December author Günter Kompa let us know what prompted him to write his new book, Parameter Extraction and Complex Nonlinear Transistor Models:
Since the eighties, parameter extraction of microwave transistors was the permanent key topic in my research group at Kassel University. Initially we dealt with packaged devices comprising the wire-bonded die in a metal or ceramic package. At that time we already experienced the challenge of reliable parameter extraction of the used devices. This seems to be a problem for all applicants having no access to device processing facilities. When the number of model elements of an equivalent circuit exceeds a critical size than optimization methods are commonly applied. However, as is well known, the more model parameters the higher the risk getting trapped in a local minimum.
Actually, a similar problem is given with high power large-size gate periphery GaN HEMTs. These multifinger structures provide additional parasitic interelectrode capacitanes that increase the complexity of the device equivalent circuit. One should not forget single-finger devices at higher operation frequencies (say > 80 GHz) where stray capacitances are becoming noticeable.
We have solved the local minimum problem by developing an innovative model parameter extraction strategy incorporating prior physics- and experimental-based knowledge of the device. Notice this coincides perfectly with the statements given by Oliver Nelles in Chapter 7, Model Complexity Optimization (page 97), of his book Nonlinear System Identification: From Classical Approaches to Neural Networks and Fuzzy Models, Springer 2001. It is a pity not having mentioned this author in my book because I had not recognized him earlier.
The extraction process encompasses a low-frequency and a high-frequency phase. The low-frequency process is similar to the conventional one but differs in the result. The reliable optimization relies on the generation of high-quality starting parameter values obtained from classical physical models´ view (Shockley and advanced versions), low-frequency measurements and modified simplex algorithm. The high-frequency extraction process provides the distribution of the low-frequency model parameters based on a model parameter scanning method. The accuracy of the extracted model parameters of the complex equivalent circuit is high so that extraction refinement by final optimization is practically not needed taking into account the specified uncertainty of commercial vector network analyzers.
For more information or to order, click here.