Industry's first SoM to include onboard hardware AI accelerator chip

Virtium Embedded Artists has launched the iMX8M Mini DX-M1, a SoM which integrates a quad-core application processor and a 25 TOPS AI hardware accelerator chip with associated memory on a board with a compact 82mm x 50mm footprint.

By integrating a DEEPX DX-M1 AI accelerator, an advanced NPU, into the SoM, the company eliminates the requirement for embedded device manufacturers that want to implement AI functions to plug a discrete AI processor module into their system. This provides multiple benefits to OEMs and designers, including board space savings, simplified systems design, streamlined BoM, and faster time-to-market.

The new SoM is based on the i.MX 8M Mini from NXP Semiconductors, an application processor which features four 1.6GHz/1.8GHz Arm Cortex-A53 CPU cores backed by 2GB of LPDDR4 memory, as well as a 400MHz Cortex-M4 controller core. The i.MX 8M Mini provides robust support for video and imaging applications, thanks to its video engine with a 1080p codec, 2D/3D graphics engine, and a four-lane MIPI-DSI interface, as well as a Gigabit Ethernet interface.

Working in tandem with the DEEPX DX-M1 NPU, the i.MX 8M provides the image processing capability needed to support AI-enabled vision systems used in popular and emerging vision applications. These include:

• Drones
• Security and surveillance
• Automated inspection and monitoring
• Transportation
• MedTech
• AgTech

The DX-M1, a high-performance AI accelerator, delivers 25 TOPS of throughput at an average power consumption of 5W, making the SoM an excellent choice for power-constrained edge computing applications. In this new Virtium Embedded Artists implementation, the SoM provides 4GB of LPDDR5 memory for the AI processor accessed via a 64-bit, four-channel data bus. This allows the DX-M1 to run multiple AI models concurrently without performance degradation.

Anders Rosvall, managing director of Virtium Embedded Artists, said: "By bringing the iMX8M Mini DX-M1 to market, Virtium Embedded Artists is meeting surging demand from edge and IoT device manufacturers for a ready-made hardware platform for low-power vision AI processing and embedded control. Now OEMs can get to market faster and more easily with new product designs which offer the extra value that vision AI brings to industrial, mobility, security and edge computing applications."

The company supplies the iMX8M Mini DX-M1 SOM in two versions: one with the DEEPX AI accelerator and i.MX 8M Mini processor, and one with the i.MX 8M Mini alone. This allows OEMs to easily scale up embedded computer systems based on the i.MX 8M Mini to also provide vision AI capabilities without altering its carrier board design.

Each of the two versions of the SOM is available with a 1.8GHz i.MX 8M Mini, for operation at temperatures between 0C and 70C, or with a 1.6GHz i.MX 8M Mini with an operating temperature range of -40C to 85C.

Alongside the iMX8M Mini DX-M1 hardware, the company provides a full technical support service. It also supplies the DEEPX iM8 Mini AI Kit, which includes reference design implementations for key interfaces. The company offers pre-designed standard carrier boards to ease integration into customers' designs.