Electrical Test Confidence


Every new test engineer eventually finds out that absolute certainty of test results can never be guaranteed, even if the most advanced automated test equipment (ATE) are involved. As such, there will always be instances wherein retest of units is necessitated, consuming valuable test capacity and increasing test cycle time without additional output.


Electrical test confidence, simply put, is a measure of how consistently an ATE system delivers the correct electrical test results. A test system with excellent test confidence makes retest unnecessary, saving precious test resources. Test confidence is therefore an indirect indicator of how efficient and productive a test process is.


If someone will collect all the test rejects after the first-pass screening of a lot and retest these, there is reasonable likelihood that some of these initially failing units will pass their second test.  Since genuine failures can not be resurrected with retesting, the only explanation for their passing the second test is that they were never really electrically bad when they were first tested.  Something, somehow, made them fail the first test.


An electrically good unit that fails electrical testing for whatever reason (and can therefore pass electrical retesting under more proper conditions) is often referred to as an 'invalid' reject or failure.  One's test confidence in an ATE system may be measured in terms of the number of invalid failures encountered from the system.


Invalid failures occur due to a multitude of reasons, which include but are not limited to:  1) improper contact between DUT leads and the contactor; 2) poor design and fabrication of test hardware; 3) improper test equipment set-up; 4) oxidation and contamination of metal surfaces used for electrical connection in any part of the system; 5) condensation or excessive moisture build-up anywhere within the system; 6) tester/instrumentation repeatability and reproducibility.  The performance variability of a marginally good device can also result in invalid failures.


Of the aforementioned causes of invalid rejections, improper electrical contact between the DUT and the contactor constitute a major slice of the pie for most companies. Improper contact between the DUT and the contactor can further be broken down into the following causes:  1) misalignment between the contactors and the DUT leads; 2) contactor wear-out or mechanical degradation; 3) oxidation/corrosion of any of the contactors or DUT leads; 4) moisture build-up or contamination on the contactors or DUT leads.     


The fact that bulk of invalid failures are caused by contactor problems has led many companies to equate electrical test confidence to the confidence in making good electrical contact between the contactor and the DUT.  After all, most of the other factors that contribute to invalid rejection can be corrected prior to the production release of the test process.  Ways of achieving this include: 1) excellent test/software design and debugging; 2) proper test guardbanding; 3) use of good test equipment; 4) use of reliable boards and small hardware; 5) use of a well-designed test floor; and 6) a sound management system.


Based on the premise that electrical test confidence depends solely on the ability to achieve good electrical connection,  'test confidence' may then be defined as the probability that good connection will be achieved every time a DUT is tested. Thus, a test confidence of 90% means that for every 100 devices tested,  10 of these devices will encounter a contact problem that will prevent them from being tested properly. 


If such a system tests 1000 devices, with 90% of them (or 900 units) being electrically good, only 810 units will pass the test.  Ninety (90) of the 900 good units will experience contact issues during testing, becoming invalid failures.  Subjecting these 90 invalid failures to retest will fail to recover all of them, since again, 10% of them (9 units) will become invalid failures.


Several retest cycles will eventually recover all the parts that have been invalidly rejected by the initial rounds of electrical testing, since all of them are good anyway.  Such a process for recovering invalid rejects, however, is inefficient and not cost-effective.  The value of recovering invalid failures through retesting is, therefore, determined by the system's test confidence as well.


The resulting first pass yield, Y1, is the product of the real or actual yield of the lot (denoted here as 'Y') multiplied by the test confidence of the ATE system (denoted here by 'C'), or Y1 = Y x C. Thus, if a test engineer knows the test confidence of a system, he can estimate the actual yield of the lot through the equation Y = Y1 / C


If the engineer decides to recover the invalid failures, he'll have to retest all the fall-outs from the first test (denoted here by 'R2').  The retest quantity R2 is equal to the initial quantity (Q) multiplied by the first pass rejection rate, (1-Y1).  Thus, R2 = Q (1-Y1).


The resulting yield of the retest (denoted here as 'Y2') is equal to the actual yield of the retest multiplied by C.  The actual yield of the retest (denoted here as 'YY') equals the number of invalid failures Rinvalid divided by the retest quantity R2. Thus, YY = Rinvalid/R2 and Y2 = (Rinvalid/R2) x C.


The number of invalid failures Rinvalid is equal to the initial test quantity Q multiplied by the difference between the actual yield of the lot Y and first pass yield of the lot Y1, or  Rinvalid = Q x (Y-Y1).  But Y = Y1/C, so Rinvalid = Q x (Y1/C - Y1).


Thus, Y2 = {[Q(Y1/C-Y1)] / [Q (1-Y1)]} x C = [(Y1/C-Y1)/(1-Y1)] x C. 

Simplifying, Y2 = (Y1(1-C))/(1-Y1), where Y2 is the expected yield of the retest based on the first pass yield Y1 and the test confidence C.


This equation can also be used by an engineer to compute for the test confidence exhibited by his test system, given first pass and retest yield data: 1-C = Y2(1-Y1)/Y1, or


C = 1 - [Y2(1-Y1) / Y1]


where C = confidence of your test system, Y1 is the first pass yield, and Y2 is the yield when the first pass rejects are retested.


According to Christopher Jones, author of the article "Analyze test Confidence to Enhance Throughput" upon which this article was based, their experience in testing millions of RF IC's every week has taught them the following:

1)  never run a test if the test confidence is less than 85%;

2)  retest of the rejects is not necessary if the test confidence exceeds 95%; and

3)  retest of the rejects is recommended if the test confidence is between 85%-95%.


Of course, the above guidelines may not be applicable to every company, since different device groups and package types are subject to different economic factors, as well as exhibit different sensitivities to contactor degradation.  It is the task, therefore, of a Test Manager to determine for his company how they can best apply the concept of test confidence management in improving their bottom lines. 


Given the very competitive atmosphere of the IC testing industry today, every Test Manager must know when to do a retest, when it is not economical to do so, and when a test system should not be used at all.  Being able to distinguish these situations from each other based on test confidence data and reacting to each of them appropriately is an important aspect of test engineering management.   


Primary Reference: Christopher Jones, M/A-COM Division of AMP, Lowell, MA, "Analyze Test Confidence to Enhance Throughput"; Test & Measurement World, 9/1/1999; through http://www.reed-electronics.com


See also:  Electrical Test Burn-in Strip TestingTest EquipmentTest Accessories


Back to Top




Copyright 2003-2005 www.SiliconFarEast.com. All Rights Reserved.