CV
General Information
Name | Hyunwuk Lee |
Contact | hyunwulee0519[at]gmail[dot]com |
Languages | Korean (native), English (fluent) |
Academic Projects
-
Neural Network Accelerators
- Complex Valued Neural Networks [First author, MICRO 2023], [First author, Under Review]
- Large Language Models [Co-author, ICCD 2024], [Co-author, On-going]
- Diffusion Models [Co-First-author, Accepted in HPCA 2025], [Co-author, On-going]
-
GPU and GPU Memory Systems
- Unified Memory Systems [Co-author, ISPASS 2023]
- GPU Register Files [Co-author, JSA] [Co-author, ISPLED 2024]
- Multi-GPU Data Transfer [Co-author, Under Review] [Co-author, JSA<\a>]
- Multi-Instance GPU [Co-author, On-going]
-
Systems for Machine Learning
- AI Workloads Acceleration in Mobile SoC [Co-author, On-going]
- Multi-Core Accelerator Scheduling [Co-author, Under Review]
- Processing in Memory for Sparse DNN [Co-author, Under Review]
-
Others
- Approximate Memory for Neural Networks [US Patent]
- Analytic Model for Multi-Core Systems [Co-author, Under Review]
- Diffusion Models [Co-author, Under Review]
Education
-
2018 - 2024 Integrated Master and Ph.D. Degree in Computer Architecture
- Yonsei University, Korea
- Advised by Won Woo Ro
- Embedded Systems and Computer Architecture Lab (eSCaL)
-
2014-2018 B.S. in Electrical and Electronic Engineering
- Yonsei University, Korea
- Advised by Won Woo Ro
Publication
-
HPCA 2025 Ditto: Accelerating Diffusion Model via Temporal Value Similarity (Accepted)
Sungbin Kim*, Hyunwuk Lee*, Wonho Cho, Mincheol Park, and Won Woo Ro (*Co-first author) -
ICCD 2024 AirGun: Adaptive Granularity Quantization for Accelerating Large Language Models [PDF]
Sungbin Kim, Hyunwuk Lee, Sungwoo Kim, Cheolhwan Kim, and Won Woo Ro -
ISLPED 2024 GUMSO: Gating Unnecessary On-Chip Memory Slices for Power Optimization on GPUs [PDF]
Seunghyun Jin, Hyunwuk Lee, and Won Woo Ro -
JSA 2024 SHREG: Mitigating Register Redundancy in GPUs [PDF]
Seunghyun Jin, Hyunwuk Lee, Jonghyun Lee, Junsung Kim, and Won Woo Ro -
MICRO 2023 Exploiting Inherent Properties of Complex Numbers for Accelerating Complex Valued Neural Networks [PDF]
Hyunwuk Lee, Hyungjun Jang, Sungbin Kim, Sungwoo Kim, Wonho Cho, and Won Woo Ro -
ISPASS 2023 Early-Adaptor: An Adaptive Framework for Proactive UVM Memory Management [PDF]
Seokjin Go, Hyunwuk Lee, Junsung Kim, Jiwon Lee, Myung Kuk Yoon, and Won Woo Ro
Professional Experience
-
Fall 2024 - System Architect
Samsung Electronic, Korea (Solution Platform Development) -
Spring 2018 Teaching Assistant
Yonsei University, Korea (EEE3530: computer architecture) -
2017 Research Intern
eSCaL at Yonsei University, Korea
Honors and Awards
-
2021 SK Hynix Industry-Academic Research Project Outstanding Invention
Industry Project
-
2024 Memory-Centric Architecture using the Reconfigurable PIM Devices
Institute for Information & communication Technology Planning & evaluation (IITP), Korea -
2022-2024 Developing Data Processing Unit for AI Workloads
Korea Evaluation Institute of Industrial Technology (KEIT), Korea -
2021-2022 Analysis on High Performance GPU Workloads and Architecture Design
Samsung Advanced Institute of Technology, Korea -
2021-2022 Development of High Performance Multi-GPU Memory System
National Research Foundation of Korea (NRF) , Korea -
2020-2024 Architectural Exploration of Parallel Execution Processing Units for Supercomputer CPU
National Research Foundation of Korea (NRF), Korea -
2018-2019 Performance Analysis of Neural Network Workloads and Development of Energy Efficient Approximate Memory for Neural Networks
SK Hynix, Korea
Patents
-
Memory device including a plurarity of area having different refresh periods, memory controller controlling the same and memory system including the same
- US Patent 11276452, Korea Application 10-2020-0045023
-
A method for neural network processing including memory-optimization techniques
- Korea Application 10-2022-0041848
-
Apparatus and method with register sharing
- US Application 18/ 315576, Korea Application 10-2022-0074653
-
Memory management unit and method of walking page table
- US Application 18/502058, Korea Application 10-2022-0175909
-
Memory management apparatus and method for UVM
- Korea Application 10-2023-0076606
-
Polar Form Aware Apparatus and Method for Complex Valued Neural Network and Rectangular Form Conversion Apparatus
- Korea Application 10-2023-0158538