Algorithm based adaptive parametric testing for outlier detection and test time reduction

Parallel test capability, enabled by numerous independent measurement channels has significantly increased throughput in parametric testing. It involves testing of numerous devices simultaneously synchronously or asynchronously. The number of devices tested for a given pad layout is increased by using higher dimensional arrays, the hallmark of which is pad sharing. Parallel testing of multiple devices with shared pads is vulnerable to device fails, where a failing device adversely affects measurement of all other devices. Information about this failing device or compromised measurement would only be evident at post analysis where a retest with a recipe change can then be ordered. In some cases retest is impossible as wafers would have already moved on to subsequent processing steps, thereby losing valuable learning opportunity. On the other hand, having to wait for post analysis requires time. Ideally failure detection and subsequent re-measure is done dynamically while the device is under test. This would require that decision making capability to be implemented in an automated tester equipment. In this work, we will discuss an algorithm based approach to adaptively change the test program allowing testing or skipping devices based on data collected real time while device is under test. The adaptive algorithm is also extended to aid in test time efficiency by eliminating tests based on measurement results of preceding tests.