Sang-Woo Jun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Interests
Sang-Woo Jun has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Jeffrey Chen
Department of Computer Science, University of California, Irvine, Orange, CA, United States
3 shared publications - Jonathan Ta
University of California, Irvine, Irvine, California, USA
2 shared publications - Se-Min Lim
Department of Computer Science University of California Irvine Irvine CA USA
2 shared publications - Aditi Mundra
University of California, Berkeley, Berkeley, CA, United States
1 shared publication - Alberto Krone Martins
Donald Bren School of Information and Computer Sciences University of California 92697 Irvine CA USA
1 shared publication - Esmerald Aliaj
University of California, Irvine, Irvine, California, USA
1 shared publication
Latest Publications
- Open Access
- 29 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 192-199
MAPLE: Flexible-Precision Processing-In-Memory Architecture for Efficient On-Device ML
- 29 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 192-199
Open AccessMAPLE: Flexible-Precision Processing-In-Memory Architecture for Efficient On-Device ML
- Open Access
Bancroft: Genomics Acceleration Beyond On-Device Memory
This paper presents Bancroft, a computational genomics acceleration platform that provides the illusion of practically infinite on-device memory capacity by compressing genomic data movement over PCIe. Bancroft introduces novel optimizations for efficient accelerator implementation to reference-based genome compression, including fixed-stride matching using cuckoo hashes and grouped header encoding, incorporated into a familiar interface supporting random accesses. We evaluate a prototype implementation of Bancroft on an affordable Alveo U50 FPGA equipped with 8 GB of HBM. Thanks to the orders of magnitude improvements in performance and resource efficiency of genomic compression, our prototype provides access to TBs of host-side genomic data at memory-class performance, measuring speeds over 30% of the on-device HBM bandwidth, an order of magnitude higher than conventional PCIe-limited architectures. Using a real-world pre-alignment filtering application, Bancroft demonstrates over 6x improvement over the conventional PCIe-attached architecture, achieving 30% of peak internal throughput of an accelerator with HBM, and 90% of the one with DDR4. Bancroft supports memory-class performance to practically infinite data capacity, using a small, fixed amount of HBM, making it an attractive solution to continued future scalability of computational genomics.
Open AccessBancroft: Genomics Acceleration Beyond On-Device Memory
This paper presents Bancroft, a computational genomics acceleration platform that provides the illusion of practically infinite on-device memory capacity by compressing genomic data movement over PCIe. Bancroft introduces novel optimizations for efficient accelerator implementation to reference-based genome compression, including fixed-stride matching using cuckoo hashes and grouped header encoding, incorporated into a familiar interface supporting random accesses. We evaluate a prototype implementation of Bancroft on an affordable Alveo U50 FPGA equipped with 8 GB of HBM. Thanks to the orders of magnitude improvements in performance and resource efficiency of genomic compression, our prototype provides access to TBs of host-side genomic data at memory-class performance, measuring speeds over 30% of the on-device HBM bandwidth, an order of magnitude higher than conventional PCIe-limited architectures. Using a real-world pre-alignment filtering application, Bancroft demonstrates over 6x improvement over the conventional PCIe-attached architecture, achieving 30% of peak internal throughput of an accelerator with HBM, and 90% of the one with DDR4. Bancroft supports memory-class performance to practically infinite data capacity, using a small, fixed amount of HBM, making it an attractive solution to continued future scalability of computational genomics.
- Open Access
- 4 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 102-107
BMoss: Reconfigurable Hardware Accelerator for Scalable Plagiarism Detection
Accurately and efficiently detecting plagiarism is a critical component of high-quality education. Many plagiarism detection algorithms are being invented and deployed, but these services often come at a significant cost due to their computational overhead. To minimize the cost of plagiarism detection, this work presents BMoss, an efficient reconfigurable hardware accelerator for detecting plagiarism of text or code. It implements the popular MOSS (Measure Of Software Similarity) algorithm, whose performance is typically bottlenecked by repeatedly calculating and comparing hashes of fixed-size substrings. We develop a configurable, pipelined architecture for both has calculation and the Winnowing algorithm, which are the core components of MOSS. We evaluate BMoss against multiple software implementations of MOSS, including our own optimized software implementation which uses a fast xxHash function to outperform all open implementations. We show that using an affordable FPGA-based Alveo U50 accelerator device, we improve the performance and power efficiency of MOSS by over an order of magnitude compared to single-thread implementations. Compared to a much costlier Xeon server, BMoss achieves almost 2× performance and over 3× power efficiency.
- 4 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 102-107
Open AccessBMoss: Reconfigurable Hardware Accelerator for Scalable Plagiarism Detection
Accurately and efficiently detecting plagiarism is a critical component of high-quality education. Many plagiarism detection algorithms are being invented and deployed, but these services often come at a significant cost due to their computational overhead. To minimize the cost of plagiarism detection, this work presents BMoss, an efficient reconfigurable hardware accelerator for detecting plagiarism of text or code. It implements the popular MOSS (Measure Of Software Similarity) algorithm, whose performance is typically bottlenecked by repeatedly calculating and comparing hashes of fixed-size substrings. We develop a configurable, pipelined architecture for both has calculation and the Winnowing algorithm, which are the core components of MOSS. We evaluate BMoss against multiple software implementations of MOSS, including our own optimized software implementation which uses a fast xxHash function to outperform all open implementations. We show that using an affordable FPGA-based Alveo U50 accelerator device, we improve the performance and power efficiency of MOSS by over an order of magnitude compared to single-thread implementations. Compared to a much costlier Xeon server, BMoss achieves almost 2× performance and over 3× power efficiency.
- Open Access
- 5 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Xyloni: Very Low Power Neural Network Accelerator for Intermittent Remote Visual Detection of Wildfire and Beyond
Wildfires are one of the most catastrophic natural disasters, causing increasingly severe ecological and economic damage. Early response is critically important for wildfire management, but also difficult due to the wide geographical area to monitor, often far from utility infrastructures such as stable power and high-bandwidth network. In this work, we present Xyloni, a very low-cost, low-power neural network accelerator for sensor nodes, which improves the cost-effectiveness and scalability of real-time wildfire detection by drastically reducing wireless data transmission and overall power consumption. Xyloni uses low-power flash and FeRAM memories to store a hardware co-optimized Neural Network model for fire and smoke detection, as well as intermediate activations during inference. It also time-shares a Field-Programmable Gate Array across different model layers for power-efficient computation. The detection model prevents benign images from consuming network traffic, allowing the use of low-bandwidth, low-power network fabrics such as a LoRa mesh network with enough range for the necessary geographical coverage. Compared to a wide range of edge and sensor platforms capable of real-time data collection, Xyloni demonstrated an order of magnitude reduction in power consumption for the network transmission reduction task, leading to a corresponding reduction in battery and deployment cost.
- 5 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessXyloni: Very Low Power Neural Network Accelerator for Intermittent Remote Visual Detection of Wildfire and Beyond
Wildfires are one of the most catastrophic natural disasters, causing increasingly severe ecological and economic damage. Early response is critically important for wildfire management, but also difficult due to the wide geographical area to monitor, often far from utility infrastructures such as stable power and high-bandwidth network. In this work, we present Xyloni, a very low-cost, low-power neural network accelerator for sensor nodes, which improves the cost-effectiveness and scalability of real-time wildfire detection by drastically reducing wireless data transmission and overall power consumption. Xyloni uses low-power flash and FeRAM memories to store a hardware co-optimized Neural Network model for fire and smoke detection, as well as intermediate activations during inference. It also time-shares a Field-Programmable Gate Array across different model layers for power-efficient computation. The detection model prevents benign images from consuming network traffic, allowing the use of low-bandwidth, low-power network fabrics such as a LoRa mesh network with enough range for the necessary geographical coverage. Compared to a wide range of edge and sensor platforms capable of real-time data collection, Xyloni demonstrated an order of magnitude reduction in power consumption for the network transmission reduction task, leading to a corresponding reduction in battery and deployment cost.
- Open Access
- 23 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Sting: Near-storage accelerator framework for scalable triangle counting and beyond
One of the most critical limitations to scalable graph mining is memory capacity, as graphs of interest continue to grow while the rate of DRAM scaling diminishes. While high-performance NVMe storage is cheap and dense enough to better support larger graphs, the relative performance limitations of secondary storage force a cost-performance trade-off. We present STING, which uses an asynchronous callback function to provide a general interface to in-storage graphs while allowing transparent near-storage acceleration. Using triangle counting, we show with transparent filtering and sorting acceleration, STING can improve state-of-the-art by 3x for cost and power efficiency.
- 23 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessSting: Near-storage accelerator framework for scalable triangle counting and beyond
One of the most critical limitations to scalable graph mining is memory capacity, as graphs of interest continue to grow while the rate of DRAM scaling diminishes. While high-performance NVMe storage is cheap and dense enough to better support larger graphs, the relative performance limitations of secondary storage force a cost-performance trade-off. We present STING, which uses an asynchronous callback function to provide a general interface to in-storage graphs while allowing transparent near-storage acceleration. Using triangle counting, we show with transparent filtering and sorting acceleration, STING can improve state-of-the-art by 3x for cost and power efficiency.
- Open Access
- 12 February 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Reconfigurable Technology and Systems
- Vol. 17 (1) , 1-25
Eciton: Very Low-power Recurrent Neural Network Accelerator for Real-time Inference at the Edge
This article presents Eciton, a very low-power recurrent neural network accelerator for time series data within low-power edge sensor nodes, achieving real-time inference with a power consumption of 17 mW under load. Eciton reduces memory and chip resource requirements via 8-bit quantization and hard sigmoid activation, allowing the accelerator as well as the recurrent neural network model parameters to fit in a low-cost, low-power Lattice iCE40 UP5K FPGA. We evaluate Eciton on multiple, established time-series classification applications including predictive maintenance of mechanical systems, sound classification, and intrusion detection for IoT nodes. Binary and multi-class classification edge models are explored, demonstrating that Eciton can adapt to a variety of deployable environments and remote use cases. Eciton demonstrates real-time processing at a very low power consumption with minimal loss of accuracy on multiple inference scenarios with differing characteristics, while achieving competitive power efficiency against the state-of-the-art of similar scale. We show that the addition of this accelerator actually reduces the power budget of the sensor node by reducing power-hungry wireless transmission. The resulting power budget of the sensor node is small enough to be powered by a power harvester, potentially allowing it to run indefinitely without a battery or periodic maintenance.
- 12 February 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Reconfigurable Technology and Systems
- Vol. 17 (1) , 1-25
Open AccessEciton: Very Low-power Recurrent Neural Network Accelerator for Real-time Inference at the Edge
This article presents Eciton, a very low-power recurrent neural network accelerator for time series data within low-power edge sensor nodes, achieving real-time inference with a power consumption of 17 mW under load. Eciton reduces memory and chip resource requirements via 8-bit quantization and hard sigmoid activation, allowing the accelerator as well as the recurrent neural network model parameters to fit in a low-cost, low-power Lattice iCE40 UP5K FPGA. We evaluate Eciton on multiple, established time-series classification applications including predictive maintenance of mechanical systems, sound classification, and intrusion detection for IoT nodes. Binary and multi-class classification edge models are explored, demonstrating that Eciton can adapt to a variety of deployable environments and remote use cases. Eciton demonstrates real-time processing at a very low power consumption with minimal loss of accuracy on multiple inference scenarios with differing characteristics, while achieving competitive power efficiency against the state-of-the-art of similar scale. We show that the addition of this accelerator actually reduces the power budget of the sensor node by reducing power-hungry wireless transmission. The resulting power budget of the sensor node is small enough to be powered by a power harvester, potentially allowing it to run indefinitely without a battery or periodic maintenance.
- 7 November 2023
- Published by Springer Nature
- p. 57-71
Myrmec: FPGA-Accelerated SmartNIC for Cost and Power Efficient IoT Sensor Networks
- 7 November 2023
- Published by Springer Nature
- p. 57-71
Myrmec: FPGA-Accelerated SmartNIC for Cost and Power Efficient IoT Sensor Networks
- Open Access
- 9 March 2022
- Published by MDPI AG in Electronics
- Vol. 11 (6) , 858
MobileNets Can Be Lossily Compressed: Neural Network Compression for Embedded Accelerators
Although neural network quantization is an imperative technology for the computation and memory efficiency of embedded neural network accelerators, simple post-training quantization incurs unacceptable levels of accuracy degradation on some important models targeting embedded systems, such as MobileNets. While explicit quantization-aware training or re-training after quantization can often reclaim lost accuracy, this is not always possible or convenient. We present an alternative approach to compressing such difficult neural networks, using a novel variant of the ZFP lossy floating-point compression algorithm to compress both model weights and inter-layer activations and demonstrate that it can be efficiently implemented on an embedded FPGA platform. Our ZFP variant, which we call ZFPe, is designed for efficient implementation on embedded accelerators, such as FPGAs, requiring a fraction of chip resources per bandwidth compared to state-of-the-art lossy compression accelerators. ZFPe-compressing the MobileNet V2 model with an 8-bit budget per weight and activation results in significantly higher accuracy compared to 8-bit integer post-training quantization and shows no loss of accuracy, compared to an uncompressed model when given a 12-bit budget per floating-point value. To demonstrate the benefits of our approach, we implement an embedded neural network accelerator on a realistic embedded acceleration platform equipped with the low-power Lattice ECP5-85F FPGA and a 32 MB SDRAM chip. Each ZFPe module consumes less than 6% of LUTs while compressing or decompressing one value per cycle, requiring a fraction of the resources compared to state-of-the-art compression accelerators while completely removing the memory bottleneck of our accelerator.
- 9 March 2022
- Published by MDPI AG in Electronics
- Vol. 11 (6) , 858
Open AccessMobileNets Can Be Lossily Compressed: Neural Network Compression for Embedded Accelerators
Although neural network quantization is an imperative technology for the computation and memory efficiency of embedded neural network accelerators, simple post-training quantization incurs unacceptable levels of accuracy degradation on some important models targeting embedded systems, such as MobileNets. While explicit quantization-aware training or re-training after quantization can often reclaim lost accuracy, this is not always possible or convenient. We present an alternative approach to compressing such difficult neural networks, using a novel variant of the ZFP lossy floating-point compression algorithm to compress both model weights and inter-layer activations and demonstrate that it can be efficiently implemented on an embedded FPGA platform. Our ZFP variant, which we call ZFPe, is designed for efficient implementation on embedded accelerators, such as FPGAs, requiring a fraction of chip resources per bandwidth compared to state-of-the-art lossy compression accelerators. ZFPe-compressing the MobileNet V2 model with an 8-bit budget per weight and activation results in significantly higher accuracy compared to 8-bit integer post-training quantization and shows no loss of accuracy, compared to an uncompressed model when given a 12-bit budget per floating-point value. To demonstrate the benefits of our approach, we implement an embedded neural network accelerator on a realistic embedded acceleration platform equipped with the low-power Lattice ECP5-85F FPGA and a 32 MB SDRAM chip. Each ZFPe module consumes less than 6% of LUTs while compressing or decompressing one value per cycle, requiring a fraction of the resources compared to state-of-the-art compression accelerators while completely removing the memory bottleneck of our accelerator.
- 1 December 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 129-138
Bandwidth Efficient Near-Storage Accelerator for High-Dimensional Similarity Search
Performance of high-dimensional similarity search systems is often determined by the rate of distance calculation between a query and candidate neighbors. This is because as the dimensionality of data becomes higher, it suffers the so-called “curse of dimensionality”, which makes conventional indexing methods inefficient. Since the sheer size of datasets of interest often mandate storing them in secondary storage, this means the performance of a similarity search may be primarily limited by the storage device bandwidth, especially if the distance calculation can be offloaded to a high-performance accelerator built with GPUs or FPGAs. In this paper, we present ZipNN, which moves the bottleneck away from the storage performance in order to make more effective use of available computation resources. We achieve this via two approaches: First, ZipNN is a near-storage accelerator which takes advantage of the high internal bandwidth of storage. Second, it implements multiple pipelines of an application-optimized, wire-speed compression algorithm which can handle full storage bandwidth. We evaluate ZipNN using real-world datasets and an FPGA-based implementation. Compared to a comparable FPGA accelerator deployed in a standalone device, it demonstrates over 6× performance improvements. It also delivers an order of magnitude improvement compared to a purely software implementation.
- 1 December 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 129-138
Bandwidth Efficient Near-Storage Accelerator for High-Dimensional Similarity Search
Performance of high-dimensional similarity search systems is often determined by the rate of distance calculation between a query and candidate neighbors. This is because as the dimensionality of data becomes higher, it suffers the so-called “curse of dimensionality”, which makes conventional indexing methods inefficient. Since the sheer size of datasets of interest often mandate storing them in secondary storage, this means the performance of a similarity search may be primarily limited by the storage device bandwidth, especially if the distance calculation can be offloaded to a high-performance accelerator built with GPUs or FPGAs. In this paper, we present ZipNN, which moves the bottleneck away from the storage performance in order to make more effective use of available computation resources. We achieve this via two approaches: First, ZipNN is a near-storage accelerator which takes advantage of the high internal bandwidth of storage. Second, it implements multiple pipelines of an application-optimized, wire-speed compression algorithm which can handle full storage bandwidth. We evaluate ZipNN using real-world datasets and an FPGA-based implementation. Compared to a comparable FPGA accelerator deployed in a standalone device, it demonstrates over 6× performance improvements. It also delivers an order of magnitude improvement compared to a purely software implementation.