Yulhwa Kim
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 Subjects
Yulhwa Kim has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Jae-Joon Kim
Seoul National University
5 shared publications - Hyungjun Kim
Department of Creative IT Engineering, Pohang University of Science and Technology, Pohang, South Korea
4 shared publications - Daehyun Ahn
Department of Convergence IT Engineering, Pohang University of Science and Technology, Pohang, South Korea
1 shared publication Eunhwan Kim
Department of Convergence IT Engineering, Pohang University of Science and Technology, Pohang, Republic of Korea
1 shared publication- Hyunmyung Oh
Department of Convergence IT Engineering, Pohang University of Science and Technology, Pohang, South Korea
1 shared publication - Inhwan Lee
Department of Convergence IT Engineering, Pohang University of Science and Technology, Pohang, Republic of Korea
1 shared publication
Latest Publications
- Top Cited Papers
- 2 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 3, 760-773
FIGNA: Integer Unit-Based Accelerator Design for FP-INT GEMM Preserving Numerical Accuracy
The weight-only quantization has emerged as a promising technique for alleviating the computational burden of large language models (LLMs) by employing low-precision integer (INT) weights, while retaining full-precision floating point (FP) activations to ensure inference quality. Despite the memory footprint reduction achieved through decreased bit-precision of weight parameters, the actual computing performance is often not improved significantly due to FP-INT multiply-accumulation (MAC) operations being performed on the floating point unit (FPU) after de quantizing the INT weight values to FP values, owing to the lack of dedicated FP- INT arithmetic units. In this study, we investigate the impact of introducing a dedicated FP-INT unit on overall performance and find that such specialization does not yield substantial improvements. As an alternative approach, we propose FIGNA, an accelerator based on INT units designed specifically for FP- INT MAC operations. A key feature of FIGNA is its ability to achieve the same numerical accuracy as the FPU while relying solely on the integer-unit, a departure from prior methods that relied on integer-units with numerical approximations for FP arithmetic results, albeit claiming similar inference accuracy through dedicated network training. Through comprehensive experiments on FP- INT quantized networks for LLMs, including OPT and BLOOM, we demonstrate the superior performance of FIGNA compared to conventional FPUs in terms of performance per area ( $TOPS/mm^{2}$ ) and energy efficiency (TOPS/W) across various input and weight precision combinations. For instance, in the FP16-INT4 case, FIGNA shows 6.34x higher $TOPS/ mm^{2}$ and 2.19x higher TOPS/W compared to the baseline.- 2 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 3, 760-773
Top Cited PapersFIGNA: Integer Unit-Based Accelerator Design for FP-INT GEMM Preserving Numerical Accuracy
The weight-only quantization has emerged as a promising technique for alleviating the computational burden of large language models (LLMs) by employing low-precision integer (INT) weights, while retaining full-precision floating point (FP) activations to ensure inference quality. Despite the memory footprint reduction achieved through decreased bit-precision of weight parameters, the actual computing performance is often not improved significantly due to FP-INT multiply-accumulation (MAC) operations being performed on the floating point unit (FPU) after de quantizing the INT weight values to FP values, owing to the lack of dedicated FP- INT arithmetic units. In this study, we investigate the impact of introducing a dedicated FP-INT unit on overall performance and find that such specialization does not yield substantial improvements. As an alternative approach, we propose FIGNA, an accelerator based on INT units designed specifically for FP- INT MAC operations. A key feature of FIGNA is its ability to achieve the same numerical accuracy as the FPU while relying solely on the integer-unit, a departure from prior methods that relied on integer-units with numerical approximations for FP arithmetic results, albeit claiming similar inference accuracy through dedicated network training. Through comprehensive experiments on FP- INT quantized networks for LLMs, including OPT and BLOOM, we demonstrate the superior performance of FIGNA compared to conventional FPUs in terms of performance per area ( $TOPS/mm^{2}$ ) and energy efficiency (TOPS/W) across various input and weight precision combinations. For instance, in the FP16-INT4 case, FIGNA shows 6.34x higher $TOPS/ mm^{2}$ and 2.19x higher TOPS/W compared to the baseline. - Open Access
- 13 October 2022
- Published by Association for Computing Machinery (ACM) in ACM Journal on Emerging Technologies in Computing Systems
- Vol. 18 (4) , 1-19
Extreme Partial-Sum Quantization for Analog Computing-In-Memory Neural Network Accelerators
In Analog Computing-in-Memory (CIM) neural network accelerators, analog-to-digital converters (ADCs) are required to convert the analog partial sums generated from a CIM array to digital values. The overhead from ADCs substantially degrades the energy efficiency of CIM accelerators so that previous works attempted to lower the ADC resolution considering the distribution of the partial sums. Despite the efforts, the required ADC resolution still remains relatively high. In this article, we propose the data-driven partial sum quantization scheme, which exhaustively searches for the optimal quantization range with little computational burden. We also report that analyzing the characteristics of the partial sum distributions at each layer gives an additional information to further reduce the ADC resolution compared to previous works that mostly used the characteristics of the partial sum distributions of the entire network. Based on the finer-level data-driven approach combined with retraining, we present a methodology for extreme partial-sum quantization. Experimental results show that the proposed method can reduce the ADC resolution to 2 to 3 bits for CIFAR-10 dataset, which is the smaller ADC bit resolution than any previous CIM-based NN accelerators.
- 13 October 2022
- Published by Association for Computing Machinery (ACM) in ACM Journal on Emerging Technologies in Computing Systems
- Vol. 18 (4) , 1-19
Open AccessExtreme Partial-Sum Quantization for Analog Computing-In-Memory Neural Network Accelerators
In Analog Computing-in-Memory (CIM) neural network accelerators, analog-to-digital converters (ADCs) are required to convert the analog partial sums generated from a CIM array to digital values. The overhead from ADCs substantially degrades the energy efficiency of CIM accelerators so that previous works attempted to lower the ADC resolution considering the distribution of the partial sums. Despite the efforts, the required ADC resolution still remains relatively high. In this article, we propose the data-driven partial sum quantization scheme, which exhaustively searches for the optimal quantization range with little computational burden. We also report that analyzing the characteristics of the partial sum distributions at each layer gives an additional information to further reduce the ADC resolution compared to previous works that mostly used the characteristics of the partial sum distributions of the entire network. Based on the finer-level data-driven approach combined with retraining, we present a methodology for extreme partial-sum quantization. Experimental results show that the proposed method can reduce the ADC resolution to 2 to 3 bits for CIFAR-10 dataset, which is the smaller ADC bit resolution than any previous CIM-based NN accelerators.
- 23 March 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 5, 70-73
Energy-Efficient In-Memory Binary Neural Network Accelerator Design Based on 8T2C SRAM Cell
We present an in-memory binary neural network (BNN) accelerator based on 8-transistor and 2-capacitor (8T2C) SRAM cell. The proposed SRAM computing-in-memory (CIM) cells rely on DRAM-like charge sharing operations to avoid undesirable static currents and potential read-disturb problems in conventional resistive SRAM-CIM designs. In addition, unlike the previous capacitive SRAM-based CIM designs, the proposed SRAM CIM does not consume energy when the input value is 0, thereby achieving the higher energy efficiency in benchmark testing. Measurement results of the 256$\times $ 64 array prototype chip in the 28-nm CMOS technology showed 3182 TOPS/W at 0.7 V which is$4.7\times $ higher energy efficiency than that of a state-of-the-art design.- 23 March 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 5, 70-73
Energy-Efficient In-Memory Binary Neural Network Accelerator Design Based on 8T2C SRAM Cell
We present an in-memory binary neural network (BNN) accelerator based on 8-transistor and 2-capacitor (8T2C) SRAM cell. The proposed SRAM computing-in-memory (CIM) cells rely on DRAM-like charge sharing operations to avoid undesirable static currents and potential read-disturb problems in conventional resistive SRAM-CIM designs. In addition, unlike the previous capacitive SRAM-based CIM designs, the proposed SRAM CIM does not consume energy when the input value is 0, thereby achieving the higher energy efficiency in benchmark testing. Measurement results of the 256$\times $ 64 array prototype chip in the 28-nm CMOS technology showed 3182 TOPS/W at 0.7 V which is$4.7\times $ higher energy efficiency than that of a state-of-the-art design.- Top Cited Papers
- 21 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 57 (6) , 1924-1935
BitBlade: Energy-Efficient Variable Bit-Precision Hardware Accelerator for Quantized Neural Networks
We introduce an area/energy-efficient precision-scalable neural network accelerator architecture. Previous precision-scalable hardware accelerators have limitations such as the under-utilization of multipliers for low bit-width operations and the large area overhead to support various bit precisions. To mitigate the problems, we first propose a bitwise summation, which reduces the area overhead for the bit-width scaling. In addition, we present a channel-wise aligning scheme (CAS) to efficiently fetch inputs and weights from on-chip SRAM buffers and a channel-first and pixel-last tiling (CFPL) scheme to maximize the utilization of multipliers on various kernel sizes. A test chip was implemented in 28-nm CMOS technology, and the experimental results show that the throughput and energy efficiency of our chip are up to 7.7$\times $ and 1.64$\times $ higher than those of the state-of-the-art designs, respectively. Moreover, additional 1.5–3.4$\times $ throughput gains can be achieved using the CFPL method compared to the CAS.- 21 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 57 (6) , 1924-1935
Top Cited PapersBitBlade: Energy-Efficient Variable Bit-Precision Hardware Accelerator for Quantized Neural Networks
We introduce an area/energy-efficient precision-scalable neural network accelerator architecture. Previous precision-scalable hardware accelerators have limitations such as the under-utilization of multipliers for low bit-width operations and the large area overhead to support various bit precisions. To mitigate the problems, we first propose a bitwise summation, which reduces the area overhead for the bit-width scaling. In addition, we present a channel-wise aligning scheme (CAS) to efficiently fetch inputs and weights from on-chip SRAM buffers and a channel-first and pixel-last tiling (CFPL) scheme to maximize the utilization of multipliers on various kernel sizes. A test chip was implemented in 28-nm CMOS technology, and the experimental results show that the throughput and energy efficiency of our chip are up to 7.7$\times $ and 1.64$\times $ higher than those of the state-of-the-art designs, respectively. Moreover, additional 1.5–3.4$\times $ throughput gains can be achieved using the CFPL method compared to the CAS. - Open Access
- 21 January 2019
- Published by Association for Computing Machinery (ACM)
- p. 645-650
In-memory batch-normalization for resistive memory based binary neural network hardware
- 21 January 2019
- Published by Association for Computing Machinery (ACM)
- p. 645-650
Open AccessIn-memory batch-normalization for resistive memory based binary neural network hardware