Online Monitoring and Performance Degradation Evaluation Method for Automotive-Grade RF Power Amplifiers Under Endurance Aging

Abstract

To meet AEC-Q102 reliability standards for automotive RF power amplifiers (PAs), traditional offline aging tests suffer from discontinuous data, inability to distinguish reversible thermal drift from permanent device damage, and oversimplified evaluation metrics. This paper proposes a complete endurance assessment system for automotive PAs. We build a multi-stress coupled online monitoring platform to continuously collect multi-dimensional performance parameters. A normalized Comprehensive Performance Degradation Index (CPDI) integrating failure physics and time-series features is proposed to quantify device degradation. A hybrid FP-LSTM prediction model constrained by failure physics is put forward to address weak generalization and low precision of conventional algorithms. A four-level hierarchical evaluation framework complying with AEC-Q102 and ISO 26262 is also formulated. Experiments on 24 GaAs/GaN PAs under 1000 h composite aging verify that the platform’s maximum measurement error is below 2.08% and can detect early hidden degradation. CPDI achieves 98.6% accuracy in identifying irreversible damage. The FP-LSTM model reduces lifespan prediction mean relative error to only 3.17%, boosting accuracy by 47.52% versus the classic Arrhenius model. The proposed system can be deployed on production lines to provide standardized quantitative support for automotive RF PA endurance certification.