使用故障區域辨識技術之多重故障矽診斷

Abstract

隨著設計複雜度增加和製程更加先進，有些錯誤無法在矽前階段就被驗證的流程所偵測到，到了晶片上才會顯現出來，因為晶片實體上的限制，存取晶片內的資料非常困難，如果有錯誤發生，耗費在矽診斷的上時間會愈來愈多，尤其是要找出故障的位置，因此有許多研究都針對找出錯誤的位置發展。針對多重故障，我們提出一個診斷的架構，我們的演算法可以辨認一個包含所有故障的區域，設法刪除其中不可能為錯誤之候選者，剩下的候選者經過評分之後，真正的故障會被排到評分表的前面，我們修正第一個候選者再測試，透過迭代診斷、修正和再次測試，可大幅增進我們演算法的效能。由實驗結果顯示，使用此診斷架構確實能夠有效地在很少的迭代中就找完所有的故障點。

While designs are getting complex and technology becomes advanced, some errors may escape the verification flows pre-silicon, but exhibit on silicon. Because of physical limitation of chips, it is difficult to access data in chips. If errors occur, silicon diagnosis takes more and more time due to the limitation, especially for locating the fault sites. Thus many researches are developed for finding the fault locations. Targeting on multiple faults, we propose a diagnosis framework. In the framework, our algorithm can identify a region covering all faults then it removes impossible candidates. After the remaining candidates are ranked, real faults can be arranged to top of the list. We repair the first one and test the circuit again. With iterative diagnosis, repairing and testing, performance of our algorithm is enhanced. The experiment results show all fault sites can be identified in a few iterations with the proposed diagnosis framework.