Yet another approach to statistical circuit design-stochastic minimax

A formulation of the statistical design centering (SDC) problem of electronic circuits is presented. The income index optimization problem formulation, introduced by L.J. Opalski and M.A. Styblinski (1986), is slightly generalized and shown to result in a stochastic minimax formulation (SMM) of the SDC problem under some assumptions. A formula for the gradient of the objective function which uses circuit-sensitivity information is developed. The formula makes it possible to calculate the stochastic gradient under much weaker assumptions about the probability density function of the circuit parameter values than before. Two solution techniques for the SMM problem are briefly presented and compared. They are direct and indirect. Using circuit-sensitivity information is shown to lead to a development of new circuit-optimization algorithms. Some implementation details of the algorithms and results of preliminary tests are also presented.<
>