Tomoya Akabe
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Tomoya Akabe has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Yasuhiko Nakashima
Division of Information Science, Nara Institute of Science and Technology, Ikoma, Japan
2 shared publications - Duong Thi Sang
Graduation School of Information Science, Nara Institute of Science and Technology (NAIST), Ikoma, Nara, Japan
1 shared publication - Ren Imamura
Graduation School of Information Science, Nara Institute of Science and Technology (NAIST), Ikoma, Nara, Japan
1 shared publication - Vu Trung Duong Le
Nara Institute of Science and Technology
1 shared publication
Latest Publications
- Open Access
- 5 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 29476-29492
Energy Consumption Optimization of Multi-Dimensional U-Nets on CGLA
The increasing demand for energy-efficient medical image processing has highlighted the limitations of traditional GPU-based solutions, particularly in mobile or remote scenarios where power consumption and computational efficiency are critical. Two-dimensional (2D) and three-dimensional (3D) convolutional neural networks (CNNs), such as U-Net (2D U-Net), have become essential in computer vision and medical diagnosis tasks. However, while 2D and 3D CNNs, widely used for medical image segmentation, offer superior performance in capturing spatial context and suffer from performance bottlenecks and high energy consumption on traditional GPUs, they significantly increase computational and memory requirements, posing challenges for conventional hardware. To address these challenges, we propose the Interposed Memory Accelerator eXtension series 3 (IMAX3), a novel chiplet-based, non-von Neumann architecture with coarse-grained linear accelerator (CGLA) to efficiently support both 2D and 3D U-Net models, delivering superior energy efficiency and computational performance. IMAX3 optimizes data processing by employing advanced data preloading mechanisms and double-buffered cache memory, reducing redundant memory retrievals and minimizing inter-chip communication. Our experimental results demonstrate that IMAX3, implemented on 28 nm technology, occupies only 12.5 mm2, consumes just 2.61–2.67 W, and significantly outperforms General-Purpose Computing on Graphics Processing Units (GPGPUs), Application-Specific Integrated Circuit (ASIC), Coarse-Grained Reconfigurable Architecture (CGRA), and Field-Programmable Gate Array (FPGA)-based designs. For 2D U-Net, IMAX3 achieves an energy efficiency of 18.68 GOPs/W, surpassing FPGA-based designs by 1.68× to 5.6× and reducing energy usage by 1.8×. In the case of ICA U-Net (3D U-Net), IMAX3 demonstrates 1.47× to 3.56× higher energy efficiency compared to ASIC and CGRA-based designs. These results establish IMAX3 as a scalable and energy-efficient solution for next-generation medical image processing applications.
- 5 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 29476-29492
Open AccessEnergy Consumption Optimization of Multi-Dimensional U-Nets on CGLA
The increasing demand for energy-efficient medical image processing has highlighted the limitations of traditional GPU-based solutions, particularly in mobile or remote scenarios where power consumption and computational efficiency are critical. Two-dimensional (2D) and three-dimensional (3D) convolutional neural networks (CNNs), such as U-Net (2D U-Net), have become essential in computer vision and medical diagnosis tasks. However, while 2D and 3D CNNs, widely used for medical image segmentation, offer superior performance in capturing spatial context and suffer from performance bottlenecks and high energy consumption on traditional GPUs, they significantly increase computational and memory requirements, posing challenges for conventional hardware. To address these challenges, we propose the Interposed Memory Accelerator eXtension series 3 (IMAX3), a novel chiplet-based, non-von Neumann architecture with coarse-grained linear accelerator (CGLA) to efficiently support both 2D and 3D U-Net models, delivering superior energy efficiency and computational performance. IMAX3 optimizes data processing by employing advanced data preloading mechanisms and double-buffered cache memory, reducing redundant memory retrievals and minimizing inter-chip communication. Our experimental results demonstrate that IMAX3, implemented on 28 nm technology, occupies only 12.5 mm2, consumes just 2.61–2.67 W, and significantly outperforms General-Purpose Computing on Graphics Processing Units (GPGPUs), Application-Specific Integrated Circuit (ASIC), Coarse-Grained Reconfigurable Architecture (CGRA), and Field-Programmable Gate Array (FPGA)-based designs. For 2D U-Net, IMAX3 achieves an energy efficiency of 18.68 GOPs/W, surpassing FPGA-based designs by 1.68× to 5.6× and reducing energy usage by 1.8×. In the case of ICA U-Net (3D U-Net), IMAX3 demonstrates 1.47× to 3.56× higher energy efficiency compared to ASIC and CGRA-based designs. These results establish IMAX3 as a scalable and energy-efficient solution for next-generation medical image processing applications.
- Open Access
- 30 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 31899-31911
IMAX: A Power-Efficient Multilevel Pipelined CGLA and Applications
With rapid advances in artificial intelligence applications, computational demands have grown in size and complexity, necessitating flexible and highly efficient hardware architectures. Although traditional hardware solutions have demonstrated high performance and efficiency for specific applications, they still face challenges in terms of flexibility and energy efficiency. To overcome these shortcomings, we propose a new architecture based on coarse-grained linear array (CGLA), called "IMAX." IMAX adopts a linear structure in which cache memory and processing units are alternately arranged, absorbing irregular memory access latencies and achieving superior performance and energy efficiency. IMAX3 introduces further optimizations in communication performance, double buffering techniques, and improved sparse matrix multiplication (SpGEMM), achieving even higher performance. Real-time evaluation results on the Xilinx VPK180 SoC demonstrate that IMAX3 achieves significant improvement over conventional graphics processing units (GPUs) and other architectures. In executing Fast Fourier Transforms (FFT), IMAX3 achieves up to 10 times the energy efficiency of Jetson AGX Orin. For SpGEMM, IMAX3 delivers up to 503 times the speed of GTX 1080Ti and up to 162 times that of Jetson AGX Orin. In comparison with related architecture, IMAX3 also exhibits significantly superior performance in calculations such as matrix multiplication and FFT. In matrix multiplication, IMAX3 achieved approximately 23 times faster execution than STRELA and 61 times faster than RipTide, with substantial improvements in energy efficiency. These results confirm IMAX3 as a flexible and efficient hardware platform, outperforming traditional GPUs and other architectures in both computational performance and energy efficiency, particularly in real-time processing and low-power environments.
- 30 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 31899-31911
Open AccessIMAX: A Power-Efficient Multilevel Pipelined CGLA and Applications
With rapid advances in artificial intelligence applications, computational demands have grown in size and complexity, necessitating flexible and highly efficient hardware architectures. Although traditional hardware solutions have demonstrated high performance and efficiency for specific applications, they still face challenges in terms of flexibility and energy efficiency. To overcome these shortcomings, we propose a new architecture based on coarse-grained linear array (CGLA), called "IMAX." IMAX adopts a linear structure in which cache memory and processing units are alternately arranged, absorbing irregular memory access latencies and achieving superior performance and energy efficiency. IMAX3 introduces further optimizations in communication performance, double buffering techniques, and improved sparse matrix multiplication (SpGEMM), achieving even higher performance. Real-time evaluation results on the Xilinx VPK180 SoC demonstrate that IMAX3 achieves significant improvement over conventional graphics processing units (GPUs) and other architectures. In executing Fast Fourier Transforms (FFT), IMAX3 achieves up to 10 times the energy efficiency of Jetson AGX Orin. For SpGEMM, IMAX3 delivers up to 503 times the speed of GTX 1080Ti and up to 162 times that of Jetson AGX Orin. In comparison with related architecture, IMAX3 also exhibits significantly superior performance in calculations such as matrix multiplication and FFT. In matrix multiplication, IMAX3 achieved approximately 23 times faster execution than STRELA and 61 times faster than RipTide, with substantial improvements in energy efficiency. These results confirm IMAX3 as a flexible and efficient hardware platform, outperforming traditional GPUs and other architectures in both computational performance and energy efficiency, particularly in real-time processing and low-power environments.