NTT unveils AI inference LSI that enables real-time AI inference processing from ultra-high-definition video on edge devices and terminals with strict power constraints.
NTT Corporation (NTT) has introduced a groundbreaking large-scale integration (LSI) chip capable of real-time AI inference processing for ultra-high-definition (4K) video at 30 frames per second (fps). Engineered with power efficiency in mind, this new technology is optimized for deployment in edge environments and devices with strict energy constraints—eliminating the need to compress high-resolution video for AI processing in real time.
When deployed in devices such as drones, this innovation allows for the identification of people and objects from altitudes up to 150 meters (492 feet)—the legal maximum for drone operation in Japan. In contrast, existing AI video inference technologies are typically limited to just 30 meters (98 feet). This extended range opens the door for enhanced applications like infrastructure inspection beyond the operator’s line of sight, streamlining operations and cutting down on labor and maintenance costs.
“The combination of low-power AI inferencing with ultra-high-definition video holds an enormous amount of potential, from infrastructure inspection to public safety to live sporting events,” said Kazu Gomi, president and CEO of NTT Research. “NTT’s LSI, which we believe to be the first of its kind to achieve such results, represents an important step forward in enabling AI inference at the edge and for power-constrained terminals.”
Also Read: DDN & Google Cloud Partner to Boost AI
Traditional edge AI devices are significantly restricted in terms of power usage—typically operating on just tens of watts, compared to the hundreds of watts consumed by GPU-based AI servers. NTT’s LSI overcomes these limitations with a proprietary AI inference engine developed in-house. This engine boosts performance by reducing computational load while preserving accuracy, using techniques such as interframe correlation and dynamic bit-precision control. It can run the widely used object detection algorithm YOLOv3 while keeping power consumption under 20 watts.
NTT has announced plans to commercialize the LSI within fiscal year 2025 through its subsidiary, NTT Innovative Devices Corporation. The technology was showcased at Upgrade 2025, NTT’s annual research and innovation summit, held in San Francisco on April 9–10.
The company’s researchers are also exploring how this LSI can integrate into the data-centric infrastructure (DCI) framework of NTT’s Innovative Optical and Wireless Network (IOWN) initiative, led in collaboration with the IOWN Global Forum. This next-generation network infrastructure uses the ultra-fast, low-latency All-Photonics Network to address challenges such as scalability, performance limitations, and energy efficiency in today’s digital landscape.
In addition, NTT is working alongside NTT DATA, Inc. to further advance this LSI’s potential by linking it with proprietary Attribute-Based Encryption (ABE) technologies. ABE supports fine-grained access controls and customizable data-sharing policies, making it ideal for secure, scalable deployment across existing systems and data environments.
