Jinkwon Kim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jinkwon Kim has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Soontae Kim
School of Computing, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea
5 shared publications - Myeongjae Jang
Korea Advanced Institute of Science & Technology (KAIST), Republic of Korea
3 shared publications - Haejin Nam
Korea Advanced Institute of Science and Technology (KAIST)
2 shared publications - Seokin Hong
Sungkyunkwan University
2 shared publications - Derrick Quinn
Cornell University, Ithaca, NY, USA
1 shared publication - E. Ezgi Yücel
Cornell University, Ithaca, USA
1 shared publication
Latest Publications
- Open Access
- 17 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 34-48
LongSight: Compute-Enabled Memory to Accelerate Large-Context LLMs via Sparse Attention
- 17 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 34-48
Open AccessLongSight: Compute-Enabled Memory to Accelerate Large-Context LLMs via Sparse Attention
- Open Access
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1746-1759
Avalanche: Optimizing Cache Utilization via Matrix Reordering for Sparse Matrix Multiplication Accelerator
- 20 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1746-1759
Open AccessAvalanche: Optimizing Cache Utilization via Matrix Reordering for Sparse Matrix Multiplication Accelerator
- Open Access
- 28 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1148-1162
HARP: Hardware-Based Pseudo-Tiling for Sparse Matrix Multiplication Accelerator
General sparse matrix-matrix multiplication (SpGEMM) is a memory-bound workload, due to the compression format used. To minimize data movements for input matrices, outer product accelerators have been proposed. Since these accelerators access input matrices only once and then generate numerous partial products, managing the generated partial products is the key optimization factor. To reduce the number of partial products handled, the state-of-the-art accelerator uses software to tile an input matrix. However, the software-based tiling has three limitations. First, a user manually executes the tiling software and manages the tiles. Second, generating a compression format for each tile incurs memory-intensive operations. Third, an accelerator that uses the compression format cannot skip ineffectual accesses for input matrices. To overcome these limitations, this paper proposes hardware-based pseudo-tiling (HARP), which enables logical tiling of the original compressed matrix without generating a compression format for each tile. To this end, HARP utilizes our proposed Runtime Operand Descriptor to point to an effectual column-row pair in a pseudo-tile. Consequently, HARP enables a user to use the accelerator as a normal SpGEMM operation does without tiling. Furthermore, HARP does not require a compression format for each tile and can skip ineffectual accesses for input matrices. To further improve the efficiency of pseudo-tiling, HARP performs super-tiling to combine pseudo-tiles and sub-tiling to further tile a pseudo-tile. Experiment results show that HARP achieves 4 ×, 35 ×, and 33 × speedup and 5 ×, 664 ×, and 233 × energy efficiency on average compared to the state-of-the-art outer product accelerator, CPU (MKL), and GPU (cuSPARSE), respectively.
- 28 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1148-1162
Open AccessHARP: Hardware-Based Pseudo-Tiling for Sparse Matrix Multiplication Accelerator
General sparse matrix-matrix multiplication (SpGEMM) is a memory-bound workload, due to the compression format used. To minimize data movements for input matrices, outer product accelerators have been proposed. Since these accelerators access input matrices only once and then generate numerous partial products, managing the generated partial products is the key optimization factor. To reduce the number of partial products handled, the state-of-the-art accelerator uses software to tile an input matrix. However, the software-based tiling has three limitations. First, a user manually executes the tiling software and manages the tiles. Second, generating a compression format for each tile incurs memory-intensive operations. Third, an accelerator that uses the compression format cannot skip ineffectual accesses for input matrices. To overcome these limitations, this paper proposes hardware-based pseudo-tiling (HARP), which enables logical tiling of the original compressed matrix without generating a compression format for each tile. To this end, HARP utilizes our proposed Runtime Operand Descriptor to point to an effectual column-row pair in a pseudo-tile. Consequently, HARP enables a user to use the accelerator as a normal SpGEMM operation does without tiling. Furthermore, HARP does not require a compression format for each tile and can skip ineffectual accesses for input matrices. To further improve the efficiency of pseudo-tiling, HARP performs super-tiling to combine pseudo-tiles and sub-tiling to further tile a pseudo-tile. Experiment results show that HARP achieves 4 ×, 35 ×, and 33 × speedup and 5 ×, 664 ×, and 233 × energy efficiency on average compared to the state-of-the-art outer product accelerator, CPU (MKL), and GPU (cuSPARSE), respectively.
- 14 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 73 (1) , 249-262
Zero and Narrow-Width Value-Aware Compression for Quantized Convolutional Neural Networks
Convolutional neural networks are normally used in systems with dedicated neural processing units for CNN-related computations. For high performance and low hardware overheads, CNN datatype quantization is applied. As an additional optimization, to further reduce DRAM accesses, compression algorithms have been used for CNN data. However, conventional zero value-aware compression algorithms suffer from a reduction in compression ratio with the latest quantized CNNs, owing to the small number of zero values. Moreover, the appropriate zero run-length code width can be changed dynamically based on the CNNs, layers, and quantization datatypes. As another compressible data value for increasing the compression ratio, the latest quantized CNNs have many narrow-width values. Because low-precision quantization reduces the data bit width, CNN data are gathered into a few discrete values and incur a biased data distribution. These discrete values become narrow-width values, and constitute a large proportion of the biased distribution. In this article, we propose an efficient compression algorithm for quantized CNNs, ENCORE, which utilizes variable zero run-length encoding and compresses narrow-width values. With the latest quantized CNNs, ENCORE shows higher compression ratios, 93.55% and 50.85% in Mobilenet v1 and Tiny YOLO v3, respectively, than conventional zero value-aware CNN data compression algorithms.
- 14 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 73 (1) , 249-262
Zero and Narrow-Width Value-Aware Compression for Quantized Convolutional Neural Networks
Convolutional neural networks are normally used in systems with dedicated neural processing units for CNN-related computations. For high performance and low hardware overheads, CNN datatype quantization is applied. As an additional optimization, to further reduce DRAM accesses, compression algorithms have been used for CNN data. However, conventional zero value-aware compression algorithms suffer from a reduction in compression ratio with the latest quantized CNNs, owing to the small number of zero values. Moreover, the appropriate zero run-length code width can be changed dynamically based on the CNNs, layers, and quantization datatypes. As another compressible data value for increasing the compression ratio, the latest quantized CNNs have many narrow-width values. Because low-precision quantization reduces the data bit width, CNN data are gathered into a few discrete values and incur a biased data distribution. These discrete values become narrow-width values, and constitute a large proportion of the biased distribution. In this article, we propose an efficient compression algorithm for quantized CNNs, ENCORE, which utilizes variable zero run-length encoding and compresses narrow-width values. With the latest quantized CNNs, ENCORE shows higher compression ratios, 93.55% and 50.85% in Mobilenet v1 and Tiny YOLO v3, respectively, than conventional zero value-aware CNN data compression algorithms.
- 28 May 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. PP (00189340) , 1
PR-SSD: Maximizing Partial Read Potential By Exploiting Compression and Channel-Level Parallelism
Recent NAND flash memories provide a partial read operation that can read a page partially and has lower latency than a normal read operation. In order to maximize the benefit of the partial read operation, compression techniques can be applied to improve performance by generating additional partial page requests by compressing pages into smaller ones. Unfortunately, existing compression support SSDs suffer from a huge decompression latency that eventually cancels the benefit of the partial read operation. In this paper, we propose Partial Read-aware SSD (PR-SSD) for fully exploiting partial read operations. In order to mitigate the decompression latency, we propose a new compression algorithm, called Dominant Pattern Compression (DPC), which has extremely low decompression latency. Because uncompressed page requests cannot exploit the partial read operation, we propose split Flash Translation Layer (FTL) that can split the requests into smaller ones and allocate them to different channels for exploiting channel-level parallelism in SSD. Experimental results reveal that PR-SSD can reduce the read response time by 18% on average and also the number of writes and write response time by 29% and 24% on average, respectively
- 28 May 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. PP (00189340) , 1
PR-SSD: Maximizing Partial Read Potential By Exploiting Compression and Channel-Level Parallelism
Recent NAND flash memories provide a partial read operation that can read a page partially and has lower latency than a normal read operation. In order to maximize the benefit of the partial read operation, compression techniques can be applied to improve performance by generating additional partial page requests by compressing pages into smaller ones. Unfortunately, existing compression support SSDs suffer from a huge decompression latency that eventually cancels the benefit of the partial read operation. In this paper, we propose Partial Read-aware SSD (PR-SSD) for fully exploiting partial read operations. In order to mitigate the decompression latency, we propose a new compression algorithm, called Dominant Pattern Compression (DPC), which has extremely low decompression latency. Because uncompressed page requests cannot exploit the partial read operation, we propose split Flash Translation Layer (FTL) that can split the requests into smaller ones and allocate them to different channels for exploiting channel-level parallelism in SSD. Experimental results reveal that PR-SSD can reduce the read response time by 18% on average and also the number of writes and write response time by 29% and 24% on average, respectively
- 14 March 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE)
ENCORE Compression: Exploiting Narrow-width Values for Quantized Deep Neural Networks
Deep Neural Networks (DNNs) become a practical machine learning algorithm running on various Neural Processing Units (NPUs). For higher performance and lower hardware overheads, DNN datatype reduction through quantization is proposed. Moreover, to solve the memory bottleneck caused by large data size in DNNs, several zero value-aware compression algorithms are used. However, these compression algorithms do not compress modern quantized DNNs well because of decreased zero values. We find that the latest quantized DNNs have data redundancy due to frequent narrow-width values. Because low-precision quantization reduces DNN datatypes to a simple datatype with less bits, scattered DNN data are gathered to a small number of discrete values and incur a biased data distribution. Narrow-width values occupy a large proportion of the biased distribution. Moreover, an appropriate zero run-length bits can be dynamically changed according to DNN sparsity. Based on this observation, we propose a compression algorithm that exploits narrow-width values and variable zero run-length for quantized DNNs. In experiments with three quantized DNNs, our proposed scheme yields an average compression ratio of 2.99.
- 14 March 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE)
ENCORE Compression: Exploiting Narrow-width Values for Quantized Deep Neural Networks
Deep Neural Networks (DNNs) become a practical machine learning algorithm running on various Neural Processing Units (NPUs). For higher performance and lower hardware overheads, DNN datatype reduction through quantization is proposed. Moreover, to solve the memory bottleneck caused by large data size in DNNs, several zero value-aware compression algorithms are used. However, these compression algorithms do not compress modern quantized DNNs well because of decreased zero values. We find that the latest quantized DNNs have data redundancy due to frequent narrow-width values. Because low-precision quantization reduces DNN datatypes to a simple datatype with less bits, scattered DNN data are gathered to a small number of discrete values and incur a biased data distribution. Narrow-width values occupy a large proportion of the biased distribution. Moreover, an appropriate zero run-length bits can be dynamically changed according to DNN sparsity. Based on this observation, we propose a compression algorithm that exploits narrow-width values and variable zero run-length for quantized DNNs. In experiments with three quantized DNNs, our proposed scheme yields an average compression ratio of 2.99.
- 17 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 70 (7) , 1132-1145
CID: Co-Architecting Instruction Cache and Decompression System for Embedded Systems
Code compression is widely used to reduce the footprint of code memory in cost-sensitive embedded systems. However, despite the small code size, the decompressor and the address translator required to support the code compression incur energy and area overheads. To reduce such overheads while still supporting code compression, we co-architect the instruction cache and decompression system (CID). In CID, each component is placed at the optimal location and the instruction cache is redesigned to recognize the compression state and retain the original address, through the cache division and address space decompression process. As a result of the cache division, the energy consumption and area overheads of the CID instruction cache are reduced. Since the decompressor overhead depends on the code compression technique, we propose a new code compression technique called entropy-based pattern code compression, which reduces overheads of the decompressor. Our experimental results show that the total energy consumption of the instruction cache and decompression system is reduced by up to 29.7% and their area is reduced by up to 15.4% compared to the post-cache architecture with almost no performance degradation, while achieving an 18.8% improvement in the compression ratio compared to the state-of-the-art code compression technique.
- 17 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computers
- Vol. 70 (7) , 1132-1145
CID: Co-Architecting Instruction Cache and Decompression System for Embedded Systems
Code compression is widely used to reduce the footprint of code memory in cost-sensitive embedded systems. However, despite the small code size, the decompressor and the address translator required to support the code compression incur energy and area overheads. To reduce such overheads while still supporting code compression, we co-architect the instruction cache and decompression system (CID). In CID, each component is placed at the optimal location and the instruction cache is redesigned to recognize the compression state and retain the original address, through the cache division and address space decompression process. As a result of the cache division, the energy consumption and area overheads of the CID instruction cache are reduced. Since the decompressor overhead depends on the code compression technique, we propose a new code compression technique called entropy-based pattern code compression, which reduces overheads of the decompressor. Our experimental results show that the total energy consumption of the instruction cache and decompression system is reduced by up to 29.7% and their area is reduced by up to 15.4% compared to the post-cache architecture with almost no performance degradation, while achieving an 18.8% improvement in the compression ratio compared to the state-of-the-art code compression technique.