CV

Download

Research Interests

I am interested in a secure and efficient architecture of hardware accelerators (such as GPU and NPU).

My research objective is to design high-performance accelerators with security guarantees. To achieve this goal, my recent studies focus on 1) hardware security and 2) performance improvement of hardware accelerators.

Hardware Security of Accelerators : As accelerators are widely used in mission-critical tasks, the importance of security gets larger. Although I extended Trusted Execution Environment to GPU and NPU in previous works, countless security weaknesses still remain. Therefore, I aim to increase the security level to resist unintended operations.

Performance Improvement of Accelerators : Since machine learning requires speedy processing, the performance improvement of accelerators is crucial. Hence, I consider both hardware and software to enhance parallelism or cut down unnecessary procedures. In a recent publication, I proposed the fine-grained scheduling algorithm in GPU by leveraging Multi-Process Service. I set the further reduction of the execution time as a future research direction.

From these two sub-goals, I target combining a trusted system with a high-performance accelerator design. It is expected to protect users from accidents (caused by attackers or extreme environments) within a reasonable latency.

Education

KAIST, Daejeon, Republic of Korea, Mar 2021 - present
  Ph.D. Student, School of Computing
  Advisor: Jaehyuk Huh

KAIST, Daejeon, Republic of Korea, Mar 2019 - Feb 2021
  Master of Science, School of Computing
  Advisor: Jaehyuk Huh
  Thesis: Hardware Security Techniques for Trusted Machine Learning Accelerators

Yonsei University, Seoul, Republic of Korea, Mar 2015 - Feb 2019
  Bachelor of Science, Computer Science

Publications

• Seonjin Na, Jungwoo Kim, Sunho Lee, and Jaehyuk Huh, "Supporting Secure Multi-GPU Computing with Dynamic and Batched Metadata Management", the 30th IEEE International Symposium on High-Performance Computer Architecture ( HPCA ), March 2024
• Jungwoo Kim, Seonjin Na, Sanghyeon Lee, Sunho Lee, and Jaehyuk Huh, "Improving Data Reuse in NPU On-chip Memory with Interleaved Gradient Order for DNN Training", the 56th IEEE/ACM International Symposium on Microarchitecture ( MICRO ), October 2023
• *Soojin Hwang, *Sunho Lee, Jungwoo Kim, Hongbeen Kim, Jaehyuk Huh, "mNPUsim: Evaluating the Effect of Sharing Resources with Multi-Core NPUs", the 2023 IEEE International Symposium on Workload Characterization ( IISWC ), October 2023 (* co-first authors)
• Seungho Lee, Sunho Lee, Jaehyuk Huh, and Sejin Kwon, "Proposal of Aerospace-informatics by Design of Ramjet Inlet Using Machine Learning", the 2023 Aerospace Europe Conference ( AEC ) joint event between the 10th European Conference for Aerospace Sciences ( EUCASS ) and the 9th Council of European Aerospace Societies ( CEAS ), July 2023
• Sunho Lee, Seonjin Na, Jungwoo Kim, Jongse Park, and Jaehyuk Huh, "Tunable Memory Protection for Secure Neural Processing Units", the 40th IEEE International Conference on Computer Design ( ICCD ), October 2022
• Seungbeom Choi, Sunho Lee, Yeonjae Kim, Jongse Park, Youngjin Kwon, and Jaehyuk Huh, "Serving Heterogeneous Machine Learning Models on Multi-GPU Servers with Spatio-Temporal Sharing", the 2022 USENIX Annual Technical Conference ( USENIX ATC ), July 2022
• Sunho Lee, Jungwoo Kim, Seonjin Na, Jongse Park, and Jaehyuk Huh, "TNPU: Supporting Trusted Execution with Tree-less Integrity Protection for Neural Processing Unit", the 28th IEEE International Symposium on High-Performance Computer Architecture ( HPCA ), Feburary 2022
• Seonjin Na, Sunho Lee, Yeonjae Kim, Jongse Park, and Jaehyuk Huh, "Common Counters: Compressed Encryption Counters for Secure GPU Memory", the 27th IEEE International Symposium on High-Performance Computer Architecture ( HPCA ), Feburary 2021

Patents

• [Pending] Jaehyuk Huh, Jungwoo Kim, Seonjin Na, Sanghyeon Lee, and Sunho Lee, "Improving the Utilization of NPU On-chip Memory with Computation Rearrangement for DNN Training", Korean Patent
• [Pending] Jaehyuk Huh, Seonjin Na, Jungwoo Kim, and Sunho Lee, "Dynamic One-time Pad Table Management for Secure Multi-GPU Communication", Korean Patent
• [Pending] Jaehyuk Huh, Sunho Lee, and Seonjin Na, "Apparatus and Method for Providing Secure Execution Environment for NPU", US Patent (with Samsung Electronics)
• [Pending] Jaehyuk Huh, Seungbeom Choi, Sunho Lee, Yeonjae Kim, Youngjin Kwon, Jongse Park, "Machine Learning Inference Time-spatial SW Scheduler Based on Multiple GPU", Korean Patent
• [Pending] Jaehyuk Huh, Sunho Lee, and Seonjin Na, "Hardware-based Security Architecture for Trusted Neural Processing Unit", Korean Patent (with Samsung Electronics)
• [10-2365263-0000] Jaehyuk Huh, Seonjin Na, Sunho Lee, Yeonjae Kim, and Jongse Park, "Efficient Encryption Method and Apparatus for Hardware-based Secure GPU Memory", Korean Patent

Research Experiences

KAIST, Daejeon, Republic of Korea, Mar 2019 - present
Ongoing Researches at CASYS (Computer Architecture and SYStem) Lab
Advisor: Jaehyuk Huh
  Accelerator Hardware-based Security
    - Memory protection optimization for GPU: Common counters for duplicate counters (Published in HPCA 2021)
    - Memory protection optimization for multi-GPU system (Published in HPCA 2024)
    - Trusted execution environment for NPU: Tensor-granularity counters (Published in HPCA 2022)
    - Memory protection optimization for NPU: Parital memory protection (Published in ICCD 2022)
    - Side-channel attack protection for NPU
    - Dynamic secure-granularity management for heterogeneous processors
  Accelerator Performance
    - Multi-tenancy support for a multi-GPU system: Time and spatial sharing (Published in USENIX ATC 2022)
    - Accurate multi-NPU simulation: Multi-NPU simulator attached with DRAMsim3 (Published in IISWC 2023)
    - On-chip memory management for training NPU: Access order rearrangement (Published in MICRO 2023)

Yonsei University, Seoul, Republic of Korea, Sep 2017 - June 2018
Undergraduate Research Intern at ELC (Embedded systems Languages and Compilers) Lab
Advisor: Bernd Burgstaller
  Parallelism
    - Accelerating big-data streaming engine: Multi-thread and shared-memory
    - Parallelization of SFA (Simultaneous Deterministic Finite Automata) construction: MPI and Huang’s algorithm

Recognition

KAIST, Daejeon, Republic of Korea
  Outstanding Teaching Assistant Award - CS311 Computer Organization, Spring 2022, Fall 2019
  Outstanding Teaching Assistant Award - CS230 System Programming, Fall 2023

Yonsei University, Seoul, Republic of Korea
  Dean’s List, Spring 2015, Spring 2018
  Undergraduate Capstone Project Award (Third Place) - Project Leader, Spring 2018
    Title: Cloud SFA: Parallel Construction of Simultaneous Deterministic Finite Automata in Distributed System

Samsung Electronics, Hwaseong, Republic of Korea
  Best Paper Award (Third Place), Summer 2022
    Title: TNPU: Supporting Trusted Execution with Tree-less Integrity Protection for Neural Processing Unit

Participation

uArch (in conjunction with ISCA 2022), New York City, United States of America
Student Panel
  Life in Grad School, June 2022

Skills

Programming Languages: C, C++, Python
NPU Simulators: SCALE-Sim, MAESTRO, Gemmini
GPU Programming: CUDA, MPS
Multi-core CPU Programming: MPI, OpenMP
Machine Learning Frameworks: Pytorch, Tensorflow

Teaching Experiences

KAIST, Daejeon, Republic of Korea
Teaching Assistant
  CS230 System Programming, Fall 2023, Fall 2021
  CS311 Computer Organization, Fall 2022, Spring 2022, Spring 2021, Fall 2019
  CS211 Digital System and Lab, Spring 2019

KAIST Education Center, Daejeon, Republic of Korea
Mentor & Lecturer
  Seocho AI College, Summer 2019, Summer 2021
  Python for Beginners, Summer 2022, Summer 2021