PROJECT
[2023] MetaVRain: A Low-power and Real-time Neural Rendering Processor and 3D Style Transfer System
Demo Video: [link] / News: [link] [link] [link] [link] [link] / Interview: [link] / Dataset: [link]
- Related paper: A 133mW Real-time Hyper-realistic-3D-NeRF Processor with 1D-2D Hybrid-Neural-Engines for Metaverse on Mobile Devices (ISSCC 2023)
- Donghyeon Han's Contribution
1) The Leader of the Project
2) Development of Bundle-Frame-Familiarity (BuFF) Architecture
3) Development of Temporal Familiarity based NeRF Acceleration Architecture
4) Architecture Simulator Design of 1D-2D Hybrid Neural Engine
5) Digital Circuit Design of 1D-2D Hybrid Neural Engine
6) Development of Centrifugal Sampling based Dynamic Neural Network Allocation Method
7) Development of Periodic Polynomial Approximation for Sinusoidal Function Approximation
8) Digital Circuit Design of Modulo-based Positional Encoding Unit
9) Back-end Design
10) MetaVRain Embedded Board Design
11) MetaVRain Demonstration Sysmte Design
[2022] HNPU-V1 & HNPU-V2: Fixed-point based Mobile Deep Neural Network Training System [AICAS Best Demonstration & Best Paper]
Demo Video: [link] / News: [link] [link] [link]
- Related paper: HNPU-V1: HNPU: An Adaptive DNN Training Processor Utilizing Stochastic Dynamic Fixed-point and Active Bit-precision Searching (JSSC 2021)
- Related paper: HNPU-V2: A 0.95 mJ/frame DNN Training Processor for Robust Object Detection with Real-World Environmental Adaptation (AICAS 2022)
- Donghyeon Han's Contribution
1) The Leader of the Project
2) Development of Layer-wise Adaptive Precision Scaling Algorithm & Related Hardware
3) Design of Fixed-point Representation based Reconfigurable Accumulation Network
4) Back-end Design
5) HNPU Embedded Board Design
6) HNPU Demonstration System Design
[2021] OmniDRL: Low-power Deep Reinforcement Learning (DRL) Processor for Mobile DRL Simulator
Demo Video: [link] / News: [link] [link]
- Related paper: A 29.3 TFLOPS/W Deep Reinforcement Learning Processor with Dual-mode Weight Compression and On-chip Sparse Weight Transposer (S. VLSI 2021)
- Donghyeon Han's Contribution
1) Pytorch based Fine-grained Bit-precision Simulator Software Design
2) 28-nm Place-and-route (PnR) of the Chip
[2020] GANPU: Energy-efficient Generative Adversarial Network (GAN) Training Processor for Mobile User-adaptation System
Demo Video: [link] / News: [link] [link]
- Related paper: 7.4 GANPU: A 135TFLOPS/W Multi-DNN Training Processor for GANs with Speculative Dual-Sparsity Exploitation (ISSCC 2020)
- Donghyeon Han's Contribution
1) Exponent based Output Prediction Algorithm Development
2) Low-area & Low-power Output Prediction IP Design
3) Design of Pipeline Structure between Main Convolution Core and Output Prediction IP
[2019] DF-LNPU: World's 1st Real-time Online Deep Neural Network Training Processor for Object Tracking in Mobile Devices
News: [link] [link] / Patent: [link]
- Related paper:
1) A 1.32 TOPS/W Energy Efficient Deep Neural Network Learning Processor with Direct Feedback Alignment based Heterogeneous Core Architecture (S. VLSI 2019)
2) DF-LNPU: A Pipelined Direct Feedback Alignment-Based Deep Neural Network Learning Processor for Fast Online Learning (JSSC 2021)
- Donghyeon Han's Contribution
1) The Leader of the Project
2) Development of New DNN Training Algorithm: Pipelined Direct Feedback Alignment (PDFA)
3) Front-end Design (Verilog) of PDFA based Heterogeneous DNN Training Cores
4) Back-end Design (PnR) of DF-LNPU
[2019] LNPU: World's 1st Deep Neural Network On-device Training Processor for Mobile Devices [ISSCC Best Demonstration]
Demo Video: [link] / News: [link]
- Related paper: LNPU: A 25.3TFLOPS/W Sparse Deep-Neural-Network Learning Processor with Fine-Grained Mixed Precision of FP8-FP16 (ISSCC 2019)
- Donghyeon Han's Contribution
1) 65-nm Place-and-route (PnR) of the Chip
2) Front-end Design (Verilog) of Batch-normalization Unit (IncludingBack-propagation Functionality)
[2018] Mobile 3D Hand Gesture Recognition (HGR) System with CNN-stereo based Depth-estimation and ICP-PSO based Hand Tracking [ISSCC Best Demonstration]
Demo Video: [link]
- Related paper: A 9.02mW CNN-Stereo based Real-time 3D Hand Gesture Recognition Processor for Smart Mobile Devices (ISSCC 2018)
- Donghyeon Han's Contribution
1) 3D Hand Model Extraction
2) ICP-PSO based Hand Tracking Algorithm Development
3) 3D Hand Sphere Model Design (UI)
[2017] K-eye Ultra-low-power Mobile Face Recognition System
Demo Video: [link] [link] / News: [link]
- Related paper: A Low-Power Convolutional Neural Network Face Recognition Processor and a CIS Integrated With Always-on Face Detector (JSSC 2017)
- Donghyeon Han's Contribution
1) Haar-like based Face Detection FPGA IP Development
2) Bluetooth FPGA IP Development for Wireless Face Detection Result Transmission
3) C# based Face Recognition UI Development