An efficient Jacobian updating technique for analog circuit optimization

A new methodology is proposed to update the Jacobian matrix of the circuit performance functions with respect to the noise parameters to minimize the number of circuit simulations at each iteration during an optimization process. This matrix is required at each iteration to estimate the variance of the performances, as the optimizer moves the nominal point in the space of designable parameters. The current method of evaluating the Jacobian at each iteration using the finite difference method is computationally too expensive. The new methodology uses the results from the previous circuit simulations to update the Jacobian matrix, and greatly reduce the number of circuit evaluations required during the optimization process. An example of the optimization of an operational amplifier is given, and the savings in computation time is described