NTT Breakthrough: Real-Time 4K AI Inference with Ultra-Low Power
TOKYO, JAPAN, 11TH APRIL , 2025 – NTT Corporation has developed an AI inference LSI (Large Scale Integration) chip for the real-time processing of ultra-high-definition 4K video on power-sensitive edge and terminal devices.
This represents a significant advance in real-time AI applications in areas such as drone surveillance, infrastructure inspections, and crowd analytics.
The chip resolves fundamental problems with video AI. In the past, the 4K video had to be down-converted before AI inference could be performed on efficient power hardware, thereby restricting the distance to which smaller objects could be detected.
NTT’s LSI removes this bottleneck to allow full 4K inference at 30 fps with power consumption of less than 20 W.
The LSI hybridizes, splitting 4K frames to detect smaller objects while analyzing a smaller full-frame image for larger target detection spanning several sections. Merging is carried out for correct detection.
The NTT-original AI inference engine manages optimization and real-time operation under high computational load by inter-frame correlation and dynamic bit-precision control.
The technology, paired with drones, allows for BVLOS navigation and infrastructure inspection at a 150-meter altitude above ground visibility – five times the conventional range. This also enhances performance analytics and tracking in public spaces.
NTT Innovative Devices Corporation is targeting LSI commercialization before the end of the fiscal year 2025. The technology has been showcased at Upgrade 2025 in San Francisco, further calling attention to NTT’s dedication to advance innovations for sustainable execution on edge AI computing.
Nissan and Wayve Team Up on Autonomous Driving Breakthrough
Capgemini’s New AI Deal With Google Could Change CX Forever
This UAE-Google Deal Could Change Global Cybersecurity Forever
