LSSVM在指数寿命型小子样IC寿命预测的应用

Application of LSSVM on Lifetime Prediction of IC with Small Sample from Exponential Distribution

针对传统寿命预测方法需要大量样本与现代高可靠集成电路(IC)在寿命试验中通常只有少量失效样本的矛盾,提出了基于最小二乘支持向量机(LSSVM)的指数寿命型小子样IC寿命预测方法。用蒙特卡罗方法研究了该方法在指数寿命型IC寿命预测应用中的可行性。同时与基于神经网络的预测方法相比。结果表明基于LSSVM的方法能更精确地预测小子样下IC的寿命,可为预测指数寿命型小子样IC的寿命提供一种新的有效途径。

It＇s becoming more and more difficult to get enough failure data sample during life test of modern integrated circuit（IC）.However traditional reliability assessment methods need a lot of failure data.In order to resolve this contradiction,a life prediction method of IC with small sample based on least squares support vector machine（LSSVM） is proposed.This method can be used to predict the lifetime of IC with small sample when the failure distributions are assumed to be exponential distribution.In addition,the effectiveness of LSSVM approach by Monte Carlo simulation is demonstrated.Error back propagation（BP） neural network is also compared with LSSVM method.The obtained results show that LSSVM method can be used to predict life of IC with small sample with high accuracy when dealing with failure data from exponential distribution.