Is It Safe To Assume That All “Passed” Die Are Actually “Good” Die?

Detecting tricky test escapes and preventing defective parts from getting into your customer’s supply chain.


In a world where Quality and Brand Protection is King, as certainly is the case for the automotive and medical device industries where strict minimal DPPM (defective parts per million) requirements are a common constraint, new methods for “escape” prevention and outlier detection are constantly being evaluated and implemented by semiconductor vendors to prevent any defective or marginal parts from leaving their manufacturing facility and later being flagged as an RMA (return material authorization) and possibly damaging the brand reputation of their customer.

With best practices being deployed to detect and discard all defective devices, it is assumed that all parts are actually tested on an ATE system before being deemed good or bad. But what if that assumption is incorrect and parts you thought were properly tested and labeled as good were never actually tested, or tested incompletely due to some software or hardware anomaly within the ATE system? Is that even possible with today’s sophisticated systems? The answer is, unfortunately yes. All hardware systems, including ATE, eventually experience issues that require periodic recalibration and preventative maintenance. If these issues are not addressed in a timely manner, they could manifest themselves in a way that may prevent certain devices from being properly tested. If this happens, there is a valid concern that these improperly tested devices, if mistakenly labeled as good, could end up in the supply chain as a “test escape.”

One such example of a test escape is a phenomenon called a Tester Freeze where the same parametric test performed on consecutive parts yields the exact same value which mathematically is very unlikely to occur (the probability that five successive parts yield the exact same measured value for a given test is 0.0218%). What happens in this case is that the value recorded by the tester is “frozen” over a small, finite number of subsequent parts. The test result that is measured and recorded before the freeze occurs is what is also recorded for subsequent parts during the freeze. More often than not, the problem may be as simple as the tester temporarily running out of resources (memory).

But how do you catch this subtle and proverbial needle in a haystack issue? Do you have to manually scour over millions of lines of tester log data to find it, if it even exists? Can the tester alert me upon encountering this phenomena? Although some testers may be able to alert you on having repeated test results on successive chips, there is typically no automated way of marking the chip/wafer/lot for retest or allowing a particular chip in question to be automatically re-binned from a good bin to a bad bin to prevent these chips from escaping. The good news is that with today’s big data analytics engines and automated rule monitoring software, catching such anomalies can be done in near real-time.

Figure 1 below shows a real-world example of the output of an analytics rule being triggered during the testing of a product with the results displayed and sent in an email to the responsible engineer in near real-time. Note that the precise dice, including their coordinates and touchdown sequence along with the precise measured test value per die, were captured. Since the measured value did not vary over a series of five (5) dice then a tester freeze has likely been identified.

Figure 1. Rule being triggered

Looking at Figure 2 below, it can be difficult, if not impossible to visually identify a tester freeze in a large sample of data. Only a big data analytics solution with an automated rules engine can easily identify these types of test anomalies and act upon them as they occur.

Figure 2. Test results per touchdown sequence

Lastly, Figure 3 shows how the test data can be cross-correlated to the die locations on the wafer map to further aid the operations team in diagnosing the issue with the ATE system. Note that the dice in Hard Bin #8 are labeled as “good.” Once we have confirmed that a tester freeze has occurred, the user can opt to automatically re-bin these dice as “bad” to prevent them from being shipped into the customer supply chain.

Figure 3. Detailed parametric test data results with cross correlation to wafer map

If quality is a significant concern for you or your end customer, then ensuring your ATE system is functioning properly is a must. Checking for anomalies such as the Tester Freeze condition should be one of several key steps you should take in securing a virtual “quality firewall” for preventing any costly and damaging escapes from occurring.

To learn more about how semiconductor companies are leveraging these types of automated technologies to aid in achieving high quality standards in their devices, visit us at