Multi-bias decomposition-based optimisation for the extraction of small-signal GaAs FET models

Van Niekerk, Cornell (1999-07)

Dissertation PhD(Ing) -- University of Stellenbosch, 1999.

Thesis

ENGLISH ABSTRACT: The availability of accurate nonlinear models is essential for the accurate and cost effective design of high frequency telecommunication systems. A key step in the creation of such models is the extraction of small signal models from measured s-parameter data. This thesis describes the development and evaluation of a new extraction algorithm. The method makes use of a decomposition-based optimiser which divides the problem into subproblems that are solved separately. The procedure is accurate, suited for handling large problems, and starting value independent. The success of the algorithm is con finned using a variety of tests. Both simulated and measured data for a GaAs MESFET and pHEMT are used. The test results also show the limitations of single bias extractions. In single bias extractions, the data measured at different bias points are handled as separate extractions. It is shown that the decomposition-based optimiser can be used to create a new multi-bias extraction algorithm. The multi-bias procedure combines the s-parameters measured at different bias points into one extraction problem, and defines the extrinsic model elements to be bias independent. This enforces a degree of physical reality onto the extraction. Tests prove the algorithm to be robust and accurate. It is able to find previously difficult to detennine elements with a high degree of precision. Results are presented that indicate how the extraction uncertainty decreases as the number of bias points used in the multi-bias extraction is increased. The new algorithm produces small signal models that accurately represents the measured s-parameters at all bias points.

AFRIKAANSE OPSOMMING: Die beskikbaarheid van akkurate nie-linere rekenaar modelle is onmisbaar vir die akkurate en koste effektiewe ontwerp van hoe frekwensie kommunikasie stelsels. Die onttrekking van kleinsein modelle uit gemete s-parameter data is 'n kern stap in die skepping van nie-linere modelle. Die tesis beskryf die ontwikkeling van 'n nuwe optimerings algoritme vir die onttrekking van kleinsein modelle uit gemete s-parameters. Die optimerings algoritme is gebasseer op 'n dekomposisie benadering wat die probleem in kleiner subprobleme verdeel en oplos. Die metode is akkuraat, geskik vir hoe dimensionele probleme en kan onakkurate optimerings beginwaardes hanteer. Die sukses van die algoritme is bevestig deur 'n verskeidenheid van toetse op gesimuleerde data, sowel as gemete data van 'n GaAs MESFET and pHEMT. Die resultate dui ook die beperkings van die enkel voorspanning ekstraksies aan. Enkel voorspanning ekstraksies hanteer die data gemeet by verskillende voorspannings as aparte ekstraksie probleme. Daar word getoon dat die dekomposisie optimeerder uitgebrei kan word om multi -voorspannings probleme te hanteer. In die multi-voorspannings geval word die data gemeet by verskillende voorspannings in een onttrekkings probleem saam gevoeg en die ekstrinsieke elemente van die model word as voorspannings onafhanklik gedefineer. Dit forseer 'n sekere mate van fisiese realiteit op die probleem af. Die multi-voorspannings benadering is robuust en akkuraat en vind ook die waardes van voorheen moeilik bepaalbare model elemente met 'n groot mate van sekerheid. Daar word getoon hoe die ekstraksie onsekerheid afneem soos die hoeveelheid voorspannings in die probleem vermeerder word. Die nuwe algoritme produseer kleinsein modelle wat die s-parameters van die transistor akkuraat modelleer by al die voorspannings.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/51344
This item appears in the following collections: