Fast deployment of alternate analog test using Bayesian model fusion
In this paper, we address the problem of limited training sets for learning the regression functions in alternate analog test. Typically, a large volume of real data needs to be collected from different wafers and lots over a long period of time to be able to train the regression functions with accuracy across the whole design space and apply alternate test with high confidence. To avoid this delay and achieve a fast deployment of alternate test, we propose to use the Bayesian model fusion technique that leverages prior knowledge from simulation data and fuses this information with data from few real circuits to draw accurate regression functions across the whole design space. The technique is demonstrated for an alternate test designed for RF low noise amplifiers.
