Video-based AI is becoming a huge part of how we manage public spaces, monitor infrastructure, and operate autonomous systems. But while high-resolution video can capture plenty of useful detail, turning that footage into real-time, usable data is difficult, especially when the processing needs to happen directly on a small device rather than a powerful server. Until now, that is.

NTT believes it has found a way to bridge the gap with a new chip designed for “edge” use: an AI inference processor that can handle full 4K video in real time, while using very little power.

On the Edge

“Edge” devices are small computers, such as those on drones or security cameras, that process data on the spot instead of sending it to the cloud. It's a useful function to have when speed, privacy, or limited connectivity are a concern. The part of AI that actually produces a result, such as detecting a person in a video, is called inference.

Right now, performing AI inference at the edge comes with tight power and space limitations, which has often meant compromising on image quality. Systems tend to shrink 4K video down to lower resolutions before running AI on it, to reduce the workload. That works up to a point, but makes it harder to spot small or distant objects. And small or distant objects are sometimes the things you most want to detect.

4K, 30fps, all under 20W

NTT’s new Large Scale Integration (LSI) chip is designed to tackle this problem directly. It enables full-resolution (4K) AI inference at 30 frames per second, while still staying under 20 watts of power consumption. That’s low enough to fit into drones or other compact systems, and efficient enough to run for extended periods without draining a battery or requiring special cooling. A big change from traditional AI setups for video analysis, which rely on GPUs, or graphics processors, that can use hundreds of watts and are too large or power-hungry for edge devices.

Not just Smaller... Smarter

So what's the secret? One of the tricks is how the video image is handled. Instead of downscaling the 4K image, the chip splits it into smaller segments and runs AI inference on each one separately. This allows it to detect small details that might otherwise be lost. At the same time, it analyzes a smaller version of the full image to capture any larger objects that cross between the segments. The results are then merged, giving a full, coherent view of what’s happening in the scene.

More and Better Processing

Although this increases the total amount of processing needed, NTT has developed a custom AI engine to keep it efficient. It uses methods such as inter-frame correlation, which finds similarities between video frames to reduce redundant calculations, and dynamic bit-precision control, which adjusts how much detail is processed depending on what’s needed. All to help the chip keep up with real-time video without gobbling up too much power.

Superior Monitoring

And the result? A chip that makes high-definition AI practical in places where it wasn’t before. For example, drones equipped with the new LSI can now perform inspections from 150 meters above the ground, compared to the 30-meter ceiling of earlier models. They can safely fly beyond the operator’s line of sight and still detect pedestrians or cars in their path, reducing labor demands and improving safety.

The chip also has potential for use in other types of application, such as wildlife monitoring in remote areas, surveillance in residential settings where privacy is a concern, and safety compliance detection on industrial sites. Or how about smart farming, where edge devices help monitor crops and livestock across large fields?

... Coming Sooner Than You Think

Commercial rollout is expected within the 2025 fiscal year through NTT Innovative Devices Corporation. In the meantime, NTT intends for its research to keep on expanding the chip’s compatibility with other AI models, which could broaden its applications even further.

As AI becomes more embedded in everyday operations, the hardware that supports it needs to evolve too. For real-time video analysis on the edge, just having more clever algorithms won't be enough; the systems running them also have to be small, fast, and efficient enough to go wherever the data is being captured. NTT’s new chip is a big step in that direction.

Innovating a Sustainable Future for People and Planet

For further information, please see this link:
https://group.ntt/en/newsrelease/2025/04/11/250411a.html

If you have any questions on the content of this article, please contact:
Public Relations
NTT IOWN Integrated Innovation Center
https://tools.group.ntt/en/rd/contact/index.php

Daniel O'Connor

Daniel O'Connor joined the NTT Group in 1999 when he began work as the Public Relations Manager of NTT Europe. While in London, he liaised with the local press, created the company's intranet site, wrote technical copy for industry magazines and managed exhibition stands from initial design to finished displays.

Later seconded to the headquarters of NTT Communications in Tokyo, he contributed to the company's first-ever winning of global telecoms awards and the digitalisation of internal company information exchange.

Since 2015 Daniel has created content for the Group's Global Leadership Institute, the One NTT Network and is currently working with NTT R&D teams to grow public understanding of the cutting-edge research undertaken by the NTT Group.