Abstract: With the dimension of the semiconductor device scaling down, the model of the devices becoming more and more complex, the number of the model parameters also increasing, traditional gradient-based parameter extraction methods have more and more difficulty to deal with the models which have hundreds of parameters. Genetic algorithm based parameter extraction methods attracte tremendous attention since they can overcome the difficulty of traditional methods. Here the theory of genetic algorithm based parameter extraction method is illustrated, and the defects and some improvements are also put forward.

Key words: semiconductor, device model, parameter extraction, genetic algorithm, hill-climbing algorithm

摘要： 随着半导体器件特征尺寸的缩小，半导体器件模型也变得越来越复杂，模型参数个数急骤增加，目标函数自变量空间的维数也变得越来越大，传统的一些基于梯度的参数提取方法已经不能很好地解决问题。遗传算法是一种应用基因工程和人工智能模拟的优化算法，近年来在半导体器件模型参数提取领域被广泛使用，这种方法能有效地克服传统参数提取方法中的一些困难。详细阐述了采用遗传算法提取半导体器件模型参数的原理，同时也指出了采用这种方法提取模型参数时的缺点和目前的一些解决方法。

关键词: 器件模型, 参数提取, 遗传算法, 爬山法