在以往電子構裝相關研究之文獻中,分析一構裝件受力後之應力、應變或壽命經常為一定值,然而實驗或實測所得的結果卻往往具有相當的離散性,本研究為了探討這樣一個隨機疲勞壽命分配從何而來的問題,就以下兩種狀況探討之,一為構裝體之外型尺寸因加工誤差而具變異性,另一為modified Coffin-Manson equation1並非一完全確定之疲勞壽命預估模型。研究的方法為在適當隨機考量下,利用有限元素軟體模擬覆晶構裝體受溫度循環負載,負載後得其最大等效塑性應變再經由modified Coffin-Manson equation求其疲勞壽命。研究之成果顯示:錫鉛凸塊半徑變異較晶片厚度變異更易影響覆晶構裝體之疲勞壽命變異現象,而在modified Coffin-Manson equation並非為一完全確定之疲勞壽命預估模型假設下,構裝體因預估模型變異而產生之隨機疲勞壽命分配現象已不可忽略,而藉由此壽命分配,我們可進一步評估構裝體之量化可靠度及可靠度隨使用時間之退化情形
In study the reliability of electronic packages from mechanics point of view, the analytical result of stress and strain obtained from finite element analysis and fatigue life prediction based on a certain rule are all constant values. However, the real outcomes of package life obtained in laboratories appear to have probability distributions and are frequently plotted in Weibull probability papers. To investigate possible causes of this contradiction, analytical work is performed in the present study. The work includes, first, a finite element analysis based on the assumption that certain geometric parameters are random variables. The maximum strain of a certain type of flip-chip package subjected to thermal-cyclic loading is found, and the fatigue life of the package is determined base on a modified Coffin-Manson equation. Both quantities are random variables owing to the randomness of the geometric parameters. It is found that, among different geometric parameters, the size of the solder bump affects the fatigue life of the package the most. It may cause the fatigue life to have a coefficient of variation (c.o.v.) of 10.65% under the assumption that the solder diameter is a random variable between 0.27 mm and 0.33 mm. In the second phase of the present study, the modified Coffin-Manson equation is considered to have a certain random nature. This can be achieved by assuming certain parameters in the equation are random variables. Through mathematical derivation, it is shown that the predicted fatigue lives may have different mean values and different variations, and the difference may be tremendous under certain assumptions. It is concluded that both random geometric configuration and random life prediction rule may cause the fatigue life of the package to have distribution following certain probability density functions as those obtained from experiments.