Research

languages, compilers, and hardware

My current research bridges programming languages, compilers, and computer architecture — co-designing the whole stack so that programs use increasingly heterogeneous hardware efficiently, reliably, and securely, without pushing that complexity onto developers.

My past projects explore diverse memory technologies—such as secure memory and non-volatile main memory to enable privacy-enhanced computation and efficient security metadata management. Additionally, I have worked on library learning for imperative programming languages. You can find more details about these projects below, or browse my full list of publications.

Past Research Projects

Leroy

ACM SIGPLAN HATRA · 2024
PDF

Automatically finding the reusable helper functions hiding in a codebase. Leroy spots repeated patterns in ordinary imperative programs (like Python) and turns them into shared library functions, making code smaller and easier to maintain.

A Midsummer Night's Tree

ACM ASPLOS · 2024
PDF

Making secure non-volatile memory both fast and chip-frugal. By nesting a small "tree within a tree" for the integrity bookkeeping, it survives power loss, cuts runtime cost by ~41%, and uses far less precious on-chip space than prior designs.

Baobab Merkle Tree

IEEE CAL · 2024
PDF

Secure memory needs a lot of bookkeeping that wastes space. Baobab remembers (memoizes) frequently repeated counters in a small on-chip table, shrinking that storage overhead by 2–4× — with no performance penalty.

Sequestered Encryption

IEEE SEED · 2022
PDF

Keeping data private even when the software is compromised: a small trusted hardware unit computes directly on encrypted data and never lets the plaintext leave it, so attackers who break the software still can't read your secrets — at near-normal speed.

VIP-Bench

IEEE SEED · 2021
PDF

A common yardstick for privacy technology. It's a suite of standardized test programs that lets researchers fairly compare different systems that compute on encrypted data — making it clear which approach is faster or more capable, today and in the future.

All publications →