Research On Parameters Extraction And Circuit Optimization With Back Propagation Algorithms

Models of semiconductor elements are playing important roles in the industry of Integrated circuits;in the meanwhile,parameters extraction is a key procedure when modeling an electrical element.Due to the progress in craftsmanship,the scale of integrated circuits became smaller and smaller.With an increasing number of microscopic elements,the physical environments around these elements and side effects became unneglectable.To deal with such complicated cases,models are growing more complexe.This complexity is a great challenge for current parameters extraction algorithms,for the traditional algorithms can not work as well as they once did with those simpler cases.Thanks to the development of integrated circuits,the computing power has been improved drastically.With the advance of computing machines,a bunch of algorithms in the category of back propagation class achieved great success in quite a lot of optimization problems in other fields.In this work,we tried to apply these back propagating algorithms to the problem of parameters extraction and circuit optimization,and proposed a method dealing with such problems with efficiency.The key contributions of this work are three-fold:1.A method based on back propagation algorithms which runs faster on complexed models dealing with circuit optimization and parameters extraction problems.2.An open source scalable SPICE-like simulation toolkit with high usability is developed.The toolkit gives out gradients information simultaneously when figuring out the target functions.This toolkit is considered as a platform for doing further researches on circuit optimization and parameters extraction problems.3.A netlist definition script language standard is defined based on the open SCS language standard.Some new instructions which are fully compatible with the SCS language specification are added into the language.These new instructions by far simplified the procedure of exploring new algorithms on circuit optimization and parameters extraction problems.In comparison with those differential algorithms and other traditional algorithms without using gradients,our method and toolkit perform better and faster,especially on complexed models.