Abstract:

It is difficult to obtain accurate evaluation results in dealing with small-sample electronic devices by using the traditional methods in practice, because the small samplecan not accord with the large sample-based hypothesis of the traditional methods. In order to solve this problem, this paper proposes a reliability evaluation method of small-sample electronic devices based on the support vector machine （SVM）. In this method, after the training of failure time of electronic devices, the optimal kernel function and parameters are selected to construct a SVM model, and the reliability parameters are evaluated according to the straight line fitted by the SVM model. Evaluated results of the life distribution of a gate oxide indicate that the proposed method is more accurate than the least square method based on large sample.

Key words: reliability, integrated circuit, support vector machine, gate oxide, least square method