Assisted test design for non-intrusive machine learning indirect test of millimeter-wave circuits

Abstract

The functional test of millimeter-wave (mm-wave) circuitry in the production line is a challenging task that requires costly dedicated test equipment and long test times. Machine learning indirect test offers an appealing alternative to standard mm-wave functional test by replacing the direct measurement of the circuit performances by a set of indirect measurements, usually called signatures. Machine learning regression algorithms are then used to map signatures and performances. In this work, we present a generic and automated methodology for finding an appropriate set of indirect measurements and assisting the designer with the necessary Design-for-Test circuit modifications. In order to avoid complex design modifications of mm-wave circuitry, the proposed strategy is targeted at generating a set of non-intrusive indirect measurements using process variation sensors not connected to the Device Under Test (DUT). The proposed methodology is demonstrated on a 60 GHz Power Amplifier designed in STMicroelectronics 55 nm BiCMOS technology.