由於電晶體微縮技術越來越艱難且缺乏經濟效益，三維積體電路 (3DIC) 被視為是一般用來提升電晶體密度的作法。但3DIC是一種昂貴的封裝技術，所以許多半導體製造商提供低成本的2.5D封裝解決方案。而整合型扇出型晶圓級晶片封裝 (InFO-WLCSP) 是其中最有潛力的解決方案。
使用無源裝置來解決電源完整性問題是一種常見的方法，而集成無源裝置 (IPD) 搭配InFO-WLCSP封裝是最有效解決方案。許多IPD被黏貼在InFO-WLCSP封裝上，由於替換IPD的成本遠高於製造IPD的成本，因此IPD的質量和可靠性必須盡可能高。
除了功能測試 (go-no-go test) 外，業內常見的做法是使用高壓應力測試 (HVST) 和可靠性篩選方法來篩選掉低可靠性的IPD。然而，可靠性篩選方法可能導致過度殺傷 (overkill)，並且過度殺傷的比例將隨著可靠性的增加而增加。為了提高篩選能力且在一個可接受的過度殺傷率 (overkill rate)，使用一個更準確的篩選方法會比僅緊縮現有方法的篩選條件更好。
在這篇論文中，為了提高IPD的質量與可靠性，我們提出了一種基於深層神經網絡 (DNN) 的篩選方法與新的測試流程。我們使用來自一片晶圓的34,204個已知擊穿電壓(BV)的IPD作為訓練集來訓練我們的DNN模型，並使用來自另一片晶圓的34,204個已知BV的IPD作為測試集來驗證DNN方法的篩選能力。結果表明，當overkill rate限制在3%時，我們這種方法可以比現有的方法多篩選出8％的壞晶片。這也表明我們的方法有可能在汽車、航空、工業、國防等安全攸關產品中提高IPD的可靠性和質量。

Since the miniaturization of transistors becomes more and more difficult and lacks economic benefits, the 3DIC has been considered as a general method for increasing transistor density. The 3DIC is an expensive packaging technology, so many semiconductor manufacturers offer low-cost 2.5D packaging solutions, and Integrated Fan-Out Wafer-Level Chip-Scale Packaging (InFO-WLCSP) is one of the most promising solution.
The use of passive devices to solve the problem of power integrity is a popular method, and InFO-WLCSP with Integrated Passive Devices (IPDs) is the most effective solution. Many IPDs are mounted onto the InFO-WLCSP package. Since the cost of replacing IPDs is much higher than the cost of manufacturing IPDs, the quality and reliability of IPD must be as high as possible.
In addition to go-no-go test, a common practice in the industry is to use High Voltage Stress Test (HVST) and reliability screening to screen the low-reliability IPDs. However, the reliability screening method may cause overkill, and the overkill rate will increase as the reliability requirement increases. To improve the screening ability under an acceptable overkill rate, using a more accurate screening method instead of just tightening the screening condition of existing methods is a better way.
In this work, in order to improve the quality and reliability of IPD, we propose a Deep Neural Network (DNN) based screening method with a new test flow. In the experiment, we use 34,204 known Breakdown Voltage (BV) IPDs from one wafer as the training set to train our DNN model, and use 34,204 known-BV IPDs from the other wafer as the test set to verify the screening capability of the DNN model. The results show that our method for this case can screen out 8% more bad dies than the existing industrial method when the overkill rate is limited to about 3%. It also shows that our method has the potential for increasing the reliability and quality of IPDs in safety-critical products, such as automotive, aviation, industrial, defense, etc.

[1] R. Courtland, "Transistors could stop shrinking in 2021," IEEE Spectrum, vol. 53, no. 9, pp. 9-11, Sept. 2016.
[2] H. Chen, H. C. Lin, C. N. Peng, and M. J. Wang, "Wafer Level Chip Scale Package copper pillar probing," in Proc. International Test Conference (ITC), Oct. 2014, pp. 1-6.
[3] M. Lee, H. C. Lin, C. N. Peng, and M. J. Wang, "Novel adaptive probing for wafer level chip scale package," in Proc. Joint e-Manufacturing and Design Collaboration Symposium (eMDC) & International Symposium on Semiconductor Manufacturing (ISSM), Sept. 2015, pp. 1-5.
[4] M. Swaminathan, D. Chung, S. Grivet-Talocia, K. Bharath, V. Laddha, and J. Xie, "Designing and Modeling for Power Integrity," IEEE Transactions on Electromagnetic Compatibility, vol. 52, no. 2, pp. 288-310, May 2010.
[5] H. Johari, and F. Ayazi, "High-Density Embedded Deep Trench Capacitors in Silicon With Enhanced Breakdown Voltage," IEEE Transactions on Components and Packaging Technologies, vol. 32, no. 4, pp. 808-815, Dec. 2009.
[6] Y. Freeman, R. Hahn, P. Lessner, and J. Prymak, "Reliability and critical applications of tantalum capacitors," in Proc. CARTS Europe 2007 Symposium, Oct. 2007, pp. 193-204.
[7] T. J. Chiu, Y. C. Lin, Y. C. Wang, C. C. Wu, H. C. Lin, and M. J. Wang, "IPD robustness test methodology for InFO," in Proc. e-Manufacturing and Design Collaboration Symposium (eMDC), Sept. 2016, pp. 1-4.
[8] R. Kawahara, O. Nakayama, and T. Kurasawa, "The effectiveness of IDDQ and high voltage stress for burn-in elimination," in Digest IEEE International Workshop on IDDQ Testing, Oct. 1996, pp. 9-13.
[9] R. Munikoti, and P. Dhar, "Highly accelerated life testing (HALT) for multilayer ceramic capacitor qualification," IEEE Transactions on Components, Hybrids, and Manufacturing Technology, vol. 11, no. 4, pp. 342-345, Dec. 1988.
[10] AEC-Q001 Rev-D, Guidelines for Part Average Testing, AEC, Dec. 2011.
[11] T. S. Barnett, M. Grady, K. Purdy, and A. D. Singh, "Exploiting defect clustering for yield and reliability prediction," IEE Proceedings - Computers and Digital Techniques, vol. 152, no. 3, pp. 407-413, May 2005.
[12] M. J. Moreno-Lizaranzu, and F. Cuesta, "Improving Electronic Sensor Reliability by Robust Outlier Screening," Sensors, vol. 13, issue 10, pp. 13521-13542, Oct. 2013.
[13] M. A. Nielsen, Neural Networks and Deep Learning, Determination Press, 2015.
[14] F. Davò, M. T. Vespucci, A. Gelmini, P. Grisi, and D. Ronzio, "Forecasting Italian electricity market prices using a Neural Network and a Support Vector Regression," in Proc. AEIT International Annual Conference (AEIT), Oct. 2016, pp. 1-6.
[15] H. Shi, M. Xu, and R. Li, "Deep Learning for Household Load Forecasting—A Novel Pooling Deep RNN," IEEE Transactions on Smart Grid, March 2017 (prepublication).
[16] N. Liu, X. Lin, and Y. Wang, "Data center power management for regulation service using neural network-based power prediction," in Proc. International Symposium on Quality Electronic Design (ISQED), March. 2017, pp. 367-372.
[17] T. Vieira da Silva, R. Vitor Arantes Monteiro, F. A. M. Moura, M. R. M. C. Albertini, M. A. Tamashiro, and G. Caixeta Guimaraes, "Performance Analysis of Neural Network Training Algorithms and Support Vector Machine for Power Generation Forecast of Photovoltaic Panel," IEEE Latin America Transactions, vol. 15, no. 6, pp. 1091-1100, June 2017.
[18] A. D. Singh, and C. M. Krishna, "On optimizing VLSI testing for product quality using die-yield prediction," IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 12, no. 5, pp. 695-709, May 1993.
[19] A. D. Singh, P. Nigh, and C. M. Krishna, "Screening for known good die (KGD) based on defect clustering: an experimental study," in Proc. International Test Conference (ITC), Nov. 1997, pp. 362-369.
[20] T. S. Barnett, A. D. Singh, and V. P. Nelson, "Burn-in failures and local region yield: an integrated field-reliability model," in Proc. VLSI Test Symposium (VTS), May 2001, pp. 326-332.
[21] T. S. Barnett, A.D. Singh, M. Grady, and K. Purdy, "Yield-reliability modeling: experimental verification and application to burn-in reduction," in Proc. VLSI Test Symposium (VTS), May 2002, pp. 75-80.
[22] T. S. Barnett, A. D. Singh, and V. P. Nelson, "Extending integrated-circuit yield-models to estimate early-life reliability," IEEE Transactions on Reliability, vol. 52, no. 3, pp. 296-300, Sept. 2003.