晶片異常往往是晶片中掃描鏈上的瑕疵引起的，而在發現晶片有異常之後，晶片設計者會用一個電路內部分析的過程來掃描鏈上瑕疵位置的情況，針對那位置的情況來對未來的製程過程做改善。掃描鏈詳細位置可以用掃描鏈檢測的演算法來估算出來，只要電路內部分析儀器能準確地抓到錯誤位置，那麼就能大幅減少整個流程的時間和所花的金錢，所以找出掃描鏈瑕疵的演算法是越準確越好，也就是能越早抓到正確瑕疵的位置是很有幫助的。之前研究上有提過許多相關於掃描鏈檢測演算法，其中包括使用測試機台輔助的方法、更改晶片硬體的方法、以及用軟體輔助的方法。三類方法之中，前兩者需要借用到測試機台或是更改電路，只有第三者可以獨立靠軟體演算法完成檢測，所以第三者是業界比較常用的。至於軟體輔助的分類下可以再分為三小類，其中包括確定性演算法、訊號處理方法、以及機器學習等方法。之前一個用歸類於機器學習方法之下的非監督式貝葉斯演算法來做檢測的方法 [1] 能夠穩定的面對一些不規則性的錯誤，但是當兩個不同的錯誤有相近的行為的時候，會導致一些比較難被判斷的情況。這篇論文中，我們提出了一個使用多層監督式類神經網路來做掃描鏈檢測的方法。在使用ITC-99系列下的待測電路來做實驗得到的結果顯示我們的方法比較於非監督式貝葉斯的方法有77%更高的機會在第一次就鎖定正確掃描鏈上瑕疵的位置。

Scan chain failure is a common cause of a failing die. Physical failure analysis is performed to look into the location of the scan chain that causes failure for further analysis. The location of the scan chain that causes failure may be predicted by a scan chain diagnosis procedure. If physical failure analysis can be performed at the right spot, it helps for the designer to understand the cause of the failure thus modify the die to increase the yield in the future die manufacturing process. However, physical failure analysis costs a lot of effort and time. Therefore, the sooner the scan chain diagnosis procedure reports the correct location that causes failure the less efforts and time are needed. Many scan chain diagnosis techniques have been proposed using: tester-based method, hardware-based method, and software-based method. While both tester-based and hardware-based scan chain diagnosis have physical constraint, software-based scan chain diagnosis are free of any outer equipment or circuits modification. Software-based scan chain diagnosis can be further categorized into deterministic based, signal profiling based, and machine learning based method. A previous work using unsupervised Bayesian technique on scan chain diagnosis [1] has the ability to handle uncertainties more robustly and intelligently compare to other software techniques, but it has difficulty to distinguish between different failing scan cells with similar behaviors. In this thesis, we proposed a scan chain diagnosis method using multiple stages of artificial neural networks with incrementally shrinking refinement. Based on the results of experiments on ITC-99 benchmark circuit, our method has up to 77% higher ratio to locate the actual failure location of the scan chain being on the top of the list of scan chain diagnosis results compare to the unsupervised Bayesian method depending on different cases under both permanent and intermittent failure.

[1] Y. Huang, B. Benware, R. Klingenberg, H. Tang, Jayant Dsouza and W.-T. Cheng, “Scan Chain Diagnosis Based on Unsupervised Machine Learning”, in Proc. ATS, 2017
[2] R. Guo, and S. Venkataraman, “A Technique for Fault Diagnosis of Defects in Scan Chains”, Proc. Int’l Test Conf., pp. 268-277, 2001.
[3] J.-S. Yang and S.-Y. Huang, “Quick Scan Chain Diagnosis Using Signal Profiling”, Proc. of Int’l Conf. On Computer Design, Oct., pp. 157-160, 2005.
[4] P. Song et al., “A Novel Scan Chain Diagnostics Technique Based on Light Emission from Leakage Current”, Proc. Int’l Test Conf., pp. 140-147, 2004.
[5] J. Hwang et al., “Deterministic Localization and Analysis of Scan Hold-Time Faults”, Proc. Int’l Symp. for Testing and Failure Analysis, pp. 396-401, 2008.
[6] K. Khusyari et al., “Diagnosis of Voltage Dependent Scan Chain Failure Using VBUMP Scan Debug Method”, Proc. Asian Test Symp., pp 271, 2008.
[7] J. Schafer et al., “Partner SRLs for Improved Shift Register Diagnostics”, Proc. VLSI Test Symp., pp. 198-201, 1992.
[8] S. Edirisooriya and G. Edirisooriya, “Diagnosis of Scan Path Failures", Proc. VLSI Test Symp., pp. 250-255, 1995.
[9] Y. Wu, “Diagnosis of Scan Chain Failures”, Proc. Int'l Symp. on Defect and Fault Tolerance in VLSI Systems, pp. 217-222, 1998.
[10] Y. Huang et al., “Using Fault Model Relaxation to Diagnose Real Scan Chain Defects”, Proc. Asia and South Pacific Design Automation Conf., pp. 1176-1179, 2005.
[11] Y. Huang, “Dynamic Learning Based Scan Chain Diagnosis”, Proc. Design, Automation & Test in Europe Conference & Exhibition, pp.510-515, 2007.
[12] J.C.-M. Li, “Diagnosis of Single Stuck-At Faults and Multiple Timing Faults in Scan Chains”, IEEE Trans. Very Large Scale Integration (VLSI) Systems, vol.13, no. 6, pp. 708-718, 2005.
[13] Y. Huang, et al., “Scan Chain Diagnosis by Adaptive Signal Profiling with Manufacturing ATPG Patterns”, Proc. Asian Test Symp., pp.35-40, 2009.
[14] J. Li, Y. Huang, W.-T. Cheng, C. Schuermyer, and D. Xiang, “A Supervised ANN Method for Memory Failure Signature Classification”, in Solid-State and Integrated Circuit Technology, pp. 1-3, 2012.
[15] Y. Huang, R Guo, W.T. Cheng, and J. C.-M. Li, “Survey of Scan Chain Diagnosis”, IEEE Design & Test of Computers, Vol. 25, NO. 3, pp.240-248, 2008.
[16] J.C.-M. Li, “Diagnosis of Multiple Hold-time and Setup-time Faults in Scan Chains”, IEEE Trans. on Computers, 54, 1467-1472, 2005.
[17] S.-Y. Huang, Chapter 7 - Logic Diagnosis of "VLSI Test Principles and Architectures - Design for Testability", Edited by L.-T. Wang, C.-W. Wu, and X. Wen, Morgan Kaufmann Publishers, pp. 397-460, 2006.
[18] C. Guojing, B. Onkar and F. Minwei, “An Efficient, Distributed Stochastic Gradient Descent Algorithm for Deep-Learning Applications”, International Conference on Parallel Processing (ICPP), pp. 11-20, 2017.
[19] D. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization”, International Conference for Learning Representations, pp. 1-15, 2014.
[20] J.-J. Hsu, C.-W. Tzeng, and S.-Y. Huang, “A New Robust Paradigm for Diagnosing Hold-Time Faults in Scan Chains,” Proc. IEEE Int’l Symp. on VLSI Design, Automation and Test, pp 171-174, 2006.
[21] TensorFlowTM. An open-source software library for Machine Intelligence. Retrieved from https://www.tensorflow.org/