Silicon Calibrated Scan Based Timing Tests for Delay Defect Detection

项目ID：CCF -0811454标题：用于延迟缺陷检测的eSilicon校准扫描定时测试PI姓名：Singh，Adit D. 机构名称：奥本大学，阿拉巴马州摘要微观制造缺陷，如局部缩小的互连线，通过空隙，和针孔绝缘栅氧化物，是一个主要的可靠性问题，在纳米数字集成电路技术。这种缺陷在生产测试期间难以检测，因为它们通常不会导致电路中的灾难性故障;相反，电路切换延迟可能沿包含缺陷的信号路径沿着略微增加。由于复杂的芯片包含数十亿个不同长度的电路路径，并且电路时钟定时被设计为适应最长的路径延迟，因此在制造后测试期间，短路径上的小延迟缺陷可能仍然隐藏在电路定时松弛内。然而，这种缺陷通常会在更坏的操作条件下导致错误，或者在现场操作的压力下进一步退化，从而导致早期寿命可靠性失效。这些小延迟缺陷可以通过更积极的时序测试来检测，使用比额定时钟频率更快的时钟频率，以捕获和暴露由于短路径沿着的过度延迟而导致的任何错误响应。这需要使用扫描设计测试方法，该方法提供对顺序设计中的内部触发器的完全访问，从而可以允许观察单周期操作的电路时序。扫描延迟测试检查是否信号传播延迟沿着目标路径在电路中，由两个矢量延迟测试模式激活，落在应用？发射？到哪里去？捕获？时钟周期然而，由于单周期扫描测试以不同于正常多周期顺序操作的方式操作电路，因此越来越多的证据表明，观察到的电路时序可能无法反映正常连续功能操作中的真实电路延迟。因素，如电源噪声，耦合和串扰从异常开关活动，在模具热分布的变化，和？时钟拉伸？都可以显著影响扫描测试定时。正常的制造工艺变化进一步增加了这些测试时序的不确定性，使得小延迟缺陷的检测极具挑战性。本研究认为，它可能是不切实际的，以观察真正的电路功能时序与足够的精度使用替代扫描的定时测试，以可靠地检测小的延迟缺陷。相反，通过比较来自匹配的部件群的定时测试结果来识别异常切换延迟（定时异常），可以更好地检测这种缺陷。这里，扫描测试不用于测试电路定时相对于固定额定时钟（例如，如速度分箱所需），而仅用于识别相对于总体的统计标准显示异常延迟的异常部分。重要的是，这种相对而不是绝对的时序评估消除了扫描时序测试的需要，以准确地匹配功能时序。通过比较测试，排除了电源和耦合噪声、时钟拉伸等共模因素的影响。非功能性测试输入也可用于测试，通过多个快速时钟，实现小延迟缺陷的高覆盖率，即使在短路径上。初步的研究实验表明，如果测试响应从相邻的模具进行比较，尺寸小于5%的关键路径的延迟缺陷可以可靠地检测。这项研究正在开发和评估一些新的方法，使这种硅？校准了吗扫描延迟测试实用且具有商业吸引力。新的延迟测试方法将使用专门设计的测试芯片进行评估。拟议研究的长期目标是与工业界合作开展实验研究，以验证批量生产零件的方法。该研究项目将有助于提高EPSCoR州亚拉巴马的研究能力。研究结果也将通过IEEE测试技术教育计划纳入PI在奥本和校外教授的高级课程。PI与位于奥本附近的历史悠久的黑人塔斯基吉大学密切合作，包括在VLSI测试领域正在进行的联合NSF计算研究基础设施（CRI）拨款合作。塔斯基吉大学的学生在奥本大学攻读研究生课程，有些学生由奥本大学的教师共同指导。塔斯基吉博士学生，谁采取了研究生课程与PI，将被鼓励参与这项研究。这将进一步加强奥本和塔斯基吉之间的研究互动，这有助于吸引美国少数民族公民攻读计算机领域的研究生。

Project ID: CCF - 0811454Title: eSilicon Calibrated Scan Based Timing Tests for Delay Defect DetectionPI name: Singh, Adit D. Institution: Auburn University, AlabamaAbstractMicroscopic manufacturing flaws, such as the localized narrowing of interconnection lines, via voids, and pin holes in insulating gate oxide, are a major reliability concern in nanometer digital integrated circuit technologies. Such defects are difficult to detect during production testing because they often do not cause catastrophic malfunction in the circuit; instead, circuit switching delays may be marginally increased along the signal paths containing the defect. Since a complex chip contains billions of circuit paths of varying lengths, and circuit clock timing is designed to accommodate the longest path delay, small delay defects on short paths can remain hidden within circuit timing slacks during post manufacturing testing. However, such defects can often cause errors under worse case operating conditions or degrade further under the stress of field operation to cause early life reliability failure. These small delay defects can potentially be detected through more aggressive timing tests, using faster than rated clock frequencies, to capture and expose any erroneous response due to the excessive delays along short paths. This requires the use of the scan design-for-test methodology which provides full access to the internal flip-flops in sequential designs, and can thereby allow circuit timing to be observed for single cycle operation. Scan delay tests check if signal propagation delays along the targeted paths in the circuit, activated by two vector delay test patterns, fall within the applied ?launch? to ?capture? clock period. However, because single cycle scan tests operate the circuit in a manner which is different from normal multi-cycle sequential operation, there is increasing evidence that the observed circuit timing may not reflect true circuit delays in normal continuous functional operation. Factors such as power supply noise, coupling and cross talk from abnormal switching activity, variations in die thermal profile, and ?clock stretching? can all significantly impact scan test timing. Normal manufacturing process variations further add to these test timing uncertainties, making the detection of small delay defects extremely challenging. This research takes the view that it may be impractical to observe true circuit functional timing with sufficient accuracy using a surrogate scan based timing test to reliably detect small delay defects. Instead, such defects are better detected by identifying abnormal switching delays (timing anomalies) through a comparison of timing test results from a matched population of parts. Here the scan tests are not used to test circuit timing against a fixed rated clock (e.g. as required for speed binning), but only to identify anomalous parts that display abnormal delays relative to the statistical norm for the population. Critically, this relative rather than absolute timing evaluation eliminates the need for the scan timing tests to accurately match functional timing. The impact of common mode factors such as power supply and coupling noise, clock stretching, etc. is factored out by the comparison test. Non-functional test inputs can also be used by the test, with multiple fast clocks, to achieve high coverage of small delay defects, even on short paths. Preliminary research has experimentally shown that if test responses from adjacent die are compared, delay defects of size less than 5% of the critical path can be reliably detected. This research is developing and evaluating a number of new methodologies to make such silicon ?calibrated? scan delay testing practical and commercially attractive. The new delay test methodologies will be evaluated using specially designed test chips. A longer term goal of the proposed research is to carry out experimental studies, in partnership with industry, to validate the methodologies developed on volume production parts. This research project will help improve the research capabilities of Alabama, an EPSCoR state. Results from the research will also be incorporated in advanced courses taught by the PI, both at Auburn and off campus, through the IEEE Test Technology Education Program. The PI works closely with the Historically Black Tuskegee University located near Auburn, including cooperating on an ongoing joint NSF Computing Research Infrastructure (CRI) grant in the VLSI testing area. Tuskegee students take graduate courses at Auburn University, and some are jointly advised by Auburn faculty. Tuskegee Ph.D. students, who have taken graduate courses with the PI, will be encouraged to participate in this research. This will further strengthen the research interaction between Auburn and Tuskegee, which is helping to attract minority U.S. citizens to graduate studies the computing field.