Regression modeling for digital test of ΣΔ modulators

Abstract

The cost of Analogue and Mixed-Signal circuit testing is an important bottleneck in the industry, due to timeconsuming verification of specifications that require state-ofthe- art Automatic Test Equipment. In this paper, we apply the concept of Alternate Test to achieve digital testing of converters. By training an ensemble of regression models that maps simple digital defect-oriented signatures onto Signal to Noise and Distortion Ratio (SNDR), an average error of 1:7% is achieved. Beyond the inference of functional metrics, we show that the approach can provide interesting diagnosis information.