Fault classification in Digital to Analog Converter using machine learning
J. Ramesh, S. Nithyadevi, Govindaraj Vellingiri
Abstract
The cost of manufacturing Integrated Circuit (IC) is affected strongly by the test time, cost of test equipment and test procedure development. Mixed Signal Integration in an IC has analog circuits and digital circuits both combined in a single die. Combining both analog circuits and digital circuits leads to sophisticated high functionality design. The cost for testing those analog circuit parts are mostly dominated in total cost of testing a Mixed Signal IC. Considerable efforts are invested by IC manufacturers to minimise production costs in the design and testing of mixed signal IC. Classification and distinguishing catastrophic faults of analog integrated circuits are gaining importance in recent days in order to reduce IC test cost. In this work, Back Propagation Neural Network (BPNN) and Random Forest are used to classify the faults in Digital-to-Analog Converters (DACs). Performance comparison of BPNN and Random Forest is made based on their fault classification efficiency in which BPNN classifies the fault with 98.1% accuracy, Random Forest classifies the fault with 98.8% accuracy.