Research
In-Situ Recovery Algorithm and Simulation Frameworks for Ferroelectric (HZO) Neural Networks
Quantized integer-arithmetic training for in-situ drift recovery on HZO FeFET analog hardware.
Master's Thesis · 2026 Laboratory for Emerging and Exploratory Devices
Python PyTorch CrossSim Streamlit Gemini
- Implemented a quantized integer-arithmetic training library for in-situ drift recovery on HZO FeFET analog hardware, recovering +45pp accuracy on MNIST after 1M cycles.
- Built CrossSim UI, a full-stack Streamlit diagnostic framework wrapping Sandia National Labs’ CrossSim simulator to characterize analog neural network inference under realistic hardware conditions.
- Integrated a Gemini LLM agent with function calling, enabling natural language control of simulation pipelines.
