Hitachi Develops Edge AI Semiconductor as the Foundational Technology for Physical AI Supporting HMAX Industry
Hitachi and Hitachi High-Tech have developed a power-efficient edge AI semiconductor for industrial equipment. Achieving 10x higher power efficiency, it enables real-time data analysis directly within machines without dedicated servers.
📋 Article Processing Timeline
- 📰 Published: April 25, 2026 at 01:10
- 🔍 Collected: April 24, 2026 at 16:31
- 🤖 AI Analyzed: April 24, 2026 at 22:31 (5h 59m after Collected)
Hitachi, Ltd. (hereinafter "Hitachi") and Hitachi High-Tech Corporation (hereinafter "Hitachi High-Tech") have developed an edge AI semiconductor*1 that can be installed in a wide range of industrial products such as manufacturing equipment, inspection systems, industrial robots, logistics equipment, building, and energy facilities, as a foundational technology supporting the next-generation solution suite "HMAX Industry" for the industrial sector. This semiconductor features high-speed processing and power savings, analyzing diverse on-site data such as images, sound, and vibration in real-time within the equipment.
In recent evaluations using actual machine data, it was confirmed that processing can be executed with over 10 times higher power efficiency compared to conventional methods*2, and that it operates stably with low power consumption suitable for in-device use. This provides the prospect of executing advanced inspection and monitoring processes directly within the equipment, which previously required dedicated servers. It has now reached a stage where it can be implemented in various on-site devices, particularly at manufacturing sites with severe constraints on installation space and power consumption.
Going forward, utilizing this edge AI semiconductor as a cross-sectional execution platform supporting the on-site application of physical AI, the companies will deploy digital services that process and utilize equipment data on the spot, leading to stabilized quality, improved yields, and increased productivity. Using an evaluation environment combining the edge AI semiconductor, lightweight AI models, and software, implementation and operation tailored to customers' equipment and manufacturing lines will proceed sequentially. As a core technology of "HMAX Industry" centered on physical AI incorporating domain knowledge cultivated in the industrial sector, it will be expanded to diverse fields such as general manufacturing, logistics, and building/energy.
*1 Edge AI semiconductor: A semiconductor chip installed directly in network terminal devices (edge devices) to execute AI inference within the equipment.
*2 Compared power efficiency for the developed lightweight AI model for edge use against state-of-the-art GPUs (GPU power efficiency calculated based on catalog values).
Background and Challenges
At industrial sites such as manufacturing, logistics, and building/energy, there is a demand to analyze data obtained from equipment in real-time to stabilize quality, improve productivity, and streamline maintenance. However, with conventional edge AI systems, power consumption, installation space, and processing loads when handling multiple sensor data have been bottlenecks, making full-scale deployment difficult in some cases. Hitachi has previously developed edge AI technology that analyzes diverse sensor data such as images, sound, and vibration in real-time with low power consumption*3. This time, realizing these results as an edge AI semiconductor installed and operated in actual industrial products, application at the sites targeted by "HMAX Industry" will begin.
*3 Developed edge AI technology to strengthen on-site application of Lumada 3.0 - R&D: October 14, 2025
Key Technology Points
1. Lightweight AI model for the edge that can be directly embedded in industrial products
Hitachi has developed a lightweight AI model for the edge designed to be embedded in industrial products. Since equipment used in factories and building facilities face power and space constraints, installing AI requiring massive computation within the equipment was a challenge. Now, by combining a CNN*4 that captures minute differences in images with a Transformer*5 that understands overall trends, they achieved both the lightweight nature required for in-device implementation and the high inference accuracy demanded in industrial applications like inspection and monitoring. Designed not to depend on specific machine models, use cases are expanding while proceeding with implementation and evaluation in inspection/measurement devices and industrial machinery.
2. Application examples reducing time and processing load for inspection and monitoring
In inspection and monitoring requiring high precision, multiple image capturing and complex analytical processing are often bottlenecks. This time, as a representative example, verification using measured data was conducted in the semiconductor inspection and measurement field, confirming the possibility of replacing high-precision measurement processes—which conventionally layered multiple images—with AI processing on a single image. This provided the prospect of ensuring necessary accuracy while reducing the number of image captures, confirming it leads to faster inline inspection/measurement*6 and reduced equipment load. Similar concepts will be sequentially applied to other inspection and monitoring equipment, such as parts visual inspection and facility condition monitoring.
Figure 1: Processing results by the lightweight AI model for edge use
3. Operation confirmation of proprietary edge AI semiconductor enabling low-power AI processing within equipment
Edge AI semiconductors can enhance power efficiency compared to general-purpose processors by optimizing arithmetic circuits and memory configurations according to the AI model executed. This time, evaluating a chip with a circuit designed to match the computation of the lightweight AI model for the edge, it was confirmed that processing can be executed with over 10 times higher power efficiency than conventional methods, and that it operates stably within the power range usable inside industrial equipment.
In recent evaluations using actual machine data, it was confirmed that processing can be executed with over 10 times higher power efficiency compared to conventional methods*2, and that it operates stably with low power consumption suitable for in-device use. This provides the prospect of executing advanced inspection and monitoring processes directly within the equipment, which previously required dedicated servers. It has now reached a stage where it can be implemented in various on-site devices, particularly at manufacturing sites with severe constraints on installation space and power consumption.
Going forward, utilizing this edge AI semiconductor as a cross-sectional execution platform supporting the on-site application of physical AI, the companies will deploy digital services that process and utilize equipment data on the spot, leading to stabilized quality, improved yields, and increased productivity. Using an evaluation environment combining the edge AI semiconductor, lightweight AI models, and software, implementation and operation tailored to customers' equipment and manufacturing lines will proceed sequentially. As a core technology of "HMAX Industry" centered on physical AI incorporating domain knowledge cultivated in the industrial sector, it will be expanded to diverse fields such as general manufacturing, logistics, and building/energy.
*1 Edge AI semiconductor: A semiconductor chip installed directly in network terminal devices (edge devices) to execute AI inference within the equipment.
*2 Compared power efficiency for the developed lightweight AI model for edge use against state-of-the-art GPUs (GPU power efficiency calculated based on catalog values).
Background and Challenges
At industrial sites such as manufacturing, logistics, and building/energy, there is a demand to analyze data obtained from equipment in real-time to stabilize quality, improve productivity, and streamline maintenance. However, with conventional edge AI systems, power consumption, installation space, and processing loads when handling multiple sensor data have been bottlenecks, making full-scale deployment difficult in some cases. Hitachi has previously developed edge AI technology that analyzes diverse sensor data such as images, sound, and vibration in real-time with low power consumption*3. This time, realizing these results as an edge AI semiconductor installed and operated in actual industrial products, application at the sites targeted by "HMAX Industry" will begin.
*3 Developed edge AI technology to strengthen on-site application of Lumada 3.0 - R&D: October 14, 2025
Key Technology Points
1. Lightweight AI model for the edge that can be directly embedded in industrial products
Hitachi has developed a lightweight AI model for the edge designed to be embedded in industrial products. Since equipment used in factories and building facilities face power and space constraints, installing AI requiring massive computation within the equipment was a challenge. Now, by combining a CNN*4 that captures minute differences in images with a Transformer*5 that understands overall trends, they achieved both the lightweight nature required for in-device implementation and the high inference accuracy demanded in industrial applications like inspection and monitoring. Designed not to depend on specific machine models, use cases are expanding while proceeding with implementation and evaluation in inspection/measurement devices and industrial machinery.
2. Application examples reducing time and processing load for inspection and monitoring
In inspection and monitoring requiring high precision, multiple image capturing and complex analytical processing are often bottlenecks. This time, as a representative example, verification using measured data was conducted in the semiconductor inspection and measurement field, confirming the possibility of replacing high-precision measurement processes—which conventionally layered multiple images—with AI processing on a single image. This provided the prospect of ensuring necessary accuracy while reducing the number of image captures, confirming it leads to faster inline inspection/measurement*6 and reduced equipment load. Similar concepts will be sequentially applied to other inspection and monitoring equipment, such as parts visual inspection and facility condition monitoring.
Figure 1: Processing results by the lightweight AI model for edge use
3. Operation confirmation of proprietary edge AI semiconductor enabling low-power AI processing within equipment
Edge AI semiconductors can enhance power efficiency compared to general-purpose processors by optimizing arithmetic circuits and memory configurations according to the AI model executed. This time, evaluating a chip with a circuit designed to match the computation of the lightweight AI model for the edge, it was confirmed that processing can be executed with over 10 times higher power efficiency than conventional methods, and that it operates stably within the power range usable inside industrial equipment.