Research Overview
Toward Reliability-Aware DTCO: developing physics-based reliability models and simulation software that predict device reliability and circuit aging from atomic defect properties.
First-principles calculations (DFT, machine-learning potentials, DASP, ILED) provide the atomic defect properties. These properties feed into the All-state Model and the simulators RASP and HSPICE to predict device reliability and circuit aging, including BTI, RTN, and related phenomena.
Electrical measurements from the eMSM protocol are inverted to recover the underlying atomic defect parameters. A support-vector / dS uniqueness criterion certifies the recovered parameters, closing the loop between simulation and experiment.
DFT / MLIP / DASP / ILED
Trap Level, Carrier Capture/Emission Rate
Electrical Measurements
SVD / Tikhonov / dS Criterion
GlobalTCAD, Sentaurus TCAD
HSPICE + OMI Compact Model
Education
Research Achievements
Atomic-Level Defect Property Simulation
- High-throughput defect calculation: Systematic first-principles study of intrinsic defect properties in SiO2, HfO2 and other gate dielectrics; established defect database.
- Software development: Led the development of ILED (Intelligent Learning Engine of Defects), an automated package that computes defect properties in IC materials. Also contributed to core modules of DASP (Defect & Dopant Ab-Initio Simulation Package).
ML-Accelerated Defect Property Calculation
- ML acceleration: Used machine-learning interatomic potentials (MLIPs) to speed up defect searches, cutting computational cost to 7.36% of DFT while keeping accuracy above 96%.
Device-Level Reliability Simulation Methods & Software
- High-throughput computation: Systematic study of oxygen vacancies in amorphous SiO2, identifying 7 stable configurations with a wide range of formation energies.
- Model innovation: Proposed the "All-state Model", a reliability model for Si/SiO2 MOSFETs that generalizes the earlier Two-state and Four-state Models. By accounting for the full set of defect configurations and transition paths in amorphous gate dielectrics, it corrects critical errors in BTI degradation prediction that the earlier models had missed. The model is being integrated into a major Chinese TCAD simulator.
- Software development: Led the development of RASP, a multi-scale reliability simulator that links atomic defect parameters to threshold voltage shifts at the device level.
Defect Parameter Extraction from Electrical Measurements
- Formulated BTI ΔVth degradation measured by the eMSM (extended Measure-Stress-Measure) protocol as an inverse problem with non-negativity constraints. Extracted NMP defect parameters using SVD, Tikhonov regularization, and PLS.
- Developed a global uniqueness certificate based on the support-vector / dS convex-hull-distance criterion, which proves that the extracted defect parameters are the unique non-negative solution over the candidate grid.
- Industry collaboration: The method covers both Si/SiO2 MOSFETs and FinFETs. Through an ongoing collaboration with the China Automotive Engineering Research Institute (CAERI), it is being applied to reliability test data from high-power automotive chips, extracting defect parameters that then feed into device-level simulations.
Circuit-Level Aging Compact Model Development
- Model innovation: Developing circuit-level aging models for MOSFETs and FinFETs based on the "All-state Model".
- Atom-to-circuit simulation: Combining the in-house defect parameter extraction methods with HSPICE to build a complete simulation chain from atomic defects to circuit aging.
Software Development
RASP
Core Developer 2024 – PresentReliability Ab initio Simulation Package
A reliability simulation package built on the All-state Model. Links first-principles defect parameters to device ΔVth predictions across multiple scales.
Documentation & User ManualILED
Core Developer 2023 – PresentIntelligent Learning Engine of Defects
An automated software package (copyrighted) for computing defect properties in IC materials. Uses machine-learning interatomic potentials, Hamiltonians, and related techniques to speed up atomic-scale calculations.
DASP
Contributor 2022 – 2023Defect & Dopant Ab-Initio Simulation Package
Contributed to core modules. Now commercially deployed with 150+ users across academia and industry, including Huawei HiSilicon, CATL, and others.
Documentation & User ManualSelected Publications
- "Si/SiO2 MOSFET reliability physics: From four-state model to all-state model." Physical Review Applied 24, 044040, 2025.
- "RASP: Reliability ab initio simulation package of MOSFETs based on all-state model." Microelectronics Reliability, Under Review.
- "An overlooked origin of NBTI in Si based MOSFETs." IEEE Electron Device Letters, Submitted.
- "Machine learning interatomic potentials accelerate defect exploration in amorphous silica." Physical Review Materials, 2025.
- "DASP: Defect and Dopant Ab-Initio Simulation Package." Journal of Semiconductors 43, 042101, 2022.