Salland has been delivering adaptive test solutions to high volume production lines since 2004. We provide turnkey solutions as well as a robust set of building blocks that allow users to configure solutions limited only by their imagination. Salland has successfully brought the vision of adaptive test to reality and we invite you to explore the potential of this new technology.
What is Adaptive Test?
Adaptive test is a broad term used to describe various statistical methods for improving semiconductor testing by adjusting the test plan based upon feedback from the current test environment. Nowadays, it is assumed that these changes are occurring on a continuous basis via programmatic methods. A test plan consists of flows and limits.
The importance of adaptive test is that it can accommodate for process variations in a more precise way than can be done using traditional manual methods. By process, we mean anything affecting the device test including the base silicon, the test program, test equipment condition, etc. When a test engineer has to set static limits based upon worse case conditions, a great deal of inefficiency gets built into the test program. By adapting the test program to the current conditions, test time can be reduced; yield and quality can be improved.
There are four basic categories of adaptive testing: Feed-back, Feed-forward, flow control and changing limits. We present example of each these here.
Feed-back
A simple example of feed-back would be if the results for final test were automatically analyzed and the limits for certain tests would be updated for the next test run at wafer probe. You can see that a Device Profile is needed that accompanies the test program and can be changed on the fly.
Feed-forward
The diagram below illustrates the basic components of a feed-forward system. Imagine that you have to retest a device a different temperatures, but analysis reveals that if a specific test at step 1 at NORMAL temperature lands in a certain zone, it will never fail at HOT. Therefore, that test could be skipped at hot and test time could be saved. You can see that in addition to the Device Profile, some on-line decision-making software is required. Another example of feed-forward is how our dynamic test cell controller, DTC-WS, can skip die based upon the results from an earlier inspection map.
Changing Flow
This technique can be used to optimize retest, recover yield loss or reduce test time via statistical sampling. The diagram below is a highly simplified view of how SwifTest-TTO works to reduce test time based upon the quality of each test in each lot.
Changing Limits
Dynamic Part Average Testing is a good example of adaptive test improving quality be automatically adjusting the test limits to screen out statistical outliers. An outlier is a measurement that is within design specs, but is significantly different for the rest of the measurements in the current batch (lot). See SwifTest-FOX.
Salland has also helped customers recover yield loss due to site-site variation by automatically normalizing shifted results (See Qualcomm case study.) The diagram below illustrates how the limits for each individual test can be adjusted for each lot in order to perform outlier detection. The product behind this case study is SwifTest-DTI.
Read more about how Salland puts this into practice on the following pages: