Qualcomm AI chips target IoT computer vision and machine learning

By Aalyia Shaukat, contributing writer

Qualcomm has unveiled two new systems-on-chip (SoCs) designed to serve smart visual applications for IoT platforms. While most IoT-based visual processing technologies repurpose processors, the QCS603  and QCS605 are tailor-made with advanced camera processing software, machine learning, and computer vision software development kits (SDKs) for smart cameras. The potential of these densely packed 10-nm FinFET-based chips can range from tracking automated equipment in Industrial IoT (IIoT) to facial recognition in smart city and home automation applications. They are the first SoCs of this type in Qualcomm’s Vision Intelligence Platform.

The SoCs support both 2 x 2 802.11ac Wi-Fi with MU-MIMO and Bluetooth 5.1 connectivity as well as six satellite navigation options including GPS, GLONASS, Beidou, Galileo, QZSS, and SBAS. With Wi-Fi and Bluetooth being the most prolific choice for non-cellular short-range communications, this chip can utilize these technologies for many commercial and consumer IoT applications. There is also support for up to 4K video at 60 frames per second (fps), or 5.7K at 30 fps with the chip’s dual 16-megapixel (MP) image sensor processors. Multiple simultaneous streams can also be enabled at lower resolutions. For significant digital signal processing (DSP) off the cloud, the chips carry an 8-core 64-bit CPU with clock speeds up to 2.5 GHz.

Qualcomm’s new AI SoCs feature in this 360 camera reference design.

“Our goal is to make IoT devices significantly smarter as we help customers bring powerful on-device intelligence, camera processing and security,” says Joseph Bousaba, vice president, product management, in their recent press release. “AI is already enabling cameras with object detection, tracking, classification and facial recognition, robots that avoid obstacles autonomously, and action cameras that learn and generate a video summary of your latest adventure, but this is really just the beginning.”

All of this hardware is really meant to be easily integrated with Qualcomm’s native third-generation AI Engine, the Qualcomm Snapdragon. This neural processing software includes multiple commonly used analysis, optimization, and debugging tools to allow companies to more simply port already trained networks into the platform. The image signal processors (ISPs) fit within the relatively nascent market of edge-processing, in which all of the complex data processing can occur at the device as opposed to on the web through a high-throughput IP link. This essentially saves time and increases the level of security because the data does not have to go through too many “hands.” This would especially be useful in IIoT applications, in which response times of equipment is on a millisecond scale. Moreover, smart facial recognition that is now used to track individuals in cities and towns can actually be applied on a smaller scale for home security.