Ambarella and Realtek have developed new smart-camera processors that can recognize faces, read license plates, and perform other object-recognition tasks on chip. The former is upgrading its chips to support next-generation 8K video, whereas the latter’s RTS3916N aims to increase AI and CV capabilities in low-cost devices.

Ambarella designed its new CV5S and CV52S for high-end surveillance systems in commercial, industrial, smart-transportation, and other venues. These chips employ 5nm technology, enabling much greater video bandwidth than the predecessor 10nm CV2-series. At 30fps, the CV5S can identify objects in video streams comprising up to 33 megapixels (one gigapixel per second) while consuming less than 5W (maximum). Typical power consumption is 2–4W, depending on the workload. The CV52S implements mostly the same architecture as the CV5S, but the company optimized it for single-sensor security cameras that stream 720Mpixels per second. Ambarella expects to sample the CV5S and CV52S in October, probably leading to volume production in 2Q22.

Realtek optimized the RTS3916N for mainstream consumer and IoT devices, such as smart doorbells and webcams. At 30fps, the processor can handle a single 5Mpixel image sensor, which is a common surveillance-camera format that provides resolution midway between full HD and 4K. Taking advantage of its networking expertise, the company equips the RTS3916N with an integrated Ethernet MAC and PHY, yielding a plug-in Internet Protocol (IP) camera. The chip is now in volume production.