Knowledge Based Fuzzy Logic Approach To Semiconductor Device Modeling

Device models are the basis of the circuit simulation and design. Along with the progress ofthe integrated circuit technology, the appearance of new materials and the diminishing of thedevice’s feature sizes, the devices’ structure and characteristic become more and more complex.How to build brief models which can describe device characteristics accurately becomes a key issueof integrated circuit design. Therefore, device modeling is always a hotspot and focus in the field ofsemiconductor and integrated circuit study.This paper first discusses the condition and the importance of device modeling, then expoundson the common modeling methods of semiconductor devices, including the physical model, theequivalent circuit model and numerical model, and then explains the characteristics of variousmethods.Mainly discussed in this paper is an unconventional modeling method—Fuzzy Logic ApproachBased Knowledge. Fuzzy logic possesses the universal approximation characteristic as neuralnetwork, it can be used in all kinds of modeling problem. But differed from the neural network,fuzzy logic is composed of a group of fuzzy inference rules mainly, which describe cause and effectof questions.Before setting up a model by using fuzzy logic, we should firstly set the structure of the modelthat contains the partition of the Fuzzy set of input and output variables, the definition ofmembership functions, the confirmation of the number of fuzzy rules and the premise andconclusion of rules. The feature of the paper is that we make full use of the priori knowledge on thedevice features. Selection membership function, the premise and conclusion of rules based thedirect current characters, then we can confirm the structure of fuzzy logic model. Once built, theinitial model has better accuracy and save the time of study training. Using the prior knowledge ofdevice, we not only simplify the structure of fuzzy logic, but also make the model structure havemore physical significance,we prove the availability of the modeling method and simplify theprocess of modeling largely, reduce the complexity of model.The paper builds the models of different devices by using the method the paper suggested. Theexperimental data can be got by circuit simulating in the environment of the hspice and tcad, whichtest the Feasibility and validity of the model fully. It is easy to model the models of different devices preferably on condition that we only make a initial model. With universality and objectivity, weprove that the fuzzy logic modeling technology possess good performance in the means ofanalyzing experimental data.