Ambarella launches 4K, 5nm Edge AI SoC for mainstream security cameras

2 mins read

Ambarella, an edge AI semiconductor company, has announced the launch of the CV72S 4K, 5nm AI vision system-on-chip (SoC).

Intended for mainstream professional security cameras, the SoC is built on the company’s CVflow 3.0 AI architecture delivering what the company claims is the security industry’s highest AI performance per watt and fusion of radar and camera data, delivering improved night-time and all-weather AI perception.

The SoC’s dedicated AI hardware can efficiently run the latest transformer neural networks (NNs), which now outperform convolutional NNs for many vision tasks. Additionally, the CV72S offers 6x greater AI performance than its predecessor, enabling it to also run Ambarella’s AISP for neural network-enhanced 4K, long-range colour night vision and HDR at very low lux levels with minimal noise and no external illumination, while leaving plenty of headroom for additional, concurrent NNs (e.g., person tracking and mask detection).

As a result of the architecture’s efficiencies, combined with 5nm process technology, the CV72S consumes less than 3W of power, offering what is said to be the mainstream security market’s highest AI performance per watt.

“Ambarella’s CV72S fits perfectly into the ongoing trend for security cameras to integrate more and more AI capabilities at the edge via camera processors, while maintaining a low power and thermal budget,” commented Florian Domengie, Senior Technology & Market Analyst, Imaging at Yole Intelligence, part of Yole Group. “The growth of infrastructure cameras powered by a processor with advanced AI acceleration is expected to increase significantly in the coming years, with a CAGR22-27 of 23%.”

“To take their mainstream professional security cameras to the next level, designers are looking for a solution that provides more AI performance, better image quality and the ability to cover wider areas, across weather and lighting conditions without increasing power budgets,” said Fermi Wang, President and CEO of Ambarella. “We are accelerating their innovation by offering our latest CVflow 3.0 architecture and 5nm process technology in the new CV72S SoC, which was purpose built for the mainstream security market using our algorithm-first approach and in-depth knowledge of vision processing, radar and AI.”

Ambarella said that it designed the CV72S to also enable higher-end cameras with advanced 16MP30 fisheye dewarping and 4x 5MP30 multi-imager AI capabilities. For single-imager cameras, the CV72S supports 4KP60 encoding for AVC and HEVC - a 2x performance increase over its predecessor.

These features are intended for a broad range of mainstream security cameras, including smart city and traffic applications, as well as monitoring crowded areas. Additionally, the CV72S provides rich details for human viewing while enabling advanced video analytics such as long-distance object detection and license plate recognition.

Ambarella’s in-house Oculii virtual aperture imaging (VAI) AI radar technology enabled the company to integrate hardware acceleration into the CV72S for fusion with camera data, as well as dynamic load switching between radar and cameras. Ambarella’s Oculii technology features 10-100x better radar resolution than other radar solutions and these radar capabilities are particularly important for applications such as perimeter security and night-time monitoring, as well as any system that needs to operate in rain, snow or fog.

Other features of the CV72S SoC include a 2x increase in CPU performance over the prior generation, via dual Arm Cortex-A76 1.6GHz cores. It also features 2x the DRAM bandwidth, with support for 32-bit LPDDR4x/LPDDR5/LPDDR5x DRAM. Additionally, this SoC integrates hardware security to prevent hacking, including secure boot, OTP and Arm TrustZone technology. The CV72S also has new high-speed PCIe and USB 3.2 interfaces, enabling more complex, multi-chip security system designs.

Ambarella’s CVflow AI development platform provides a full set of tools for easy porting and supports all common CV frameworks.