The Research Of Intelligent Algorithms Based Parameter Extraction Of Semiconductor Devices

Extraction and optimization of semiconductor device parameters is an important issue in device modeling and simulation. In the integrated circuit (IC), the semiconductor device is the elementary part. IC design and analysis depend on device model, especially the accuracy and simplicity of the model parameters. Regardless how sophisticated a semiconductor device model is, the model is useless and inaccurate if the viable parameter extraction method is not in place. At the same time, reliability in devices has to face the reduction of size of components and the increase of the number of parameters in semiconductor devices, in other words, the dimensions of space in which the device model function exists become larger and larger.Mathematical modeling is a crucial element in the development of semiconductor technology. Semiconductor devices modeling has received much attention over the last decades in an attempt to better understand the phenomena occurring within the device. Normally, semiconductor device model is a complex nonlinear, multi-variable system that is hard to construct by conventional methods. In order to improve the accuracy of the models and make the models reflect the actual semiconductor performance better, it is necessary to identify the parameters of the models using optimization techniques.However, the accuracy and reliability of some commonly used parameter extraction techniques, e.g. least squares algorithm and Nelder-Mead simplex search method, are restricted by the measured data, whose errors are introduced by the numerical differentiation and simplified formulae are used in parameter determination as well. On the other hand, most of these methods are step-by-step procedures that essentially rely on extracting each parameter from restricted regions of the measured current-voltage (I-V)characteristics where the effect of other parameters is assumed to be negligible. Theintelligent algorithms are therefore proposed to overcome the disadvantages of thetraditional parameter extraction methods in this thesis.Recently, the methods based on intelligent evolution algorithm have attracted risingattention in the area of the semiconductor device parameter extraction. For example,genetic algorithm (GA), particle swarm optimization (PSO), differential evolutionary (DE),ant colony algorithm (ACA), artificial neural networks (ANN) and artificial immunealgorithm (AIA) are the popular technologies in dealing with global optimizationproblems.The advantages of the intelligent algorithms:(1) Particularly necessitate initial guesses as close as possible to solutions are not essential, required only is a broad range specified for each of the parameters.(2) Handling both discrete and continuous variables, nonlinear objective and constrain functions without requiring gradient information. It is known that many semiconductor models are complicated nonlinear functions, so the intelligent algorithms can do good jobs in this area.(3) A deeper understanding of the semiconductor model is needless. Meanwhile, the simplification of the model function and some complicated calculation like derivation in the process of parameter extraction are also unnecessary.The main points we will concern in this thesis including:(1) Compile algorithm programs and apply them to extract the parameters of semiconductor devices.(2) Improve the global searching ability of the standard PSO in order to prevent this method falling into local optimum at the end of the search.(3) Make detailed comparisons of the performance for every intelligent algorithm in the same semiconductor device parameter extraction problem so as to find the most suitable one. The comparative results indicate that the performance of parameters extracted from both simulation and experimental tests by intelligent algorithms are confirmed to be valid with respect to good agreements that can be found between the original data and fitted ones even in the presence of measuring noise.