Enhancing confidence in indirect analog/RF testing against the lack of correlation between regular parameters and indirect measurements

The greedy specification testing remains mandatory for analog and Radio Frequency (RF) integrated circuits because of the accuracy of the sorting based on these measurements. Unfortunately, to be implemented, this kind of testing method often incurs very high costs (expensive instruments, long test time...). A common approach, in the literature, is the so-called indirect/alternate test strategy. This strategy consists in deriving targeted specifications from low-cost Indirect Measurements (IMs). During the industrial test phase, the estimation of regular specifications using IMs is based on a correlation model that has been built previously, during a training phase. Despite the substantial test cost reduction offered by this strategy, its deployment in industry is limited, mainly because of a lack of confidence in the accuracy of estimations made by the correlation model. A solution to increase the confidence in the estimation of specifications using the indirect approach is to implement redundancy in the prediction phase. In this paper, we demonstrate that the redundancy implementation brings more than identifying rare misjudged circuits from a high-correlated model. Indeed redundancy massively increases the accuracy despite of the lack of accurate models that have been assumed in previous implementations of redundant indirect testing. This approach is illustrated on a real case study for which we have experimental measurements on a set of 10,000 devices. KeywordsTest; Analog/RF integrated circuits, alternate test, indirect test, prediction error, machine-learning algorithm