Hi! I'm Meron Zerihun Demissie, a PhD student in Computer Science and Engineering at the University of Michigan, advised by Prof. Ronald Dreslinski.

I work in computer architecture, on hardware for reliable and trustworthy computing. Right now I study how GPUs can protect the memory that large-scale AI training depends on. I grew up in Addis Ababa, Ethiopia, and earned my B.Sc. in Software Engineering from the Addis Ababa Institute of Technology. I'm always happy to chat about research: mdemissi@umich.edu.

News

  • Our paper on HEIR, a universal compiler for homomorphic encryption, is now on arXiv.
  • Joined the CSE Graduate Committee (CSEG) board as Service Co-chair.
  • Back as a Graduate Student Instructor for EECS 370: Introduction to Computer Organization.
  • Started a graduate technical internship at Intel, working on the HEIR compiler.
Older news
  • 🎉 Received the CSE DEI Service Award.
  • Graduate Student Instructor for EECS 573: Advanced Topics in Computer Architecture.
  • Our paper Exploring the Efficiency of Data-Oblivious Programs appeared at ISPASS 2023.

Research

Current

Reliable memory for AI training

Large-scale AI training keeps a lot of state in GPU memory, and memory errors are a common reason training jobs stop. I study which corruptions of that state actually harm a model, and how hardware can use what it knows about the workload to protect training more efficiently.

Earlier

Hardware for privacy and integrity

Hardware that lets the cloud compute on sensitive data without exposing it, and that gives users verifiable evidence that a remote computation ran correctly, including safe, selective disclosure of results.

Publications

  • In submissionDash: A Hardware Approach for Safe Data Disclosures
    M. Zerihun Demissie, L. Biernacki, D. Ramotsoela, P. Grubbs, T. Austin
  • arXiv 2025HEIR: A Universal Compiler for Homomorphic Encryption
    A. Ali, J. Choi, B. Gipson, S. Gorantala, J. Kun, W. Legiest, L. Lim, A. Viand, M. Zerihun Demissie, H. Zheng
  • ISPASS 2023Exploring the Efficiency of Data-Oblivious Programs
    L. Biernacki, B. Mengist Tiruye, M. Zerihun Demissie, F. Assamnew Andargie, B. Reagen, T. Austin
  • SEED 2022Sequestered Encryption: A Hardware Technique for Comprehensive Data Privacy
    L. Biernacki, M. Zerihun Demissie, K. Birkayehu Workneh, F. Assamnew Andargie, T. Austin
  • SEED 2021VIP-Bench: A Benchmark Suite for Evaluating Privacy-Enhanced Computation Frameworks
    L. Biernacki, M. Zerihun Demissie, K. Birkayehu Workneh, G. Basha Namomsa, P. Gebremedhin, F. Assamnew Andargie, B. Reagen, T. Austin
Posters: DAC 2023 · GREPSEC · WISE 2022

Experience

  • Graduate Technical Intern, Intel Summer 2024 · HEIR, an MLIR compiler for homomorphic encryption
  • Graduate Student Instructor EECS 370 (Computer Organization) · EECS 573 (Advanced Computer Architecture)

Service

  • CSE Graduate Committee (CSEG) Service Chair · community and service events for graduate students
  • Founder, VoiCSEs Peer support for skill development and community
  • Research mentor, UMAPS and AURA Visiting scholars and undergraduate researchers