Dependence of stretched exponential function-based lifetime prediction accuracy on trapping dynamics in enhancement-mode oxide thin-film transistors

Accurate degradation prediction due to bias stress is crucial for assessing the reliability of thin-film transistors (TFTs). The typical stretched exponential function (SEF) is a widely used prediction model in academia and industry. This study reveals that the typical SEF could lead to significant errors in long-term degradation prediction due to time-dependent trapping dynamics of indium oxide (In 2 O 3 ) semiconductors. It systematically investigates time-dependent trap dynamics and the resulting SEF-based lifetime prediction accuracy by obtaining periodic transfer characteristics when applying positive bias stress for up to 5 × 10 4 s in In 2 O 3 TFTs. Rigorous analysis reveals significant fluctuations in the linearity due to extrapolation within the fitting range. To elucidate the physical origin, the recoverable shallow trap and the permanent deep trap components are experimentally decomposed using multiple stress-recovery analyses. The results show a gradual transition from early degradation (dominated by shallow traps) to late degradation (dominated by deep traps), violating the SEF assumption of statistically stable dynamics. Based on this insight, a time-windowed SEF strategy is proposed that excludes the early shallow trap-dominated region in threshold voltage shift under long-term stress. This approach improves long-term prediction error compared to conventional SEF by balancing bias rejection, noise sensitivity, and throughput while maintaining practical measurement time. • Conventional stretched exponential function (SEF)–based lifetime prediction can be intrinsically inaccurate when early-time data dominate the fitting window. • Long-term positive bias stress measurements of indium oxide thin-film transistors demonstrate this limitation as pronounced fitting-range dependence in SEF-based extrapolation of threshold voltage shift (ΔV TH ). • Stress–recovery analyses experimentally separate ΔV TH into recoverable shallow-trap and persistent deep-trap components, showing a progressive transition in the dominant degradation mechanism with stress time. • This time-dependent transition violates the implicit SEF assumption of statistically stable trapping dynamics, leading to fitting-range-dependent β and τ parameters and degraded long-term prediction accuracy. • A time-windowed SEF (x t2 –x t3 ) strategy that excludes the shallow-associated components improves target-time ΔV TH prediction accuracy while preserving practical measurement time.