Arrow announced a pair of 96Boards CE Extended SBCs that run Yocto with Linux 4.14: an AI-ML Board with a quad -A35 i.MX8X and a Thor96 SBC with a quad -A53 i.MX8M and support for multi-channel audio, ZigBee, and Thread.

Like the DragonBoard 820c, the new AI-ML Board and Thor96 SBCs are collaborations between Arrow and eInfochips that use the 100 x 85mm 96Boards CE Extended form factor. Arrow also recently announced a CE Extended collaboration with manufacturer DH Electronics: an Avenger96 SBC with ST’s new STM32MP1 SoC.

Thor96
(click image to enlarge)

The AI-ML Board and higher-end Thor96 are the first 96Boards SBCs built around NXP’s 64-bit i.MX8 family of processors. In addition, the AI-ML Board is the first i.MX8X SBC of any kind that we’ve encountered.

They both feature a Linux stack built around Yocto Project code and Linux 4.14. A Yocto stack “allows designers to customize and build their own Linux distro best suited for their system,” says Arrow. “This allows for improved speed by removing unneeded software, provides a custom solution for each project and can even help to improve security by minimizing the number of possible entry points into the system.” (What Arrow doesn’t mention is that Yocto development requires a fair amount of expertise, but for the ultimate in flexibility there’s nothing better.)

Like most 96Boards SBCs, Arrow’s SBCs don’t have a lot of real-world ports. However, there are a variety of third-party Mezzanine add-on boards that plug into the 96Boards CE standard 40-pin low-power or 60-pin high-power GPIO connectors. Since all officially sanctioned 96Boards have open specifications, you can also design your own Mezzanines.

Arrow’s two new SBCs are not yet listed on the 96Boards website. However, it’s very likely that when the boards ship around the beginning of June, the SBCs will win the 96Boards seal of approval.

 
AI-ML Board

The AI-ML Board taps the lowest-end i.MX8X member of the i.MX8 family. The SoC supplies up to 4x 64-bit Cortex-A35 cores clocked to 1.2GHz. The power-efficient i.MX8X also integrates a Cortex-M4F core and Tensilica HiFi 4 DSP.

i.MX8X security features include HAB, SRTC, TrustZone, SHA-256, 3DES, ARC4, MD5, Tamper and Inline Enc Engine. There are several i.MX8X compute modules, including Kontron’s SMARC-sAMX8X.

AI-ML Board and GPIO details
(click images to enlarge)

The AI-ML Board uses the quad-core QuadXPlus i.MX8X model with 4-shader Vivante GPU and multi-format VPU. The SBC is targeted at robotics, drones, local AI/machine learning systems, monitoring systems, and 4K signage.

The references to artificial intelligence perhaps relates the potentiality of the DSP, although Arrow offers no details. In any case, the i.MX8X looks like a good choice for camera-oriented robotics due to the combination of low power consumption, DSP, real-time Cortex-M4F, extensive low-power I/O, and a GPU with H.265 4K@30 decode and H.264 1080@60 encode. Arrow touts both of its boards for drones, stating: “The ability to read camera information and encode on the fly could easily allow for real-time processing of the environment for obstacle avoidance, while the Cortex-MF4 can be used for reading sensor data such as orientation and acceleration.”

The AI-ML Board ships with a 2GB industrial-temp LPDDR4 module. Although it’s not listed in the specs, the image shows a microSD slot for storage. Arrow also fails to mention the Ethernet port, but it’s clearly visible. It’s unclear if it’s Fast Ethernet or Gigabit Ethernet, which is supported by the i.MX8X. There’s also a WiFi-ac module with Bluetooth.

The AI-ML Board is further equipped with an HDMI port, dual MIPI-DSI interfaces, 2x USB 3.0 ports, and a micro-USB. Camera boards can be connected via the high-speed GPIO connector. (See the diagram above for expansion GPIO details.)

Specifications listed for the AI-ML Board include:

• Processor – NXP i.MX8X QuadXPlus (4x -A35 @ 1.2GHz); Vivante GT7000Lite GPU (4x Vec4 shaders with 16 EUs); Vulkan 2D Blit engine with OpenGL; multi-format VPU; Tensilica HiFi 4 DSP; 266MHz Cortex-M4F
• Memory/storage — 2GB LPDDR4 @ 1,600MHz (industrial-temp Micron module); microSD slot
• Wireless — 802.11ac and dual-mode Bluetooth (Laird/LSR 450-0169C module with Cypress WiFi)
• Networking — Ethernet port
• Other I/O:
  • HDMI port at up to 4K@30 (H.265)
  • 2x MIPI-DSI
  • 2x USB 3.0 host ports
  • Micro-USB port
• Expansion — 40- low-power and 60-pin high-power GPIO connectors (96Boards CE)
• Dimensions — 100 x 85mm (96Boards CE Extended)
• Operating system — Yocto Project with Linux 4.14

 
Thor96

The Thor96 taps the up to quad-core version of the 64-bit, up to 1.5GHz Cortex-A53 i.MX8M. The SoC also includes a Cortex-M4F real-time core and a 4K-ready Vivante GC7000Lite/GC7000VLX GPU. Other i.MX8M-based SBCs include Estone Technology’s EMB-2238.

The board has the same 2GB RAM module as the AI-ML Board. It provides Cypress WiFi-ac and BT via a different module, and its LAN port is listed as GbE. The HDMI port goes to 4K@60 and a camera board connected to the 60-pin GPIO connector is supported with 1080p@60 encoding. Unlike the AI-ML, the board supports 2x MIPI-CSI cameras.

The Thor96 provides dual MIPI-DSI interfaces, as well as “DSI to HDMI,” suggesting you could rig a second 4K HDMI port. Like the AI-ML Board, it provides 2x USB 3.0 ports and a micro-USB port.

The Thor96 adds a Silicon Labs MGM111 Thread/Zigbee module, which enhances the board’s recommended application for building automation gizmos. In addition, there’s an A2B module from Analog Devices that supports low latency digital audio, speakers, and multiple microphone arrays. States Arrow: “Running both power and multi-channel digital audio over a single unshielded twisted pair of cables reduces weight, cost and complexity while enabling advanced voice recognition and active noise cancelation applications.”

As with the AI-ML, there were no details on power inputs or operating range. The Thor96 image suggests that a fan option is available.

Specifications listed for the Thor96 include:

• Processor – NXP i.MX8M (4x Cortex-A53 @ 1.5GHz); Vivante GC7000Lite/GC7000VLX for OpenGL/ES 3.1, OpenGL 3.0, Vulkan, OpenCL 1.2 GPU; Vulkan High-Performance 2D Blit engine; Cortex-M4 @ 266MHz
• Memory/storage — 2GB LPDDR4 @ 1,600MHz (industrial-temp Micron module); microSD slot
• Wireless:
  • 802.11ac and dual-mode Bluetooth (Murata Sterling-LWB5 module with Cypress WiFi)
  • Silicon Labs MGM111 Thread/Zigbee module
• Networking — Gigabit Ethernet port
• Other I/O:
  • HDMI port at up to 4K@30 (H.265)
  • 2x MIPI-DSI
  • Multi-channel digital audio I/O via Analog Devices A2B module
  • 2x USB 3.0 host ports
  • Micro-USB port
• Expansion — 40- low-power and 60-pin high-power GPIO connectors (96Boards CE)
• Dimensions — 100 x 85mm (96Boards CE Extended)
• Operating system — Yocto Project with Linux 4.14

 
Further information

The AI-ML Board (IMX8_AI_ML) and Thor96 (I.IMX8_THOR96) SBCs will be available in single quantities starting around June 1. (Update on Dec. 26: The AI-ML is still not in stock and the Thor96 is available for $368.25 via the link below.) More information may be found in the Arrow announcement and its AI-ML Board and Thor96 shopping pages. eInfochips product pages are also provided for the AI-ML Board and Thor96.