AI continues to rapidly spread and evolve, contributing significantly to efficiency and advancement across various industries and today, it has become an indispensable element of society. However, challenges, such as the exponential growth of learning data, increased power consumption, communication latency, and security concerns, are becoming increasingly evident. SEC is tackling these issues by focusing on Edge AI.
This approach leverages technologies like FPGA, a low-power, rewritable circuit, reservoir computing, which enables learning from small amounts of data in a short time, and XR (Extended Reality) as next-generation devices. These technologies hold the potential to shape the future of AI and are being developed for real-world implementation.
Edge AI circuit implementation technology
SEC has the technology to utilize FPGAs, which are high-speed and power-saving integrated circuits, as implementation devices for edge AI. SEC has established a method that enables software engineers to implement intelligent processing on the FPGA development environment, which is premised on circuit design. The technological infrastructure that transcends the boundaries between hardware and software is SEC's great strength.
Edge AI algorithms that learn with small amounts of data
Reservoir computing is a machine learning algorithm that does not require large amounts of training data and can be trained in a short time. This is especially effective at sites where a large amount of training data cannot be acquired or for learning highly individualized events. By combining this technology, which is suitable for edge AI applications, with hardware implementations, highly practical edge AI is realized.
Edge AI implementation capabilities cultivated through real-time technology
Edge AI demonstrates its true value in fields where immediate decision-making and control are required. The fields that SEC has been working in for many years, such as space, robotics, automobiles, and social infrastructure, are exactly the areas where edge AI is expected to play an active role. We are opening up new possibilities for edge AI based on the real-time technology we have cultivated over many years.
We are collaborating with Kyushu Institute of Technology and other organizations to develop and implement ultra-low-power edge AI chips and to build a software development environment for use with these chips.
AI has become an essential technology for progress in society, but it still presents many challenges in terms of processing power, energy efficiency, and other areas. To tackle these issues, it is essential to take the latest insight from neuroscience and neurophysiology, as well as the computer theory that can put this insight into practice, and to integrate these concepts into hardware research.
We are working to develop an edge AI chip with an integrated circuit that can run models based on reservoir computing—a type of machine learning algorithm that expresses the mechanisms of the human brain as formulas. We also seek to utilize these chips and realize real-world implementation with AI-powered edge devices in the fields of robotics and IoT. At the same time, we are engaging in initiatives to build a software development environment to achieve our goal of implementation.
Reservoir computing is a machine learning algorithm that realizes AI. The ability to finish training in a short period with a small amount of data is a significant feature, and this technology is attracting attention as a technology that overturns the conventional wisdom of AI, which requires a large amount of training data and computation.
AI is being used to detect failures in manufacturing equipment and industrial robots, but because of differences in equipment and installation environments, AI that has already been trained is unable to accurately detect failures. With reservoir computing, learning can be completed in a short period of time after installation for each device, thus accommodating variations in devices and environments.
We are tackling development in collaboration with Kyushu Institute of Technology and others of ultra-low-power consumption edge AI chips that are equipped with reservoir computing. We are working on the practical application of edge AI chips and building a software development environment to use edge AI chips.
FPGAs are programmable integrated circuits. Usually, a central processing unit (CPU) is used as the computing device in a computer system, but the circuit configuration of the CPU itself is not flexible, and flexibility is ensured by the software that runs it. In contrast, FPGAs enable users to change (program) the circuit configuration itself and parallel computation processing enables processing that is faster than CPUs. While graphics processing units (GPUs) are programmable, high-speed computing devices, FPGAs consume less power and have advantages over GPUs as implementation devices for edge AI for which power consumption constraints are anticipated.
Even though FPGAs are rewritable integrated circuits, rewriting circuits is not as easy as software engineers writing programs. In 2017, we began collaborating with the Kyushu Institute of Technology on the topic of circuitry for intelligent processing for robots using FPGAs and we acquired a method for software engineers to implement programs on FPGAs. Based on the knowledge cultivated in this way, we are also working on the development of an edge AI chip that implements reservoir computing on FPGAs in “Technological Development of AI Chips and Next-generation Computing that Enable High-efficiency and Fast Processing” (a NEDO publicly solicited project).
We are engaged in research and development of XR (MR, AR, VR) as one of the application fields of Edge AI.
In 2017, SEC began joint research on MR technology with the Japan Aerospace Exploration Agency (JAXA) to develop security technology that can safely manage and project sensitive 3D models, as well as advanced MR technology utilizing AI and various sensors.
SEC has become a ROHM's Solist-AI™ ecosystem partner.
Solist-AI™ is ROHM's on-device AI solution for the edge computing field. It enables real-time learning and inference processing on a stand-alone edge device without relying on a cloud server, and features a compact design and low power consumption.
The Solist-AI™ microcontroller is an on-device AI microcontroller suitable for reservoir computing implementations. SEC has implemented reservoir computing, which is under research and development, in the Solist-AI™ microcontroller.
AI is being used to detect failures in manufacturing equipment and industrial robots, but because of differences in equipment and installation environments, AI that has already been trained is unable to accurately detect failures. The Solist-AI™ microcontroller, which is equipped with reservoir computing, can complete training quickly with minimal data after installation, enabling flexible support of variations in equipment and environments.
- Kyushu Institute of Technology
- Japan Aerospace Exploration Agency (JAXA)
- ROHM Co., Ltd.