小子样正态寿命型IC的可靠性评估

Reliability Evaluation of IC with Small Sample from Normal Distribution

面对集成电路(IC)在寿命试验中难以得到大量失效数据的小子样情形,提出了基于最小二乘支持向量机(LSSVM)的小子样正态寿命型IC可靠性评估方法。该方法的主要思想是基于寿命试验数据建立最小二乘支持向量机回归模型,根据该模型计算出正态分布的参数,从而进行可靠性评估。用蒙特卡罗方法研究了截尾失效情况下该方法在正态寿命型IC平均寿命评估应用中的可行性,同时与常用的最小二乘回归(LSR)法和极大似然估计(MLE)法相比,结果表明,基于LSSVM的方法能更精确地反映小子样下IC的可靠性,能为评估小子样正态寿命型IC的可靠性提供一种新的有效途径。

In view of the small failure data samples obtained from life test of electronic components, a reliability evaluation method based on least squares support vector machine （LSSVM） was proposed, which can be used to evaluate the reliability of electronic components with small sample when the failure distributions were assumed to be normal distribution. Based on the failure data, an LSSVM model can be constructed, and the normal distribution parameters can be evaluated by the LSSVM model, the reliability can be evaluated accordingly. In addition, Monte Carlo simulations were carried out to demonstrate the effectiveness of the LSSVM evaluation method in evaluating reliability when dealing with the small sample censored failure data from the normal distribution. Both of least squares regression （LSR） and maximum-likelihood estimation （MLE） were also compared with LSSVM evaluation method. The results show that LSSVM evaluation method has higher accuracy and can provide a new effective way for evaluating the reliability of small samples IC, when dealing with censored failure data from the normal distribution.