Sequential linear programming (SLP) consists of linearizing the objective
and constraints in a region around a nominal operating point by a Taylor
series expansion. The resulting linear programming problem is then solved
by standard methods such as the Simplex [21] or the
Interior-Point [31] methods. After a preliminary testing of a
prototype implementation of the SLP method for TCAD models, it was found
necessary to limit parameter variations by enforcing bound constraints to
ensure the validity and accuracy of the linearization. At every iteration the
size of the linear trust region is adjusted automatically depending on the
agreement between the TCAD models and their linearized counterpart.