Seongyon Hong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Seongyon Hong has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Hoi-Jun Yoo
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
6 shared publications - Sangjin Kim
Korea Advanced Institute of Science and Technology (KAIST)
6 shared publications - Sangyeob Kim
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
5 shared publications - Jiwon Choi
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
3 shared publications - Soyeon Kim
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
3 shared publications - Wooyoung Jo
School of Electrical Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
3 shared publications
Latest Publications
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. PP (00189200) , 1-13
MIDAS: An Energy-Efficient Microscaling Quantization-Based Digital CIM Accelerator With Spatio-Temporal Cyclic Alignment for Generative AI
Digital compute-in-memory (DCIM) and the microscaling (MX) data format have emerged as promising hardware and algorithmic solutions, respectively, for energy-efficient generative AI (GenAI) processing. However, their combination introduces challenges, including redundant computation from mantissa alignment, excessive exponent handling overhead, and accuracy loss from static approximation schemes that fail to accommodate block-wise dynamic bit-significance. In this article, we present MIDAS, an MX-optimized DCIM-based GenAI accelerator with three key features: 1) spatio-temporal cyclic alignment (STCA) with sign-magnitude-based assembled column addition (SACA) enables compact weight mapping, high input reusability, and low-latency computation, achieving $1.72\times $ higher macro energy efficiency; 2) MX-aware minimum exponent post-detection (MEPD) eliminates redundant maximum exponent detection by exploiting the inherent characteristics of MX quantization, reducing exponent handling power, area, and latency by 68.8%, 63.2%, and 22.5%, respectively; and 3) sense amplifier-based dynamic bit (SADB)-significance handling with shared scale-aware channel regrouping (SSCR) dynamically adjusts approximation bit-positions according to each block’s shared scale, improving energy efficiency by 27.3% while minimizing accuracy loss. MIDAS is fabricated in Samsung 28-nm CMOS technology and occupies a 7.56 mm2 die area. MIDAS achieves macro energy and area efficiencies of 96.3 TFLOPS/W and 0.69 TFLOPS/mm2, respectively, under MXFP4. Compared to state-of-the-art DCIMs, MIDAS demonstrates $1.58\times $ higher system energy efficiency and $3.32\times $ higher macro energy efficiency, validating its effectiveness as an MX-optimized AI accelerator for next-generation GenAI applications.
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. PP (00189200) , 1-13
MIDAS: An Energy-Efficient Microscaling Quantization-Based Digital CIM Accelerator With Spatio-Temporal Cyclic Alignment for Generative AI
Digital compute-in-memory (DCIM) and the microscaling (MX) data format have emerged as promising hardware and algorithmic solutions, respectively, for energy-efficient generative AI (GenAI) processing. However, their combination introduces challenges, including redundant computation from mantissa alignment, excessive exponent handling overhead, and accuracy loss from static approximation schemes that fail to accommodate block-wise dynamic bit-significance. In this article, we present MIDAS, an MX-optimized DCIM-based GenAI accelerator with three key features: 1) spatio-temporal cyclic alignment (STCA) with sign-magnitude-based assembled column addition (SACA) enables compact weight mapping, high input reusability, and low-latency computation, achieving $1.72\times $ higher macro energy efficiency; 2) MX-aware minimum exponent post-detection (MEPD) eliminates redundant maximum exponent detection by exploiting the inherent characteristics of MX quantization, reducing exponent handling power, area, and latency by 68.8%, 63.2%, and 22.5%, respectively; and 3) sense amplifier-based dynamic bit (SADB)-significance handling with shared scale-aware channel regrouping (SSCR) dynamically adjusts approximation bit-positions according to each block’s shared scale, improving energy efficiency by 27.3% while minimizing accuracy loss. MIDAS is fabricated in Samsung 28-nm CMOS technology and occupies a 7.56 mm2 die area. MIDAS achieves macro energy and area efficiencies of 96.3 TFLOPS/W and 0.69 TFLOPS/mm2, respectively, under MXFP4. Compared to state-of-the-art DCIMs, MIDAS demonstrates $1.58\times $ higher system energy efficiency and $3.32\times $ higher macro energy efficiency, validating its effectiveness as an MX-optimized AI accelerator for next-generation GenAI applications.
- 15 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (10) , 3802-3815
C-Transformer: An Energy-Efficient Homogeneous DNN-Transformer/SNN-Transformer Processor for Large Language Models
In this article, we propose a new language model processor named the C-Transformer to address the external memory bottleneck of language models. It consists of three key functional blocks: 1) homogeneous deep-neural-network (DNN)–Transformer/spiking-Transformer core (HDSC) with a hybrid multiplication/accumulation unit (HMAU) to enhance hardware utilization; 2) output spike speculation unit (OSSU) to increase the energy efficiency of spike-domain processing; and 3) implicit weight generation unit (IWGU) with extended sign compression (ESC) to eliminate the external memory bottleneck. The chip is fabricated in Samsung’s 28-nm 1P8M CMOS technology and operates at a supply voltage of 0.7–1.1 V with a maximum frequency of 200 MHz and supports various tasks such as language modeling, translation, and summarization. For models like generative pre-trained transformer 2 (GPT-2), multilingual text-to-text transfer transformer (mT5), text-to-text transfer transformer (T5), and FairSeq MachineTranslation (FSMT), the C-Transformer achieves $0.21{\times }$ – $0.33{\times }$ computation energy and $0.37{\times }$ – $0.41{\times }$ external memory access (EMA) energy compared to the baseline. Our chip demonstrates 13.6% lower energy consumption than the previous state-of-the-art, despite having $2.1{\times }$ more parameters. Moreover, it consumes 63.8% less energy with a similar parameter size. The C-Transformer can complete various language model tasks with <1 s latency, notably FSMT in 0.09 s and GPT-2 in 0.656 s. By combining DNN-Transformer and Spiking-Transformer architectures, the C-Transformer enhances computational energy efficiency, eliminates external memory bottlenecks, and enables language models such as GPT-2 to achieve state-of-the-art performance on mobile devices.
- 15 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (10) , 3802-3815
C-Transformer: An Energy-Efficient Homogeneous DNN-Transformer/SNN-Transformer Processor for Large Language Models
In this article, we propose a new language model processor named the C-Transformer to address the external memory bottleneck of language models. It consists of three key functional blocks: 1) homogeneous deep-neural-network (DNN)–Transformer/spiking-Transformer core (HDSC) with a hybrid multiplication/accumulation unit (HMAU) to enhance hardware utilization; 2) output spike speculation unit (OSSU) to increase the energy efficiency of spike-domain processing; and 3) implicit weight generation unit (IWGU) with extended sign compression (ESC) to eliminate the external memory bottleneck. The chip is fabricated in Samsung’s 28-nm 1P8M CMOS technology and operates at a supply voltage of 0.7–1.1 V with a maximum frequency of 200 MHz and supports various tasks such as language modeling, translation, and summarization. For models like generative pre-trained transformer 2 (GPT-2), multilingual text-to-text transfer transformer (mT5), text-to-text transfer transformer (T5), and FairSeq MachineTranslation (FSMT), the C-Transformer achieves $0.21{\times }$ – $0.33{\times }$ computation energy and $0.37{\times }$ – $0.41{\times }$ external memory access (EMA) energy compared to the baseline. Our chip demonstrates 13.6% lower energy consumption than the previous state-of-the-art, despite having $2.1{\times }$ more parameters. Moreover, it consumes 63.8% less energy with a similar parameter size. The C-Transformer can complete various language model tasks with <1 s latency, notably FSMT in 0.09 s and GPT-2 in 0.656 s. By combining DNN-Transformer and Spiking-Transformer architectures, the C-Transformer enhances computational energy efficiency, eliminates external memory bottlenecks, and enables language models such as GPT-2 to achieve state-of-the-art performance on mobile devices.
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (4) , 1299-1310
Dyamond: Compact and Efficient 1T1C DRAM IMC Accelerator With Bit Column Addition for Memory-Intensive AI
This article proposes Dyamond, a one transistor, one capacitor (1T1C) dynamic random access memory (DRAM) in-memory computing (IMC) accelerator with architecture-to-circuit-level optimizations for high memory density and energy efficiency. The bit column addition (BCA) dataflow introduces output bit-wise accumulation to exploit varying accuracy and energy characteristics across different bit positions. The lower BCA (LBCA) reduces analog-to-digital converter (ADC) operations to enhance energy efficiency with inter-column analog accumulation. The higher BCA (HBCA) improves accuracy through signal enhancement and minimizes energy consumption per ADC readout with signal shift (SS). The design maximizes memory density by dedicating 1T1C cells solely to memory and integrating a compact computation circuit adjacent to the bitline sense amplifier. The memory access power is further reduced with a big-little array structure and a switchable sense amplifier (SWSA), which trades off retention time and energy consumption. Fabricated in 28-nm CMOS, Dyamond integrates 3.54-MB DRAM in a 6.48-mm2 area, achieving 27.2 TOPS/W peak efficiency and outstanding performance in advanced models such as BERT and GPT-2.
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (4) , 1299-1310
Dyamond: Compact and Efficient 1T1C DRAM IMC Accelerator With Bit Column Addition for Memory-Intensive AI
This article proposes Dyamond, a one transistor, one capacitor (1T1C) dynamic random access memory (DRAM) in-memory computing (IMC) accelerator with architecture-to-circuit-level optimizations for high memory density and energy efficiency. The bit column addition (BCA) dataflow introduces output bit-wise accumulation to exploit varying accuracy and energy characteristics across different bit positions. The lower BCA (LBCA) reduces analog-to-digital converter (ADC) operations to enhance energy efficiency with inter-column analog accumulation. The higher BCA (HBCA) improves accuracy through signal enhancement and minimizes energy consumption per ADC readout with signal shift (SS). The design maximizes memory density by dedicating 1T1C cells solely to memory and integrating a compact computation circuit adjacent to the bitline sense amplifier. The memory access power is further reduced with a big-little array structure and a switchable sense amplifier (SWSA), which trades off retention time and energy consumption. Fabricated in 28-nm CMOS, Dyamond integrates 3.54-MB DRAM in a 6.48-mm2 area, achieving 27.2 TOPS/W peak efficiency and outstanding performance in advanced models such as BERT and GPT-2.
- 30 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (10) , 3330-3341
LOG-CIM: An Energy-Efficient Logarithmic Quantization Computing-In-Memory Processor With Exponential Parallel Data Mapping and Zero-Aware 6T Dual-WL Cell
This article introduces the logarithmic quantization computing-in-memory processor (LOG-CIM), an advanced processor that extends the capabilities of conventional linear quantization (INT) by integrating logarithmic quantization (LOGQ) into computing-in-memory (CIM) architectures. Emphasizing three key features, the LOG-CIM significantly achieves high energy efficiency and memory density using LOGQ: 1) exponential parallel data mapping (EPDM) not only facilitates efficient shifted alignment and data accumulation, optimizing the processing pipeline for LOGQ complex computations but also achieves a 51.5% improvement in energy efficiency as a result; 2) the implementation of a zero-aware dual-word-line (ZADU) cell, which not only enhances parallelism but also reduces computation power by lowering read power 88.9% for 89.3% bit sparsity; and 3) shift-addition peripheral (SAP) logic, which maximizes throughput by performing the addition of different bit positions from two data read by the ZADU cell before they enter the adder tree, thereby enhancing 24.3% energy efficiency. The LOG-CIM is fabricated in 28-nm CMOS technology and occupies a 5.832 mm 2 with a total of 48-kB on-chip CIMs. It achieves 192 TOPS/W using conventional data mapping at 4-b INT weight and 4-b input activation and 116.4 TOPS/W using EPDM at 4-b LOGQ weight and 8-b INT input activation at 120-MHz clock and 0.85-V supply voltage with the ResNet-50 @ ImageNet dataset. When using the 4-b LOGQ weight for inference, the energy efficiency was 3.54 times higher than the 8-b INT weight with similar accuracy. During training, using 4-b LOGQ weight to reach the same accuracy increases epoch compared with using only 8-b INT weight, but consequentially saves 68.7% in relative total energy. Furthermore, it achieves 599-kb/mm 2 memory density.
- 30 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (10) , 3330-3341
LOG-CIM: An Energy-Efficient Logarithmic Quantization Computing-In-Memory Processor With Exponential Parallel Data Mapping and Zero-Aware 6T Dual-WL Cell
This article introduces the logarithmic quantization computing-in-memory processor (LOG-CIM), an advanced processor that extends the capabilities of conventional linear quantization (INT) by integrating logarithmic quantization (LOGQ) into computing-in-memory (CIM) architectures. Emphasizing three key features, the LOG-CIM significantly achieves high energy efficiency and memory density using LOGQ: 1) exponential parallel data mapping (EPDM) not only facilitates efficient shifted alignment and data accumulation, optimizing the processing pipeline for LOGQ complex computations but also achieves a 51.5% improvement in energy efficiency as a result; 2) the implementation of a zero-aware dual-word-line (ZADU) cell, which not only enhances parallelism but also reduces computation power by lowering read power 88.9% for 89.3% bit sparsity; and 3) shift-addition peripheral (SAP) logic, which maximizes throughput by performing the addition of different bit positions from two data read by the ZADU cell before they enter the adder tree, thereby enhancing 24.3% energy efficiency. The LOG-CIM is fabricated in 28-nm CMOS technology and occupies a 5.832 mm 2 with a total of 48-kB on-chip CIMs. It achieves 192 TOPS/W using conventional data mapping at 4-b INT weight and 4-b input activation and 116.4 TOPS/W using EPDM at 4-b LOGQ weight and 8-b INT input activation at 120-MHz clock and 0.85-V supply voltage with the ResNet-50 @ ImageNet dataset. When using the 4-b LOGQ weight for inference, the energy efficiency was 3.54 times higher than the 8-b INT weight with similar accuracy. During training, using 4-b LOGQ weight to reach the same accuracy increases epoch compared with using only 8-b INT weight, but consequentially saves 68.7% in relative total energy. Furthermore, it achieves 599-kb/mm 2 memory density.
- 15 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (1) , 157-172
C-DNN: An Energy-Efficient Complementary Deep-Neural-Network Processor With Heterogeneous CNN/SNN Core Architecture
In this article, we propose a complementary deep-neural-network (C-DNN) processor by combining convolutional neural network (CNN) and spiking neural network (SNN) to take advantage of them. The C-DNN processor can support both complementary inference and training with heterogeneous CNN and SNN core architecture. In addition, the C-DNN processor is the first DNN accelerator application-specific integrated circuit (ASIC) that can support CNN–SNN workload division by using their magnitude–energy tradeoff. The C-DNN processor integrates the CNN–SNN workload allocator and attention module to find a more energy-efficient network domain for each workload in DNN. They enable the C-DNN processor to operate at the energy optimal point. Moreover, the SNN processing element (PE) array with distributed L1 cache can reduce the redundant memory access for SNN processing, resulting in a 42.2%–49.1% reduction. For high energy-efficient DNN training, the C-DNN processor integrates the global counter and local delta-weight (LDW) unit to eliminate power-consuming counters for a forward delta-weight generation. Furthermore, the forward delta-weight-based sparsity generation (FDWSG) is proposed to reduce the number of operations for training by 31%–79% The C-DNN processor achieves an energy efficiency of 85.8 and 79.9 TOPS/W for inference with CIFAR-10 and CIFAR-100, respectively (VGG-16). Moreover, the C-DNN processor achieves ImageNet classification with state-of-the-art energy efficiency of 24.5 TOPS/W (ResNet-50). For training, the C-DNN processor achieves the state-of-the-art energy efficiency of 84.5 and 17.2 TOPS/W for CIFAR-10 and ImageNet, respectively. Furthermore, it achieves 77.1% accuracy for ImageNet training with ResNet-50.
- 15 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (1) , 157-172
C-DNN: An Energy-Efficient Complementary Deep-Neural-Network Processor With Heterogeneous CNN/SNN Core Architecture
In this article, we propose a complementary deep-neural-network (C-DNN) processor by combining convolutional neural network (CNN) and spiking neural network (SNN) to take advantage of them. The C-DNN processor can support both complementary inference and training with heterogeneous CNN and SNN core architecture. In addition, the C-DNN processor is the first DNN accelerator application-specific integrated circuit (ASIC) that can support CNN–SNN workload division by using their magnitude–energy tradeoff. The C-DNN processor integrates the CNN–SNN workload allocator and attention module to find a more energy-efficient network domain for each workload in DNN. They enable the C-DNN processor to operate at the energy optimal point. Moreover, the SNN processing element (PE) array with distributed L1 cache can reduce the redundant memory access for SNN processing, resulting in a 42.2%–49.1% reduction. For high energy-efficient DNN training, the C-DNN processor integrates the global counter and local delta-weight (LDW) unit to eliminate power-consuming counters for a forward delta-weight generation. Furthermore, the forward delta-weight-based sparsity generation (FDWSG) is proposed to reduce the number of operations for training by 31%–79% The C-DNN processor achieves an energy efficiency of 85.8 and 79.9 TOPS/W for inference with CIFAR-10 and CIFAR-100, respectively (VGG-16). Moreover, the C-DNN processor achieves ImageNet classification with state-of-the-art energy efficiency of 24.5 TOPS/W (ResNet-50). For training, the C-DNN processor achieves the state-of-the-art energy efficiency of 84.5 and 17.2 TOPS/W for CIFAR-10 and ImageNet, respectively. Furthermore, it achieves 77.1% accuracy for ImageNet training with ResNet-50.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 44 (1) , 28-37
COOL-NPU: Complementary Online Learning Neural Processing Unit
The authors propose a complementary online learning neural processing unit (COOL-NPU) to implement a highly accurate and high-energy-efficient online learning system. It reduces the energy consumption by combining the training methods of CNN and SNN, and eliminates the power overhead due to the redundant weight update by training trigger with SNN-gradient. The proposed SNN core reduces the energy consumption of SNN-gradient generation by 2-step encoding, and reduces inference power by hierarchical cache with LUT-mode. In addition, it supports neuron-level event-driven backward operation to maximize the effect of the training trigger. Fabricated with Samsung 28-nm CMOS technology, the COOL-NPU achieves 6.94 mJ/frame and 0.73 mAP for object detection, resulting in 47.7% energy reduction with a slight accuracy loss compared to previous state-of-the-art.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 44 (1) , 28-37
COOL-NPU: Complementary Online Learning Neural Processing Unit
The authors propose a complementary online learning neural processing unit (COOL-NPU) to implement a highly accurate and high-energy-efficient online learning system. It reduces the energy consumption by combining the training methods of CNN and SNN, and eliminates the power overhead due to the redundant weight update by training trigger with SNN-gradient. The proposed SNN core reduces the energy consumption of SNN-gradient generation by 2-step encoding, and reduces inference power by hierarchical cache with LUT-mode. In addition, it supports neuron-level event-driven backward operation to maximize the effect of the training trigger. Fabricated with Samsung 28-nm CMOS technology, the COOL-NPU achieves 6.94 mJ/frame and 0.73 mAP for object detection, resulting in 47.7% energy reduction with a slight accuracy loss compared to previous state-of-the-art.