity preservation. Piecewise quadratic Bernstein polynomials are then used to approximate voltage-charge or voltage-current relationships. During Newton iterations, in solving the nonlinear ordinary differential equations, the preservation of convexity is advantageous assuring greater robustness and convergence speed. The numerical model developed in this way is proved to be somewhat faster than the SPICE program (which implements an analytical model) -- as shown by several computations.