Sungju Kim (김성주)

Graduate Research Assistant, CASLAB, Yonsei University

Education

• Ph.D. Student in Electrical and Electronic Engineering, Yonsei University, Mar. 2025–Present

  • Advisor: Prof. Daehoon Kim

• M.S. in Electrical Engineering and Computer Science, DGIST, Feb. 2025

  • Advisor: Prof. Daehoon Kim

  • Thesis: “MX-Format Quantization Optimization for Lightning-NeRF”

• B.S. in Transdisciplinary Studies, DGIST, Aug. 2022

Research Interest

• Computer Architecture and Systems

  • ML Systems: Optimization of machine learning models and workloads across hardware and software stacks, including 3D Gaussian Splatting

  • Hardware Security: Defense mechanisms against RowHammer attacks

News

• 2026.07.24 

  • A paper was accepted for publication in IEEE Computer Architecture Letters (CAL). 

• 2026.07.08 

  • A paper was accepted to MICRO 2026. 

• 2026.06.07 

  • A paper was accepted for publication in IEEE Computer Architecture Letters (CAL).

• 2026.03.28 

  • Two papers were accepted to ISCA 2026. 

• 2026.03.23 

  • A paper was accepted for publication in IEEE Computer Architecture Letters (CAL).

• 2025.05.19 

  • A paper was accepted to ISLPED 2025.

• 2024.07.09 

  • A paper was accepted to PACT 2024.

Skills

• Programming Languages: C, C++, Python, Bash, CUDA 

• Machine Learning Frameworks: PyTorch

• Architecture and Memory Simulators: Ramulator, Ramulator 2, ChampSim, DRAMSim3

Teaching Experience

• Teaching Assistant, SYS3209: System Programming, Yonsei University, Spring 2026 

• Teaching Assistant, GEV6154: Multi-core and GPU, Yonsei University, Fall 2025 

• Teaching Assistant, SYS4211: Advanced Operating Systems, Yonsei University, Fall 2025 

• Teaching Assistant, SYS3209: System Programming, Yonsei University, Spring 2025 

• Teaching Assistant, CSE305: Computer Architecture, DGIST, Spring 2024

• Teaching Assistant, CSE305: Computer Architecture, DGIST, Spring 2023 

• Teaching Assistant, R&E Program with Daegu Il Science High School, DGIST, Mar.–Dec. 2023

Projects

• Chips and AI Innovation Center, IITP-Yonsei Univ, 2025–2030 

• Industry–Academia Collaboration Project, Samsung Electronics, 2025–2026 

• ML Compiler for On-Device AI, IITP–ETRI, 2025–2027 

• Data Utilization Planning and Verification for Fine-Tuning and Inference of On-Device Generative Models, NIA, 2024 

• Development of an On-Device NPU Architecture for Accelerating Generative AI Models, Samsung Research, 2024 AI Training Data Construction Project, NIA, 2023 

• Research on Memory-Centric Next-Generation Computing System Architectures, IITP, 2022

Contact

Email: sungju.kim [at] yonsei.ac.kr

Address : Room 212D, Engineering Research Park, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul, Republic of Korea.

ORCID · Google Scholar

Publication (International, *Corresponding author.) 

• ResidualEvil: Exploiting Post-Squash MSHR Contention for Transient Execution Attacks

• IEEE Computer Architecture Letters (CAL), 2026

• Minwoo Jang, Hyosang Kim, Sungju Kim, and Daehoon Kim*

• AMG: AMX–GPU Cooperative Acceleration of Billion-Scale ANNS via GEMM Reformulation

• IEEE/ACM International Symposium on Microarchitecture (MICRO), 2026

• Minho Kim, Houxiang Ji, Hwanjun Lee, Jaeyoung Kang, Sungju Kim, Hyunkyun Shin, Daehoon Kim*, and Nam Sung Kim

• NDPool: Correctness-Preserving Shared Execution for Efficient LLM Inference on CXL-NDP Systems

• IEEE Computer Architecture Letters (CAL), 2026

• Yeji Jung, Hwanjun Lee, Sungju Kim, Seulki Kim, Yunhyeong Jeon, and Daehoon Kim*

• DarkStream: Exploiting Internal Throughput Contention in Data Streaming Accelerator for Timing Attacks

• International Symposium on Computer Architecture (ISCA), 2026

• Hyosang Kim, Ki-Dong Kang, Gyeongseo Park,  Sungju Kim,  Daehoon Kim*

• SMOOTH: Hardware-Assisted Fine-Grained On-Chip Memory Management for Efficient On-Device LLM Inference

• International Symposium on Computer Architecture (ISCA), 2026

• Seulki Kim, Bokyeong Kim, Kyeonghyeon Ryu, Yeji Jung , Hwanjun Lee,  Sungju Kim, Yunhyeong Jeon, Daehoon Kim*

• Nighthawk: Zero-Copy Cache Quarantine for Invisible Speculation

• IEEE Computer Architecture Letters (CAL), 2026

• Minwoo Jang, Sungju Kim, and Daehoon Kim*

• Jack Unit: An area- and energy-efficient multiply-accumulate (MAC) unit supporting diverse data formats

• IEEE/ACM International Symposium on Low Power Electronics and Design (ISLPED), 2025

• Seock-Hwan Noh, Sungju Kim, Seohyun Kim, Daehoon Kim, Jaeha Kung* and Yeseong Kim*

• vSPACE: Supporting Parallel Network Packet Processing in Virtualized Environments through Dynamic Core Management

• International Conference on Parallel Architectures and Compilation Techniques (PACT), 2024

• Gyeongseo Park, Minho Kim, Ki-Dong Kang, YunHyeong Jeon, Sungju Kim, Hyosang Kim, and Daehoon Kim*.

Publication (Domestic, *Corresponding author.) 

• CPU 오프로딩을 통한 NPU 기반 대규모 언어 모델 서비스의 성능 최적화 방안 연구

• KSC(한국소프트웨어종합학술대회), 2025

• 장예원, 정찬섭, 전윤형, 김성주, 김대훈*.

• 온디바이스 생성모델 어댑터 기반 파인튜닝 및 추론 성능 분석

• KSC(한국소프트웨어종합학술대회), 2024

• 정예지, 김슬기, 김정보, 유경현, 김성주, 신현균, 오선무, 좌훈승, 김대훈*.