ML-Predicted EOTPR Reference Waveforms from CAD with Layout-Aware Correlation for Efficient Fault Isolation in Advanced Packages
ML-Predicted EOTPR Reference Waveforms from CAD with Layout-Aware Correlation for Efficient Fault Isolation in Advanced Packages
Tuesday, October 6, 2026: 1:50 PM
Summary:
Electro-Optical Terahertz Pulse Reflectometry (EOTPR) is widely used for fault isolation in advanced semiconductor packages, where interpretation of measured waveforms relies on comparison with reference responses. These reference waveforms are traditionally generated through physics-based simulation using trace models derived from CAD layout data, a process that remains time-consuming and inherently performed on a per-signal basis. In this work, we present a machine learning–based approach for predicting EOTPR reference waveforms directly from CAD-derived interconnect features. The model is trained on simulation-generated reference waveforms, enabling it to learn the relationship between interconnect geometry and waveform characteristics. This eliminates the need for per-signal simulation during inference and enables rapid, on-the-fly waveform generation within the production CAD environment. In addition, a layout-aware analysis capability is introduced to map waveform features to interconnect geometry, enabling interactive correlation between signal-domain behavior and physical structure. The proposed framework significantly accelerates waveform generation, expands signal coverage, and improves interpretability in failure analysis workflows.
Electro-Optical Terahertz Pulse Reflectometry (EOTPR) is widely used for fault isolation in advanced semiconductor packages, where interpretation of measured waveforms relies on comparison with reference responses. These reference waveforms are traditionally generated through physics-based simulation using trace models derived from CAD layout data, a process that remains time-consuming and inherently performed on a per-signal basis. In this work, we present a machine learning–based approach for predicting EOTPR reference waveforms directly from CAD-derived interconnect features. The model is trained on simulation-generated reference waveforms, enabling it to learn the relationship between interconnect geometry and waveform characteristics. This eliminates the need for per-signal simulation during inference and enables rapid, on-the-fly waveform generation within the production CAD environment. In addition, a layout-aware analysis capability is introduced to map waveform features to interconnect geometry, enabling interactive correlation between signal-domain behavior and physical structure. The proposed framework significantly accelerates waveform generation, expands signal coverage, and improves interpretability in failure analysis workflows.