Satoshi Kawakami
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
Satoshi Kawakami has a total of 130 co-authors. Here are the top co-authors by number of shared publications:
- Koji Inoue
Faculty of Information Science and Electrical Engineering, Kyushu University, Fukuoka, Japan
7 shared publications - Masamitsu Tanaka
Graduate School of Engineering, Nagoya University, Nagoya, Japan
4 shared publications - Teruo Tanimoto
Faculty of Information Science and Electrical Engineering, Kyushu University, Fukuoka, Japan
4 shared publications - Akira Takahara
Department of Pharmacology and Therapeutics, Faculty of Pharmaceutical Sciences, Toho University, 2–2–1 Miyama, Funabashi, Chiba 274–8510, Japan
3 shared publications - Ilkwon Byun
Faculty of Information Science and Electrical Engineering, Kyushu University, Fukuoka, Japan
3 shared publications Megumi Aimoto
Dept. Pharmacol. Therap., Fac. Pharmaceut. Sci. Toho Univ
3 shared publications
Latest Publications
- Open Access
- 12 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 32 (2) , 1-25
Effective Concurrent Attack Detection Framework through Laser Fault Informed Balanced Optical Neural Network Observation
Silicon Photonics-based AI Accelerators (SPAAs) have been considered as promising AI accelerators achieving high energy efficiency and low latency. While many researchers focus on improving SPAAs’ energy efficiency and latency, their physical security has only recently received attention. While it is essential to deliver strong optical neural network inferencing approaches, their success and adoption are predicated on their ability to deliver a secure execution environment. Towards this end, this article extends PrometheusFree to PrometheusFreev2 , an optical neural network framework that is capable of concurrent detection of laser fault injection attacks. This article first presents an illustrative threat of laser fault injection attacks on SPAAs, capable of subjecting the optical neural network to misclassifications. PrometheusFreev2 then addresses this threat by developing techniques for concurrent detection of the laser fault injection attacks. Furthermore, PrometheusFreev2 introduces an application of the Wavelength Division Perturbation (WDP) technique where wavelength-dependent Vector Matrix Multiplication (VMM) results are utilized to boost fault attack detection accuracy. This article goes on to introduce a novel optical vulnerability that overpowers the PrometheusFree detection capability, the swapping attack that exploits an architectural vulnerability of SPAAs. To address this practical and critical attack, PrometheusFreev2 features a dummy padding technique which injects random values to weight matrices to improve attack detectability. Simulation results show that PrometheusFreev2 achieves over 92% attack-caused misprediction recall as the use of the WDP technique squashes the attack success rate by 31–47% on average by using three wavelengths. Compared with prior art, PrometheusFreev2 constricts the average attack success ratio to 0.030, yielding a 92.8–95.0% reduction. The experimental results confirm the superiority of the concurrent detection and the boost in attack detection abilities imparted by the WDP and the dummy padding approaches.
- 12 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Design Automation of Electronic Systems
- Vol. 32 (2) , 1-25
Open AccessEffective Concurrent Attack Detection Framework through Laser Fault Informed Balanced Optical Neural Network Observation
Silicon Photonics-based AI Accelerators (SPAAs) have been considered as promising AI accelerators achieving high energy efficiency and low latency. While many researchers focus on improving SPAAs’ energy efficiency and latency, their physical security has only recently received attention. While it is essential to deliver strong optical neural network inferencing approaches, their success and adoption are predicated on their ability to deliver a secure execution environment. Towards this end, this article extends PrometheusFree to PrometheusFreev2 , an optical neural network framework that is capable of concurrent detection of laser fault injection attacks. This article first presents an illustrative threat of laser fault injection attacks on SPAAs, capable of subjecting the optical neural network to misclassifications. PrometheusFreev2 then addresses this threat by developing techniques for concurrent detection of the laser fault injection attacks. Furthermore, PrometheusFreev2 introduces an application of the Wavelength Division Perturbation (WDP) technique where wavelength-dependent Vector Matrix Multiplication (VMM) results are utilized to boost fault attack detection accuracy. This article goes on to introduce a novel optical vulnerability that overpowers the PrometheusFree detection capability, the swapping attack that exploits an architectural vulnerability of SPAAs. To address this practical and critical attack, PrometheusFreev2 features a dummy padding technique which injects random values to weight matrices to improve attack detectability. Simulation results show that PrometheusFreev2 achieves over 92% attack-caused misprediction recall as the use of the WDP technique squashes the attack success rate by 31–47% on average by using three wavelengths. Compared with prior art, PrometheusFreev2 constricts the average attack success ratio to 0.030, yielding a 92.8–95.0% reduction. The experimental results confirm the superiority of the concurrent detection and the boost in attack detection abilities imparted by the WDP and the dummy padding approaches.
- Open Access
- 1 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Computer
- Vol. 59 (2) , 142-149
Toward Ultra-High-Speed, Ultra-Low-Power Cryogenic Superconductor Computing
- 1 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Computer
- Vol. 59 (2) , 142-149
Open AccessToward Ultra-High-Speed, Ultra-Low-Power Cryogenic Superconductor Computing
- 26 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 24 (2) , 253-256
Approximate SFQ-Based Computing Architecture Modeling With Device-Level Guidelines
Single-flux-quantum (SFQ) logic has emerged as a promising post-Moore technology thanks to its ultra-fast and lowenergy operation. However, despite progress in various fields, its feasibility is questionable due to the prohibitive cooling cost. Proven conventional ideas, such as approximate computing, may help to resolve this challenge. However, introducing such ideas has been impossible due to the complex performance, power, and error trade-offs originating from the unique SFQ device characteristics. This work introduces approximate SFQ-based computing (AxSFQ) with an architecture modeling framework and essential design guidelines. Our optimized device-level AxSFQ showcases 30˜ 100 times energy efficiency improvement, which motivates further circuit and architecture-level exploration
- 26 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 24 (2) , 253-256
Approximate SFQ-Based Computing Architecture Modeling With Device-Level Guidelines
Single-flux-quantum (SFQ) logic has emerged as a promising post-Moore technology thanks to its ultra-fast and lowenergy operation. However, despite progress in various fields, its feasibility is questionable due to the prohibitive cooling cost. Proven conventional ideas, such as approximate computing, may help to resolve this challenge. However, introducing such ideas has been impossible due to the complex performance, power, and error trade-offs originating from the unique SFQ device characteristics. This work introduces approximate SFQ-based computing (AxSFQ) with an architecture modeling framework and essential design guidelines. Our optimized device-level AxSFQ showcases 30˜ 100 times energy efficiency improvement, which motivates further circuit and architecture-level exploration
- 5 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 24 (1) , 81-84
Data-Pattern-Driven LUT for Efficient In-Cache Computing in CNNs Acceleration
The lookup table (LUT)-based Processing-in-Memory (PIM) solutions perform computations by looking up precomputed results stored in LUTs, providing exceptional efficiency for complex operations such as multiplication, making them highly suitable for energy- and latency-efficient Convolutional Neural Network (CNN) inference tasks. However, including all possible results in the LUT naively demands exponential hardware resources, significantly limiting parallelism and increasing hardware area, latency, and power overhead. While decomposition and compression techniques can reduce the LUT size, they also introduce considerable memory access overhead and additional operations. To address these challenges, we conduct an extensive analysis to identify which data portions significantly impact accuracy in CNNs. Based on the insight that key data is concentrated in a small range, we propose a data-pattern-driven (DPD) optimization strategy, which approximates less critical data to drastically reduce LUT size while preserving computational efficiency with acceptable accuracy loss.
- 5 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Computer Architecture Letters
- Vol. 24 (1) , 81-84
Data-Pattern-Driven LUT for Efficient In-Cache Computing in CNNs Acceleration
The lookup table (LUT)-based Processing-in-Memory (PIM) solutions perform computations by looking up precomputed results stored in LUTs, providing exceptional efficiency for complex operations such as multiplication, making them highly suitable for energy- and latency-efficient Convolutional Neural Network (CNN) inference tasks. However, including all possible results in the LUT naively demands exponential hardware resources, significantly limiting parallelism and increasing hardware area, latency, and power overhead. While decomposition and compression techniques can reduce the LUT size, they also introduce considerable memory access overhead and additional operations. To address these challenges, we conduct an extensive analysis to identify which data portions significantly impact accuracy in CNNs. Based on the insight that key data is concentrated in a small range, we propose a data-pattern-driven (DPD) optimization strategy, which approximates less critical data to drastically reduce LUT size while preserving computational efficiency with acceptable accuracy loss.
- 1 January 2025
- Published by American Association for the Advancement of Science (AAAS) in Advanced Devices & Instrumentation
- Vol. 6
Photonic Compute-in-Wire: Remotely Driven Photonic Deep Neural Network with a Single Nonlinear Loop
The development of deep learning has motivated the advancement of unconventional computing that leverages analog physical systems such as analog electronics, spintronics, and photonics. These technologies have also led to the development of unique computational paradigms harnessing the features of analog devices, including compute-in-memory for nonvolatile devices and compute-in-sensor for analog electronics. What, then, are the computational paradigms that can exploit the characteristics of photonics? Optical computing has emerged as a promising candidate as it offers low-latency and low-power computation by utilizing the inherent parallelism of light. Additionally, the low-loss medium of optical fibers allows for the transmission of information over long distances. In this paper, we experimentally demonstrate a remotely driven optical neural network that combines these advantageous features. Namely, computations can be executed with data transfer over a photonic network, which provides a computational paradigm named photonic compute-in-wire. As a demonstration, we constructed an optoelectronic benchtop with a 20-km fiber access line. To facilitate the implementation of this concept, we extended the folded-in-time deep neural network (FiT-DNN) approach, which utilizes a single delay line with a nonlinear device, a dispersive element, and a modulator. This allows the insertion of a photonic DNN within a single transmission line. The parameters of our constructed photonic FiT-DNN were trained by using a digital twin of nonlinear optoelectronic dynamics. Through our benchmark experiments, we confirmed good accuracy for image recognition tasks. Our approach broadens the opportunities to utilize optical computation from local edge computing to in-network computing for low-latency and low-energy computation.
- 1 January 2025
- Published by American Association for the Advancement of Science (AAAS) in Advanced Devices & Instrumentation
- Vol. 6
Photonic Compute-in-Wire: Remotely Driven Photonic Deep Neural Network with a Single Nonlinear Loop
The development of deep learning has motivated the advancement of unconventional computing that leverages analog physical systems such as analog electronics, spintronics, and photonics. These technologies have also led to the development of unique computational paradigms harnessing the features of analog devices, including compute-in-memory for nonvolatile devices and compute-in-sensor for analog electronics. What, then, are the computational paradigms that can exploit the characteristics of photonics? Optical computing has emerged as a promising candidate as it offers low-latency and low-power computation by utilizing the inherent parallelism of light. Additionally, the low-loss medium of optical fibers allows for the transmission of information over long distances. In this paper, we experimentally demonstrate a remotely driven optical neural network that combines these advantageous features. Namely, computations can be executed with data transfer over a photonic network, which provides a computational paradigm named photonic compute-in-wire. As a demonstration, we constructed an optoelectronic benchtop with a 20-km fiber access line. To facilitate the implementation of this concept, we extended the folded-in-time deep neural network (FiT-DNN) approach, which utilizes a single delay line with a nonlinear device, a dispersive element, and a modulator. This allows the insertion of a photonic DNN within a single transmission line. The parameters of our constructed photonic FiT-DNN were trained by using a digital twin of nonlinear optoelectronic dynamics. Through our benchmark experiments, we confirmed good accuracy for image recognition tasks. Our approach broadens the opportunities to utilize optical computation from local edge computing to in-network computing for low-latency and low-energy computation.
- 25 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Design & Test
- Vol. 41 (6) , 5-7
Open-Source Silicon—Unleashing Innovation and Collaboration
- 25 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Design & Test
- Vol. 41 (6) , 5-7
Open-Source Silicon—Unleashing Innovation and Collaboration
- 18 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 30, 65-72
Design of The Ultra-Low-Power Driven VMM Configurations for μW Scale IoT Devices
Operating IoT devices by supplying power from an energy harvester and installing AI accelerators in IoT devices are required. Nevertheless, conventionally selected architectures for AI processing require a large amount of power, making it difficult to operate in low-power bands such as IoT devices or even impossible to operate in the first place. Therefore, driving AI accelerators with power that energy harvesters can supply is an issue. However, there has been no past exploration of AI accelerators in the driven of the μW scale. In this paper, we analyze the configuration of Vector Matrix Multiplier, mainly used for AI accelerators, and show the effective configuration for μW scale IoT devices. Using 180nm CMOS to synthesize the four architectures of various sizes, we characterize device performance and analyze energy consumption and circuit area. As a result of the analysis, it shows that the configuration in which all calculations are deployed on the circuit can have the lowest energy consumption. In addition, we found that when there is a limit on circuit area, a configuration in which some calculations are performed in the time domain by lowering the voltage is suitable.
- 18 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 30, 65-72
Design of The Ultra-Low-Power Driven VMM Configurations for μW Scale IoT Devices
Operating IoT devices by supplying power from an energy harvester and installing AI accelerators in IoT devices are required. Nevertheless, conventionally selected architectures for AI processing require a large amount of power, making it difficult to operate in low-power bands such as IoT devices or even impossible to operate in the first place. Therefore, driving AI accelerators with power that energy harvesters can supply is an issue. However, there has been no past exploration of AI accelerators in the driven of the μW scale. In this paper, we analyze the configuration of Vector Matrix Multiplier, mainly used for AI accelerators, and show the effective configuration for μW scale IoT devices. Using 180nm CMOS to synthesize the four architectures of various sizes, we characterize device performance and analyze energy consumption and circuit area. As a result of the analysis, it shows that the configuration in which all calculations are deployed on the circuit can have the lowest energy consumption. In addition, we found that when there is a limit on circuit area, a configuration in which some calculations are performed in the time domain by lowering the voltage is suitable.
- Open Access
- 9 May 2023
- Published by Association for Computing Machinery (ACM)
- p. 219
Evaluating floating-point multipliers with opto-electrical hybrid circuits
Nanophotonic technology has ultra-low latency and low energy consumption characteristics, and several previous research have shown the potential of photonic-based computation. However, existing optical circuits are based on low-bit-width integer operations with analog manners, so there is less research on optical floating-point operations. Providing the benefits of analog optical circuits (i.e., low latency and low energy) to floating-point multipliers (FMs) requires careful consideration. In particular, introducing optical circuits unnecessarily increases the overhead of optical-to-electrical (O/E) and analog-to-digital (A/D) conversion, thus reducing system performance efficiency. Appropriate processing balancing between electrical and optical circuits is an important issue in OE hybrid circuit design. This study evaluates FM latency and energy consumption with different combinations of optical and electrical circuits. We design two versions of an Opto-Electrical FM (OEFM): M-OEFM, an optical implementation of an integer multiplier in an FM, and MA-OEFM, an optical implementation of an integer multiplier and adder functions. These designs are based on a policy of priority optical circuit implementation of the dominant components of conventional electrical circuits. Experimental results indicate that the M-OEFM achieved a 56 % reduction in latency and a 41 % reduction in energy consumption compared with conventional electric circuits. The MA-OEFM achieved an 88 % reduction in latency and a 19 % reduction in energy consumption compared with conventional electric circuits. The MA-OEFM is superior to M-OEFM in terms of energy-delay product (EDP). The results also indicate that appropriate co-design of analog optical and digital electrical circuits enables highly efficient FMs.
- 9 May 2023
- Published by Association for Computing Machinery (ACM)
- p. 219
Open AccessEvaluating floating-point multipliers with opto-electrical hybrid circuits
Nanophotonic technology has ultra-low latency and low energy consumption characteristics, and several previous research have shown the potential of photonic-based computation. However, existing optical circuits are based on low-bit-width integer operations with analog manners, so there is less research on optical floating-point operations. Providing the benefits of analog optical circuits (i.e., low latency and low energy) to floating-point multipliers (FMs) requires careful consideration. In particular, introducing optical circuits unnecessarily increases the overhead of optical-to-electrical (O/E) and analog-to-digital (A/D) conversion, thus reducing system performance efficiency. Appropriate processing balancing between electrical and optical circuits is an important issue in OE hybrid circuit design. This study evaluates FM latency and energy consumption with different combinations of optical and electrical circuits. We design two versions of an Opto-Electrical FM (OEFM): M-OEFM, an optical implementation of an integer multiplier in an FM, and MA-OEFM, an optical implementation of an integer multiplier and adder functions. These designs are based on a policy of priority optical circuit implementation of the dominant components of conventional electrical circuits. Experimental results indicate that the M-OEFM achieved a 56 % reduction in latency and a 41 % reduction in energy consumption compared with conventional electric circuits. The MA-OEFM achieved an 88 % reduction in latency and a 19 % reduction in energy consumption compared with conventional electric circuits. The MA-OEFM is superior to M-OEFM in terms of energy-delay product (EDP). The results also indicate that appropriate co-design of analog optical and digital electrical circuits enables highly efficient FMs.
- 2 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Applied Superconductivity
- Vol. 33 (4) , 1-11
50-GFLOPS Floating-Point Adder and Multiplier Using Gate-Level-Pipelined Single-Flux-Quantum Logic With Frequency-Increased Clock Distribution
We demonstrate the functioning of a high-throughput, gate-level-pipelined floating-point adder and multiplier over 50 GHz. The gate-level-pipelined floating-point adder and multiplier requires dedicated circuit blocks to wait until other circuit blocks complete calculations because of the dependence between their sign, exponent, and significand parts. We revealed that the resultant delay difference of the waiting circuit blocks hinders high-frequency operation if the predesigned circuit blocks with the fixed clock distribution are connected in a simple manner. We showed that clock distribution needs to synchronize with every pipeline stage regardless of the circuit blocks to minimize the delay difference between the circuit blocks for circuits containing the waiting circuit blocks (e.g., the floating-point adder and multiplier). We designed a 5-bit floating-point adder and multiplier to demonstrate the effectiveness of the clock distribution experimentally. The test chips were fabricated using AIST 10-kA/cm $\boldsymbol{^{2}}$ Advanced Process 2. We verified the high-speed operation at over 50 GHz in the floating-point adder and multiplier. The maximum clock frequency and throughput of the floating-point adder were 56 GHz and 56 GFLOPS, respectively. The corresponding values for the floating-point multiplier were 63 GHz and 63 GFLOPS, respectively.
- 2 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Applied Superconductivity
- Vol. 33 (4) , 1-11
50-GFLOPS Floating-Point Adder and Multiplier Using Gate-Level-Pipelined Single-Flux-Quantum Logic With Frequency-Increased Clock Distribution
We demonstrate the functioning of a high-throughput, gate-level-pipelined floating-point adder and multiplier over 50 GHz. The gate-level-pipelined floating-point adder and multiplier requires dedicated circuit blocks to wait until other circuit blocks complete calculations because of the dependence between their sign, exponent, and significand parts. We revealed that the resultant delay difference of the waiting circuit blocks hinders high-frequency operation if the predesigned circuit blocks with the fixed clock distribution are connected in a simple manner. We showed that clock distribution needs to synchronize with every pipeline stage regardless of the circuit blocks to minimize the delay difference between the circuit blocks for circuits containing the waiting circuit blocks (e.g., the floating-point adder and multiplier). We designed a 5-bit floating-point adder and multiplier to demonstrate the effectiveness of the clock distribution experimentally. The test chips were fabricated using AIST 10-kA/cm $\boldsymbol{^{2}}$ Advanced Process 2. We verified the high-speed operation at over 50 GHz in the floating-point adder and multiplier. The maximum clock frequency and throughput of the floating-point adder were 56 GHz and 56 GFLOPS, respectively. The corresponding values for the floating-point multiplier were 63 GHz and 63 GFLOPS, respectively.