Sarabjeet Singh
Photo of Sarabjeet Singh

Sarabjeet Singh

Designing custom hardware —
for AI and for privacy-preserving computing

I am a Senior Engineer in Performance Architecture at Samsung Semiconductor in San Jose, California. I work on our hardware performance simulator, and I study how trends in large language models (LLMs) change hardware metrics, so that those trends can drive architectural decisions for next-generation AI inference hardware.

My work spans computer architecture, memory systems, and hardware/software co-design, applied to AI inference and privacy-preserving computing.

Before Samsung, I was a Scientist at Duality Technologies (2024–2026). There I designed and implemented components of TREBUCHET, a $15M DARPA DPRIVE effort (2021–2025) that built a custom 12 nm fully homomorphic encryption (FHE) accelerator ASIC for encrypted machine and deep learning, and I was Technical Lead for ARPA-H SHARE (Sharing Data for Rapid Collaboration on Encrypted Data), a privacy-preserving platform for healthcare collaboration built with ICF and Dana-Farber Cancer Institute.

I received my Ph.D. in Computer Science from the University of Utah in 2024, advised by Prof. Rajeev Balasubramonian in the Utah Arch Research Group. My thesis was on accelerating modern cryptography, with a focus on data movement. During my Ph.D. I was a research co-op at NVIDIA Research (2022) and AMD Research (2020). I received my B.Tech. from IIT Gandhinagar in 2017.

News

Show older newsHide older news

Publications

Conference Journal U.S. government conference Workshop Citations on Google Scholar

Conference and journal papers

  1. IEEE S&P2024

    Hyena: Balancing Packing, Reuse, and Rotations for Encrypted Inference

    Sarabjeet Singh, Shreyas Singh, Sumanth Gudaparthi, Xiong Fan, Rajeev Balasubramonian

    IEEE Symposium on Security and Privacy (S&P), 2024

  2. IEEE Micro2023

    XCRYPT: Accelerating Lattice-Based Cryptography with Memristor Crossbar Arrays

    Sarabjeet Singh, Xiong Fan, Ananth Krishna Prasad, Lin Jia, Anirban Nag, Rajeev Balasubramonian, Mahdi Nazm Bojnordi, Elaine Shi

    IEEE Micro, vol. 43, no. 5, 2023

  3. ACM TACO2023

    HyGain: High-Performance, Energy-Efficient Hybrid Gain Cell-Based Cache Hierarchy

    Sarabjeet Singh, Neelam Surana, Kailash Prasad, Pranjali Jain, Joycee Mekie, Manu Awasthi

    ACM Transactions on Architecture and Code Optimization (TACO), vol. 20, no. 2, 2023

  4. IEEE HPCA2022

    CANDLES: Channel-Aware Novel Dataflow-Microarchitecture Co-Design for Low Energy Sparse Neural Network Acceleration

    Sumanth Gudaparthi, Sarabjeet Singh, Surya Narayanan, Rajeev Balasubramonian, Visvesh Sathe

    IEEE International Symposium on High-Performance Computer Architecture (HPCA), 2022

  5. IEEE IISWC2019

    Efficacy of Statistical Sampling on Contemporary Workloads: The Case of SPEC CPU2017

    Sarabjeet Singh, Manu Awasthi

    IEEE International Symposium on Workload Characterization (IISWC), 2019

    Traces of the simulation points are available; see Software.

  6. ICPE2019

    Memory Centric Characterization and Analysis of SPEC CPU2017 Suite

    Sarabjeet Singh, Manu Awasthi

    ACM/SPEC International Conference on Performance Engineering (ICPE), 2019

  7. ACM JETC2019

    PANE: Pluggable Asynchronous Network-on-Chip Simulator

    Sneha N. Ved, Sarabjeet Singh, Joycee Mekie

    ACM Journal on Emerging Technologies in Computing Systems (JETC), vol. 15, no. 1, 2019

Government conference papers

  1. GOMACTech2025

    Accelerating FHE to Enable Privacy-Preserving Deep Learning

    Ahmad Al Badawi, Yuriy Polyakov, Ronen Cohen, Carlo Pascoe, Sarabjeet Singh, David Bruce Cousins, Kurt Rohloff, Matthew French, Benedict Reynwar, Troy Hicks, Kellie Canida, Clynn Mathew, Ajey Jacob, Kenneth M. Zick, Akshath Mahajan, Mukund Ramakrishnan, Rugved Mhatre, Austin Ebel, Brandon Reagen

    Government Microcircuit Applications & Critical Technology Conference (GOMACTech), Pasadena, CA, March 2025, paper 37.2

Workshop papers and other writing

  1. CogArch2022

    EPIC: Efficient Packing for Inference using Cheetah

    Sarabjeet Singh, Shreyas Singh, Rajeev Balasubramonian

    6th Workshop on Cognitive Architectures (CogArch), held with HPCA, 2022

  2. Blog2021

    Post Quantum Cryptography

    Sarabjeet Singh

    ACM SIGARCH Computer Architecture Today, February 2021

Experience

  1. Sep 2026 – Present

    Senior Engineer, Performance Architecture

    Samsung Semiconductor · San Jose, CA

    • Work on the hardware performance simulator for next-generation AI inference hardware.
    • Study how trends in large language models (LLMs) change hardware metrics, to drive architectural decisions.
  2. Jul 2024 – May 2026

    Scientist

    Duality Technologies

    TREBUCHET: 12 nm FHE accelerator ASIC · DARPA DPRIVE

    A $15M DARPA effort (Apr 2021 – Nov 2025, completed) to build a custom 12 nm ASIC for homomorphic machine and deep learning, carried to a fully verified RTL implementation and top-level floorplan (176 mm², 1 GHz timing closure). The design includes a custom ISA, a microcode scheduler, and a node-array architecture, and was validated on encrypted CNN training and inference and AES transciphering.

    My work: microcode generators for FHE kernels, kernel-to-hardware mapping, dataflow optimization, and a KeySwitching dataflow that improves data reuse; I also managed the technical collaboration with the NYU subcontractor team led by Prof. Brandon Reagen on simulator development.

    Paper: “Accelerating FHE to Enable Privacy-Preserving Deep Learning,” GOMACTech 2025.

    ARPA-H SHARE: Sharing Data for Rapid Collaboration on Encrypted Data

    ARPA-H award to Duality Technologies (Scalable Solutions Office; up to $6.3M; started September 2024).

    My role: Technical Lead. Privacy-preserving healthcare analytics across distributed patient-data owners, combining FHE, federated learning, differential privacy, and post-quantum cryptography, with ICF and Dana-Farber Cancer Institute.

  3. May 2022 – Aug 2022

    Research Co-op

    NVIDIA Research · Homomorphic Encryption at Modern GPU Datacenter Scale

    • Explored mappings of FHE kernels on multi-GPU architectures to maximize shared-memory reuse and reduce communication overheads.
    • Evaluated acceleration opportunities using in-switch compute, NVSwitch, AllReduce/Broadcast/Transpose, and kernel fusion.
  4. May 2020 – Aug 2020

    Research Co-op

    AMD Research · Processing-in-Memory

    • Profiled BERT/NLP workloads to identify code segments suitable for processing-in-memory acceleration.
    • Automated a PIM code-generation flow translating x86 instructions to PIM instructions, memory requests, and PIM/DRAM commands.
  5. Aug 2019 – Jun 2024

    Graduate Research Assistant

    University of Utah · Utah Arch Research Group · Advisor: Prof. Rajeev Balasubramonian

    • Ph.D. research on accelerating modern cryptography with a focus on data movement, memory systems, and hardware/software co-design.
    • Accelerator and memory-system techniques for FHE, lattice-based cryptography, sparse neural networks, and cache/memory hierarchies.
    Other Ph.D. research projectsHide projects
    • Efficient Integrity Verification using Custom DIMM: reducing the cost of integrity verification in trusted environments with near-memory support.
    • ZPU, Zero-Knowledge Proof Processing Units: high-throughput processing units for MSM/NTT kernels in zero-knowledge proofs.
    • Reducing Expensive Key Movement in FHE using Processing-in-Memory: partitioning FHE kernels on a CPU-UPMEM model.
    • Secure AI using Samsung’s AxDIMM (Oct 2021 – Jun 2022): finalist, Samsung Open Innovation Contest for AxDIMM Technology.
    • Efficient Metadata for Memory Protection (Aug 2020 – Mar 2021).
    • AMBOP, Adaptive Multiple Best Offset Prefetcher (Mar 2019 – Aug 2019).

Earlier positions

Service and Recognition

Software and Artifacts

Education

Contact

Reach me at sarab@cs.utah.edu. I’d love to talk, about work or otherwise.