Ambarella’s new, low-power chip  will enable AI processing on the device itself rather than the cloud.

Ambarella, Inc. (NASDAQ: AMBA) has unveiled its new CV28M SoC (System on Camera) chip for intelligent applications required to run “on edge.” These include smart home security, retail monitoring, consumer robotics, occupancy monitoring and in smart cars. (VB)

Ambarella’s CV Flow engine

In 2015, Ambarella decided to leverage its experience in image processing and compression to develop a fast, intelligent, and powerful system capable of running the latest neural networks with minimum effort, high precision, and low power consumption.

In a line, the company wanted to develop a low-power consuming, dedicated AI accelerator for embedded systems at the edge that could give cloud-based systems a run for their money in terms of performance.

What followed was the company’s range of computer vision chips known as CVflow.

Ambarella’s CV28M SoC

• CVflow Computer Vision Engine
• Combines state-of-the-art computer vision technology with image processing
• Single, low-power design.
• Fabricated using advanced 10 nm process technology
• Provides deep neural network (DNN) computer vision processing and 4KP30 video encoding
• Runs a multitude of computer vision applications on the edge
• AVC and HEVC video formats
• Streaming high-resolution video at low bit rates
• Next-generation image signal processor (ISP) to deliver outstanding imaging in low-light conditions
• High dynamic range (HDR) processing extracts maximum image detail in high-contrast scenes
• Advanced security features for on-device physical security
• Comes with a software development kit that offers a complete set of tools
• 64-bit dual-core Arm® Cortex®-A53 CPU up to 1 GHz

CV28M suited for consumer video applications with better privacy than cloud

According to VB, the CV28M consumes only 500 milliwatts of power, enabling wide use of “on edge” applications.

“The significance of this chip for the consumer video market is its ability to process more on the camera,” Blake Kosak, an analyst at market researcher Omdia, said in an email to VentureBeat. “Today, most consumer security cameras send video to the cloud for processing, which is expensive for the camera manufacturer. Chips that allow for more edge-based processing will benefit brands that are looking to avoid charging customers monthly for cloud storage and advanced analytics.”

According to Kosak, rising concerns of privacy, particularly in Europe and North America, also boost the case for on-edge processing because it reduces the need to send video data to a remote cloud server.

Related Story:  Intel First To Deploy AI “On Edge” In Space

Image Credit for CV28M Chip: Ambarella