NTT Corp. has unveiled a real-time AI inference large-scale integration (LSI) processing chip designed for edge and power-constrained terminal deployments that require compression of high-definition 4K resolution at 30 frames per second.
In these applications, AI devices are typically limited to power consumption at an order of magnitude lower than that of GPUs used in AI services. The LSI chip overcomes this through its AI inference engine that reduces computational complexity while enabling detection accuracy and improving efficiency using interframe correlation and dynamic bit-precision control, NTT said.
NTT gave an example of a drone equipped with the LSI chip that can detect individuals or objects from up to 492 feet above ground. A conventional AI video inferencing chip would allow for only 98 feet by comparison, NTT said. This could lead to drones being used more for infrastructure inspection for operations beyond the visual line of sight (BVLOS).
The company plans to develop the LSI technologies to expand supported inference models and use cases. NTT aims to commercialize the LSI this year.
NTT demonstrated the LSI chip at Upgrade 2025, taking place this week in San Francisco.
