Sunwook Hwang, Ph.D.

NPU Performance Architect for AI Systems

Digital IP/Circuit Chip Design · SOC IP Development Team (S.LSI)
Samsung Electronics

I work across the AI systems stack, from model behavior and workload characteristics to accelerator architecture and inference optimization. At Samsung Electronics, I drive architecture and optimization work for flagship Exynos NPUs, including performance modeling, workload characterization, compiler-aware scheduling, memory/dataflow tuning, and inference-time optimization.

My research has focused on AI systems for 3D perception and autonomous driving, covering semi-supervised object detection, model inversion attacks on 3D feature representations, video analytics, and V2X communication. This background gives me a cross-layer view of how AI workloads are designed, represented, and executed on real systems. I have published at ICCV and ICLR and hold multiple patents in the US and Korea.

Employment

- Samsung Electronics, Korea

• SOC IP Development Team (S.LSI)
• Sept. 2024 ~ Current

- Seoul National University (SNU), Seoul, Korea

• Postdoctoral Researcher (Advisor - Prof. Saewoong Bahk)
• Sept. 2023 ~ Aug. 2024

---

Focus Areas

• AI accelerator performance architecture
  NPU architecture, workload characterization, performance modeling, and bottleneck analysis for production AI systems.

• HW/SW co-design for AI inference
  Compiler-aware scheduling, memory/dataflow optimization, and cross-stack analysis from LLM and vision model behavior to accelerator execution.

• AI systems for 3D perception and autonomous driving
  Semi-supervised 3D object detection, model inversion on 3D feature representations, video analytics, and V2X communication.

---

Education

- Ph.D. in ECE from SNU in Aug. 2023

• Advisor
  • Saewoong Bahk
  • Sunghyun Choi, Advisor before moving to NETLAB
    (Currently Corporate EVP and Head of Advanced R&D Team, Samsung Networks)

- B.S. in EE from POSTECH in Feb. 2016

---

Current Collaborators

• Jeongjun Park (Ph.D., Samsung Electronics [CP S/W Development])
• Jin Mo Yang (Ph.D. Candidate, SNU)
• Youngseok Kim (Ph.D. Candidate, SNU)

---

Journals & Conferences

• [IEEE Access 2026] ALERT Open Dataset and Input-Size-Agnostic Vision Transformer for Driver Activity Recognition using IR-UWB [Paper] [arXiv]
  Jeongjun Park, ^Sunwook Hwang, Hyeonho Noh, Jin Mo Yang, Hyun Jong Yang, and ^Saewoong Bahk (^Corresponding authors)

• [ICLR 2025] ConcreTizer: Model Inversion Attack via Occupancy Classification and Dispersion Control for 3D Point Cloud Restoration [Paper] [arXiv]
  *Youngseok Kim, *Sunwook Hwang, Hyung-Sin Kim, and Saewoong Bahk (*Equal Contribution)

• [IEEE Access 2025] FrameBoost: Advanced Video Analytics with Inference Trigger Frame Selection via Tracking Error Estimation [Paper]
  Jin Mo Yang, ^Sunwook Hwang, Jeongjun Park, and ^Saewoong Bahk (^Corresponding authors)

• [IEEE/CVF ICCV 2023] UpCycling: Semi-supervised 3D Object Detection without Sharing Raw-level Unlabeled Scenes [Paper]
  Sunwook Hwang, Youngseok Kim, Seongwon Kim, Saewoong Bahk, and Hyung-Sin Kim

• [IEEE TVT 2020] Beyond Vision: Hidden Car Detector with On-demand Relaying in Vehicular Communications [Paper]
  Sunwook Hwang, Seongwon Kim, Hoyoung Yoon, Byungjun Kim, Sunghyun Choi, and Saewoong Bahk

• [IEEE VTS APWCS 2022] A Study on the Feature-level Perception Sharing of Autonomous Vehicles
  Youngseok Kim, Sunwook Hwang, Saewoong Bahk

• [IEEE WCNC 2021] PRESS: Predictive Assessment of Resource Usage for C-V2V Mode 4 [Paper]
  Jin Mo Yang, Hoyoung Yoon, Sunwook Hwang, and Saewoong Bahk

• [IEEE Access 2019] Nearest-First: Efficient Relaying Scheme in Heterogeneous V2V Communication Environments
  Byungjun Kim, Seongwon Kim, Hoyoung Yoon, Sunwook Hwang, M. Xavier Punithan, Byeong Rim Jo, and Sunghyun Choi

• [IEEE DySPAN 2017] COTA: Channel Occupancy Time Adaptation for LTE in Unlicensed Spectrum
  Kangjin Yoon, Taejun Park, Jihoon Kim, Weiping Sun, Sunwook Hwang, Ingab Kang, and Sunghyun Choi

• [IEEE ICTC 2017] Channel Switching Operation of LTE-LAA in Unlicensed Spectrum
  Sunwook Hwang, Kangjin Yoon, and Sunghyun Choi

---

Intellectual Properties

- Patents

• Method for transmission resource allocation based on transmission resource prediction via V2V communication and apparatus for performing the method
  Jin Mo Yang, Saewoong Bahk, Hoyoung Yoon, Sunwook Hwang
  Korean Patent 10-2698048-0000, Aug. 2024.

• Semi-supervised learning method for object detection in autonomous vehicle and server for performing semi-supervised learning for object detection in autonomous vehicle
  Sunwook Hwang, Youngseok Kim, Saewoong Bahk, and Hyung-sin Kim
  US 11,495,012, Nov. 2022.
  Korean Patent 10-23-402, Apr. 2022.

• Apparatus and method for using multiple carriers in wireless communication system
  Byounghoon Jung, Jihoon Kim, Sunghyun Choi, Seung-Hoon Park, Jungsoo Jung, Taejun Park, Kangjin Yoon, Jaehong Yi, Sunwook Hwang
  US 11,330,585 B2, May 2022.

• Method and apparatus for communication between vehicles and apparatus for using the same
  Sunwook Hwang, Seongwon Kim, Hoyoung Yoon, Byungjun Kim, Sunghyun Choi, and Saewoong Bahk
  US 11,032,682 B2, June, 2021.
  Korean Patent 10-1975759, Apr. 2019.

• Method, apparatus and computer readable record media for collision-aware link adaptation through clustering
  Kangjin Yoon, Sunwook Hwang, and Sunghyun Choi
  Korean Patent 10-2099376, Apr. 2020.

• Method and apparatus for message relaying
  Seungil Park, Sunwook Hwang, Hoyoung Yoon, Byungjun Kim, and Sunghyun Choi
  PCT/KR2019/008328, July 2019.
  Korean Patent 10-1935230, Dec. 2018.

• Method, apparatus and computer readable record media for sharing radio resource on unlicensed band
  Kangjin Yoon, Sunwook Hwang, Taejun Park, Jihoon Kim, and Sunghyun Choi
  Korean Patent 10-1865390, May 2018.

• Apparatus and method for operating a plurality of carriers in wireless communication system
  Byounghoon Jung, Jihoon Kim, Sunghyun Choi, Seunghoon Park, Jungsoo Jung, Jaehong Yi, Kangjin Yoon, and Sunwook Hwang
  Korean Patents Application 10-2017-0111389, filed Aug. 2017, Patent Pending.

- Software Intellectual Properties

• LAA Channel Occupancy Time Adaptation Algorithm for fair coexistence with WLAN
  Kangjin Yoon, Sunwook Hwang, Taejun Park, Jihoon Kim, and Sunghyun Choi
  Korea Copyright Commission C-2017-024231, Oct. 2017.

• WLAN Saturation Detection Algorithm
  Kangjin Yoon, Sunwook Hwang, Taejun Park, Jihoon Kim, and Sunghyun Choi
  Korea Copyright Commission C-2017-000564, Jan. 2017.

---

Research Projects

• Research on distributed learning and extended-vision based 3D object detection model for autonomous driving in 5G networks
  National Research Foundation grant funded by the Korea government (MSIT), 2020-2023.

• Research and Education for Integrated Technology of 5G/Autonomous Vehicles
  5G Unmanned Vehicle Research Center funded by the Institute for Information & Communications Technology Promotion, 2018-2019.

• Dual Interface Synchronized Hybrid V2X Research by Simulation
  LG Electronics, 2018.

• Spectrum Sensing and Future Radio Communication Platforms
  Institute for Information & communications Technology Promotion grant funded by the Ministry of Science, ICT and Future Planning (MSIP) of the Republic of Korea, 2016–2017.

---

Professional Activities

- Reviewer

• International Conferences
  • The 18th European Conference on Computer Vision (ECCV) 2026
  • IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026
  • IEEE Wireless Communications and Networking Conference (WCNC) 2020
  • IEEE Wireless Communications and Networking Conference (WCNC) 2019
  • IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM) 2019
• International Journal
  • IEEE Transactions on Vehicular Technology (TVT)
  • IEEE Transactions on Cognitive Communications and Networking (TCCN)

---

Tutorials

• Tutorial on “Wired LAN in ns-3” at 2019 KICS ns-3 lecture, Feb. 2019.
• Tutorial on “Low-Level Tracing and TCP/UDP/IP in ns-3” at 2019 KICS ns-3 lecture, Feb. 2019.
• Tutorial on “Low-Level Tracing and TCP/UDP/IP in ns-3” at 2018 KICS ns-3 lecture, Feb. 2018.