provigil japan, provigil para que es, what to tell doctor to get provigil, can you take provigil and nuvigil together, canadian generic provigil, provigil tablet image

Getting A Clearer Picture

Statistical enhancement improves scan test diagnosis results and can turn failing test cycles into valuable data.


Scan test diagnosis is an established software-based methodology for localizing defects causing failures in digital semiconductor devices. Using structural test patterns (such as ATPG) and the design description, diagnosis turns failing test cycles into valuable data. Exactly how valuable this data is depends on the quality of the diagnosis results. A result that points to a small group of nets as potential defect locations may be a good starting point for physical failure analysis (PFA) of a customer return. However, it may not be sufficient to determine what exactly caused the failure to happen prior to PFA.

In this article we explore how a statistical enhancement process called Root Cause Deconvolution (RCD) can be used to enhance scan test diagnosis results and better identify the underlying root cause.

The quality of diagnosis results is measured by accuracy and resolution. Accuracy is a measurement of whether the diagnosis result contains the actual defect or not. Resolution measures the area of the location or locations captured by a suspect. Traditional logic-only diagnosis would identify the logic net(s) suspected of causing the test failures. The advent of layout-aware diagnosis improved the resolution to a net segment and provided a significantly better capability to differentiate between different defect types such as bridges and opens. Still, there is ambiguity in the diagnosis results. Typically, about 80% of diagnosis results contain multiple suspect locations or explanations of the failure. Each explanation can in turn be explained by multiple root causes. For instance, according diagnosis, failures observed for one device could be caused by a bridge between two specific nets. These two nets may be within line of sight in five metal layers. That means that there are five potential root causes (a bridge in in route_1, a bridge in route_2, etc.) that could explain this diagnosis result.

RCD is the next step in diagnosis resolution enhancement. It is a statistical enhancement process that works by analyzing multiple layout-aware diagnosis reports together to identify the underlying defect distribution (root cause distribution) that is most likely to explain this set of diagnosis results. RCD does not rely on foundry/process data or any other data beyond what is needed for layout-aware diagnosis. RCD processes the design layout and test patterns to extract statistics such as ‘critical area per tested net segment per layer’ and ‘tested via count per net segment per layer’. In addition to identifying the underlying defect distribution, the results are also back-annotated to the individual diagnosis suspects. Where layout-aware diagnosis alone points to a segment, RCD can isolate a particular root cause in that segment, as illustrated in Figure 1. There are two fundamental assumptions for this technology. First, RCD assumes that root cause instances are randomly distributed. A prerequisite for RCD analysis is therefore to filter out diagnosis results with systematic locations. The other assumption RCD makes is that there is one underlying root cause distribution. It is therefore recommended to analyze devices from different process conditions separately.

The practical aspects of RCD have been explored in several published articles. In [1], RCD was applied to the early stages of a 28nm yield ramp to estimate the root cause distribution for several lots. RCD found that the yield loss was dominated by metal4 and metal6 shorts, which was confirmed by PFA. After addressing these defects in manufacturing, RCD was used again to confirm the effect of the process change. In [2], RCD was used to examine and find defectivity trends and differences across designs and IP. In one example, the root cause of static leakage on a 20 nm test chip was found by comparing RCD on two populations with low and high static leakage. In a second example, two versions of the same design were produced on the same wafers with different yields. RCD revealed that the low-yielding design had one additional fail mechanism not found in the high-yielding design.

With the ability to identify root cause of yield loss from fail data alone, RCD increases the PFA relevance and success rate dramatically and reduces the PFA cycle time from months to days. RCD also enables “virtual PFA”, the ability to determine defect distribution for a population of failing devices before any failure analysis is performed. This is something that has been virtually impossible for fabless semiconductor companies to do in the past. RCD results can easily be added to existing yield monitoring flows and systems.

1. B. Benware,, “Determining a Failure Root Cause Distribution From a Population of Layout-Aware Scan Diagnosis Results”, IEEE Design & Test of Computers, 01/2012.
2. Y. Pan,, “Leveraging Root Cause Deconvolution Analysis for Logic Yield Ramping”, ISTFA, 2013.

Figure 1:Layout-aware diagnosis improves scan diagnosis resolution to a net segment. Root Cause Deconvolution (RCD) leverages statistical enhancement to identify the underlying root cause, in this example an open VIA defect.

Mentor SemiEng Fig 1

Geir Eide earned an MS in electrical and computer engineering from the University of California at Santa Barbara and is a product marketing manager in the Silicon Test Solutions group at Mentor Graphics Corporation, 8005 SW Boeckman Rd., Wilsonville, OR 97070 USA; ph.: 503-685-7943; e-mail: [email protected]


[…] Getting a Clearer Picture (Semiconductor Engineering) […]

Leave a Reply