research
Computer arithmetic, compilation, and physical implementation
My research concerns the representation, transformation, and implementation of numerical computations. I study how arithmetic choices affect accuracy, energy, throughput, and circuit area.
Research directions
- Arithmetic compilation: preserve numerical structure through MLIR transformations and lower it to explicit datapaths with HAriCo, FloPoCo, and CIRCT.
- Tensor accelerators: generate systolic arrays with configurable number formats and accumulation schemes.
- Physical evaluation: use open EDA flows to evaluate generated circuits and compare compiler decisions against implementation results.
Work and resources
- Publications: 20 papers, posters, talks, and thesis records.
- Software: components used from model import to physical implementation.
- Open-silicon projects: designs, shuttle participation, and public artifacts.
- Talks and teaching: presentations and course material.
- Results and contributions: publications and upstream changes.