GOALI: A Machine-Learning Approach to Built-In Self-Test of Mixed-Signal/RF Circuits

Yale University, New Haven CT

Investigators

Abstract

ECS-0622081 Y. Makris, Yale University We will develop Built-In Self-Test (BIST) solutions for mixed-signal/RF circuits. BIST is a very important capability of electronic circuits, which allows them to examine their operational health in the field of operation and report potential malfunctions. The current state-of-the-art lacks BIST solutions for mixed-signal/RF circuits, mainly because most known test methods for these circuits rely on a functional test approach, which is impossible to implement on-chip. To mitigate this problem, in this project we will follow an alternative test approach based on machine-learning, wherein a neural classifier, trained through a representative chip population, will examine a set of simple measurements and will decide whether the chip is healthy or not. Efforts will be directed to two main areas, namely the design of on-chip circuitry for generation of test stimuli and acquisition of discriminative measurements and the design of neural classifiers for on-chip machine learning. Two mixed-signal/RF integrated circuits, namely a switched-capacitor filter and a low-noise amplifier will be designed and fabricated through National Semiconductor, the industrial collaborator of this project, in order to demonstrate the feasibility and effectiveness of machine learning-based BIST. The proposed research will be complemented by various educational and outreach activities, including the development of a new graduate-level seminar on Applications of Machine-Learning in Computer Aided Design and Test, participation of undergraduates in research, and promotion of active learning in the design of testable and reliable electronics. Intellectual Merit: This project aims to develop Built-In Self-Test (BIST) solutions for mixed-signal/RF circuits. BIST is a very important capability of electronic circuits, which allows them to examine their operational health in the field of operation and report potential malfunctions. The current state-of-the-art lacks BIST solutions for mixed-signal/RF circuits, mainly because most known test methods for these circuits rely on a functional test approach, which is impossible to implement on-chip. To mitigate this problem, this project follows an alternative test approach based on machine-learning, wherein a neural classifier, trained through a representative chip population, examines a set of simple measurements and decides whether the chip is healthy or not. Efforts will be directed to two main areas, namely the design of on-chip circuitry for generation of test stimuli and acquisition of discriminative measurements and the design of neural classifiers for on-chip machine learning. Two mixed-signal/RF integrated circuits will be designed and fabricated to demonstrate the feasibility and effectiveness of machine-learning-based BIST. Broader Impact: This project will facilitate the realization of testable and reliable electronic circuits and systems, thus extending their deployment in a broad range of applications, enabling reliable computing, and fostering technology trustworthiness. The proposed research is complemented by various educational and outreach activities, including the development of a new graduate-level seminar on Applications of Machine-Learning in Computer Aided Design and Test, participation of undergraduates in research, and promotion of active learning in the design of reliable electronics through involvement with the Yale University solar car racing Team Lux.