All News | Boards | Chips | Devices | Software | Archive | About | Contact | Subscribe
Follow LinuxGizmos:
Twitter Facebook Pinterest RSS feed
*   get email updates   *

Linux-driven 4K camera SoC targets the low-power AI edge

Nov 3, 2020 — by Eric Brown 2,090 views

Ambarella has begun sampling a 10nm “CV28M” camera SoC for edge AI that runs Linux on dual 1GHz Cortex-A53 cores and offers CVflow CNN processing, a 320MP/s ISP, 4Kp30 encoding, and security features.

Five years have passed since we covered a new Ambarella camera SoC, which is a shame since it’s so much fun saying “Ambarella.” Since the announcement of its HD-ready, Cortex-A9 based S2Lm, we have mentioned the company in passing for its part in a Linux- and Jetson-driven Teal One drone, which uses Ambarella’s [email protected] ready, quad -A53 CV2.

Now Ambarella has launched the CV28M, the latest in its CVflow family of AI-enabled computer vision processors. Like the other Ambarella SoCs, it is available with a Linux SDK and evaluation kit.


Equipped with 2x up to 1GHz Cortex-A53 cores, the CV8M is optimized for image processing AI applications on the edge, offering local processing of images without the security, privacy, connectivity, and latency issues involved with cloud processing. Like the recent, quad -A53 CV25S (PDF), this is a 10nm fabricated chip. Among other differences, it offers lower power consumption with better support for battery-operated cameras, robots, and other mobile or low-power edge devices.


Applications include smart home security, retail monitoring, consumer robotics, and occupancy monitoring. On a robot, the CV28M can be connected to sensors including visible, structured light, and time-of-flight (ToF) to capture, and then process data required for navigation.

The CVflow Computer Vision Engine offers CNN/DNN inference acceleration for detection and classification analytics including people counting, face recognition, license plate recognition, and human, pet, and vehicle classification. You can port to frameworks including Caffe, TensorFlow, and ONNX.

CV28M block diagram
(click image to enlarge)

The CV28M offers H.265 and H.264 encoding and enables 4Kp30 AVC and HEVC encoding with multi-stream support. AI-based SmartAVC and SmartHEVC intelligent rate controls can “optimize image quality while reducing video storage and network bandwidth requirements,” says Ambarella.

The CV28M is equipped with a 320MP/s ISP with high dynamic range (HDR) and advanced low-light processing. Special effects include image stabilization, lens distortion correction (LDC), hardware dewarping, 3D motion-compensated temporal filtering (MCTF), and more. Security features include secure boot with TrustZone, TRNG, OTP, DRAM scrambling, and virtualization.

The SoC supports triple-sensor video input with high-speed SLVS, MIPI-CSI, and LVCMOS interfaces. There is a GbE controller and support for USB 2.0 host and device, 2x SD/SDXC controllers, MIPI-DSI, I2S, and DMIC.

Other I/Os include SSI/SPI, I2C, UART, GPIO, PWM, IR, and ADC. The 11 x 12mm, 0.65-pitch BGA processor supports -25 to 85°C temperatures.

The CV28M shares a common Linux SDK and computer vision (CV) tools with Ambarella’s CV25, CV22, and CV2 CVflow SoCs. The SDK is built around a 64-bit Linux 4.14 kernel. CV tools include a compiler, debugger, porting tools, and guidelines for CNN optimization. There is also a reference application with C source code.

Ambarella supplied only limited specs for its CV28M Camera Development Platform. The mainboard connects to a sensor/lens daughterboard with support for sensors from vendors including Sony, ON Semi, and Omnivision. The eval kit is available with BOM, schematics, and layout files.

Further information

CV28M samples are available now at an undisclosed price. More information may be found in Ambarella’s announcement and product page (PDF).


(advertise here)

Print Friendly, PDF & Email

Please comment here...