SolidRun Partners with Hailo for a 20 TOPS Hailo-15H-Powered Edge AI System-on-Module

Embedded hardware specialist SolidRun has announced the release of a system-on-module (SOM) built around the Hailo-15 neural processing unit (NPU) — bringing up to 20 tera-operations per second (TOPS) to bear on edge artificial intelligence (edge AI) and on-device machine learning workloads.

“Our vision at SolidRun has always been to simplify the complex world of embedded systems and empower engineers to innovate without constraints," claims Atai Ziv, PhD, SolidRun's chief executive officer. "Collaborating with Hailo to develop the Hailo-15H SOM, we've introduced an integrated solution that achieves just that. This market-ready solution seamlessly combines cutting-edge AI inferencing capabilities with ease of integration, ultimately reducing engineering time and costs. Most importantly, it enables developers and engineers to devote their energies entirely to realizing their product vision."

"SolidRun’s SOM based on our Hailo-15H SOC represents a significant milestone for in-camera edge AI processing and provides engineers with a compact solution that delivers unparalleled performance and energy efficiency for a wide range of AI vision applications," adds Hailo's Yaron Ofer. "Working with SolidRun, we’re able to empower customers to quickly unlock the full potential of AI at the edge, and drive innovation across countless industries."

Hailo unveiled the Hailo-15 back in March last year, promising up to 20 tera-operations per second (TOPS) of compute, part-dependent, for on-device machine learning and edge artificial intelligence (edge AI) workloads — delivering the ability, according to the company's own testing, to run the ResNet-50 model on-device at 700 frames per second. SolidRun's last Hailo-equipped device, meanwhile, was the Hummingboard 8P Edge AI combining an NXP i.MX 8M Plus with the older Hailo-8.

SolidRun has confirmed core variants of the Hailo-15 SOM: the Hailo-15M SOM uses a lower-end version of the Hailo-15 chip offering 11 TOPS of compute, while the Hailo-15H SOM features the top-end 20 TOPS version. Both models include a quad-core Arm Cortex-A53 application processor running at 1.3GHz, up to 8GB of LPDDR4 memory, and eMMC storage from 8GB to 256GB alongside a 32Mb QSPI flash module. For network connectivity there's an on-board dual-band 802.11ac and Bluetooth 5.0 radio alongside gigabit Ethernet, with USB 3.1 for external hardware.

The modules include a vision subsystem with a dual-camera 12 megapixel image signal processor (ISP) with hardware H.264/H.265 encoder supporting high-dynamic range (HDR) video and up to 4k30 footage, two MIPI Camera Serial Interface (CSI) four-lane ports, and compatibility with the Sony IMX334 and IMX678 image sensors. There's a four-name MIPI Display Serial Interface (DSI) for video output, Secure Digital Input/Output (SDIO) support, audio capabilities, and SPI, I2C, I2S, and UART buses alongside general-purpose input/output (GPIO) pins brought out to three Hirose DF40 connectors on the board's underside.

Initially, the Hailo-15 SOMs are only available to order as bare modules; the company has confirmed plans to release a Hummingboard IIoT carrier featuring the modules and bringing out all major features. Pricing for the carrier bundles and Hailo-15M SOM has yet to be confirmed, with orders now open for the Hailo-15H SOM starting at $149 with 4GB of LPDDR4 RAM and 32GB of eMMC storage.

More information is available on the SolidRun website, while the company will be demonstrating the devices at Embedded World in Nuremberg, Germany April 9th-11th, Hall 1 Booth #126, and at ISC West in Las Vegas, April 10th-12th, Booth #31065.