A Structural Causal Model for Electronic Device Reliability: From Effects to Counterfactuals

TL;DR

This paper develops a structural causal model to analyze electronic device reliability, enabling estimation of causal effects and counterfactuals from limited and censored data, with applications demonstrated through synthetic data.

Contribution

It introduces a novel parametric causal model that integrates observational and experimental data to assess device failure mechanisms under various stress conditions.

Findings

Effective estimation of causal effects from censored data

Ability to predict counterfactual device failure scenarios

Demonstrated model applicability with synthetic data

Abstract

Electronic devices exhibit changes in electrical resistance over time at varying rates, depending on the configuration of certain components. Since measuring overall electrical resistance requires partial disassembly, only a limited number of measurements are performed over thousands of operating hours. This leads to censored failure times, whether under natural stress or under accelerated stress conditions. To address these challenges, including device-specific failure thresholds, a parametric structural causal model is developed to extract information from both observational and experimental data, with the aim of estimating causal effects and counterfactuals, regardless of the applied stress regime. Synthetic data are used to illustrate the main findings.