Quality Control of Lifetime Drift in Discrete Electrical Parameters in Semiconductor Devices via Transition Modeling

TL;DR

This paper introduces a semi-parametric, scalable model for predicting lifetime drift in discrete electrical parameters of semiconductors, improving early warning and quality control during production.

Contribution

It presents a novel, data-driven model specifically designed for discrete parameters, extending existing continuous models with faster computation and better handling of discrete data.

Findings

Enhanced prediction accuracy for discrete parameter drift

Faster calculation methods compared to existing models

Effective early warning system for semiconductor quality control

Abstract

Semiconductors are widely used in various applications and critical infrastructures. These devices have specified lifetimes and quality targets that manufacturers must achieve. Lifetime estimation is conducted through accelerated stress tests. Electrical parameters are measured at multiple times during a stress test procedure. The change in these Electrical parameters is called lifetime drift. Data from these tests can be used to develop a statistical model predicting the lifetime behavior of the electrical parameters in real devices. These models can provide early warnings in production processes, identify critical parameter drift, and detect outliers. While models for continuous electrical parameters exists, there may be bias when estimating the lifetime of discrete parameters. To address this, we propose a semi-parametric model for degradation trajectories based on longitudinal stress test data. This model optimizes guard bands, or quality guaranteeing tighter limits, for discrete electrical parameters at production testing. It is scalable, data-driven, and explainable, offering improvements over existing methods for continuous underlying data, such as faster calculations, arbitrary non-parametric conditional distribution modeling, and a natural extension of optimization algorithms to the discrete case using Markov transition matrices.