末端結構設計對電致遷移生命期測試之影響

Abstract

晶圓等級的電致遷移效應測試因為可以大幅縮短測試時間，在近幾年廣受重視。但是極重的加速測試條件，造成傳統的末端結構比待測導線更容易故障，適當的末端結構設計是晶圓等級的電致遷移效應測試可以實際應用的先決條件。 本論文以H. A. Schafft在1987年提出的四端點量測金屬導線電致遷移效應的主測試結構為基礎，再加上短路監控及HRR兩種監控電致遷移效應的結構，設計了一組完整的電致遷移效應測試結構。除了目前在參考文獻中所看過的標準NIST、三分支兩種末端結構之外，再設計等距箭尾、倍寬箭尾、10um分支間距、20um分支間距及30um分支間距等五種末端結構，以了解末端結構對電致遷移生命期測試的影響。 由數值模擬結果得知，10um分支間距、20um分支間距及30um分支間距等三種末端結構是因為分兩次將電流導入待測導線，因此所造成的電流密度梯度最小；其次是由待測導線兩倍線寬接45度緩衝區的標準NIST，由於有45度的緩衝區，所以電流密度的梯度也不大；再其次是三分支結構、等距箭尾與倍寬箭尾。因為這三種末端結構是將所有的電流在末端結構與待測導線的介面同時導入待測導線，箭尾結構中間的分支幾乎沒有作用。 在實際量測的結果方面，七種末端結構可以分為兩大族群：第一個族群是標準NIST、10um分支間距、20um分支間距及30um分支間距，第二個族群是三分支、等距箭尾、倍寬箭尾。第一個族群的生命期大約是第二個族群生命期的兩倍，除了10um分支間距及20um分支間距兩種末端結構之外，故障發生的位置幾乎都集中在陰極的末端結構或是陰極末端結構與待測導線的接合處。 將故障位置發生在待測導線與在末端結構上的樣品生命期分別統計，可以發現當故障發生在待測導線上時，不同末端結構對整個導線的生命期並沒有很明顯的影響當故障發生在末端結構上，不同的末端結構設計對測試結構的生命期影響甚鉅，標準NIST、10um分支間距、20um分支間距及30um分支間距等四種末端結構的生命期遠大於三分支、等距箭尾、倍寬箭尾等三種末端結構。這些結果與數值模擬的結果完全吻合。 由上述的實驗結果可以得知，末端結構的優劣可以從電流密度梯度判斷，不同的末端結構設計對電致遷移效應可靠度測試有決定性的影響。由於各種末端結構均無法避免故障的位置發生在末端結構上，所以在進行晶圓等級電致遷移效應量測時，必須逐一檢視樣本，剔除末端結構故障的數據，才能得到正確的結論。

Based on the electromigration test structure proposed by H. A. Schafft in 1987, a fully functioned test structure including extrusion monitor line and HRR structure is designed. Beside the standard NIST and 3-branch endsegments in the reference, we also design five endsegment including equidistant arrow, double line width arrow, distance=10um branch, distance=20um branch, and distance=30um branch, to observe the differences of electromigration lifetime test between 7 kinds of endsegments. MatLab and Origin softwares were also used to get the mathematics simulation of the current distribution on endsegments and metal lines. From the results of simulation, endsegments in distance=10um branch, distance=20um branch, and distance=30um branch cause the smallest current density gradient because the current is fed into the test line at two sections. As for the standard NIST structure, because of the 45˚ buffer area, it also has small current density gradient. And then, come the 3-branch, equidistant arrow, and double line width arrow. There is almost no function in the middle branch of the arrow endsegment, so the two arrow endsegments can be considered the same with 3-branch structure. In the results of the real tests, we gather statistics of the sample lifetime according to the place of the failures. The results show that when the failure occurs on the test line, different endsegments have no remarkable influences on the lifetime of the test line as expected. However, when the failure occurs on the endsegment, different endsegments have large impact on the life time of the test structures. The lifetime of the standard NIST, distance=10um branch, distance=20um branch, and distance=30um branch is much longer than that of the 3-branch, equidistant arrow, and double line width arrow. These test results completely match with the simulation results. According to the above experiment result, we realize that the quality of the endsegment can be judged from current density gradient. The different endsegments have decisive influence on electromigration reliability test. All endsegments can’t be excluded from occurring failures, so the samples in the wafer level reliability test for electromigration must be examined one by one carefully to pick off the failure data of endsegment. In this way, we can get the accurate conclusion.