Your circuit design (analog, maybe mixed-signal) has already the best circuit topology. There might be still some room for parameter selection, e.g. the geometries of transistors or values of passive elements, to best fit the specific purpose. This is, what (automatic) circuit optimization will deliver. In addition you may fine-tune, optimize and verify the circuit over voltage, process or temperature corners. So circuit optimization is a valuable tool in the hands of an experienced designer. It will relieve you from the routine task of 'endless' repetitions of re-simulating your design.

You have to choose circuit variables as parameters to be varied during optimization (e.g. device size, component values, bias inputs etc.). Then you may pose performance constraints onto you circuits (e.g. Vnode < 1.2V, gain > 50 etc.). Optimization objectives are the variables to be minimized or maximized. The n objectives and m constraints are assembled into a cost function.

The optimization flow is now the following: The circuit is loaded. Several (perhaps only one) simulations are started with a suitable starter set of variables. Measurements are done on the simulator output to check for the performance constraints and optimization objectives. These data are fed into the optimizer to evaluate the cost function. A sophisticated algorithm now determines a new set of circuit variables for the next simulator run(s). Stop conditions have to be defined by the user to tell the simulator when to finish (e.g. fall below a cost function value, parameter changes fall below a certain threshold, number of iterations exceeded).

The optimizer algorithms, its parameters and the starting point influence the convergence behavior. The algorithms have to provide measures to reaching the global optimum, not to stick to a local one, and thus are tantamount for the quality of the optimizer.

ngspice does not have an integral optimization processor. Thus this chapter will rely on work done by third parties to introduce ngspice optimization capability.ngspice provides the simulation engine, a script or program controls the simulator and provides the optimizer functionality.

Four optimizers are presented here, using ngspice scripting language, using tclspice, using a Python script, and using ASCO, a c-coded optimization program.