Combining adaptive alternate test and multi-site

Testing analog, mixed-signal and RF circuits represents one of the main cost components for complex SoCs. Multisite Testing is widely accepted as a straightforward technique to reduce the effective test time. This paper shows that an adaptive Alternate Test approach can be compatible with a multisite strategy. The proposed solution consists in ordering offline the signatures acquisition sequence and training incremental regression models for each new feature. These models can be used to diagnose the circuit as good, provided that the estimate of the performance is larger than the specification plus a guard-band related to the model error. If all the sites are diagnosed as good, the test program can be halted before completion. This decision is taken on-line and makes this scheme adaptive. We provide an analytical study of the expected test time reduction and of the test escape penalty that is incurred. Results obtained from post-layout MonteCarlo simulations of an LNA demonstrate the validity of the approach and show that significant test time improvements can be obtained, even for large number of sites, whenever the manufacturing yield is sufficiently high.