There are already plenty of NXP i.MX 8M Mini boards and system-on-modules, but so far none of those included a neural network accelerator. But that’s now an option as SolidRun and Gyrfalcon teamed up to design SolidRun i.MX 8M Mini system-on-module that combines NXP i.MX 8M Mini processor with Gyrfalcon Lightspeeur 2803S NPU (Neural Processing Unit) in order to speed up AI workloads at the edge.
- SoC (one or the other)
- NXP i.MX8M Mini S single core Arm Cortex-A53 processor up to 1.8 GHz, Arm Cortex-M4 core up to 400MHz, GC NanoUltra 3D GPU + GC320 2D GPU
- NXP i.MX8M Mini D dual-core Arm Cortex-A53 processor up to 1.8 GHz, Arm Cortex-M4 core up to 400MHz, GC NanoUltra 3D GPU + GC320 2D GPU
- NXP i.MX8M Mini Q quad-core Arm Cortex-A53 processor up to 1.8 GHz, Arm Cortex-M4 core up to 400MHz, GC NanoUltra 3D GPU + GC320 2D GPU
- NPU – Gyrfalcon Technology Lightspeeur 2803S artificial intelligence accelerator
- System Memory
- i.MX8M Mini Solo – Up to 3GB LPDDR4-3000
- i.MX8M Mini Dual/Quad – Up to 4GB LPDDR4-3000
- Storage – Optional eMMC flash, and optional QSPI-NOR flash
- Decode – 1080p60 VP9, VP8, HEVC/H.265 decoder, AVC/H.264
- Encode – 1080p60 AVC/H.264 encoder, VP8 encoder
- Atheros Gigabit Ethernet transceiver
- Optional wireless module with 802.11b/g/n WiFi 4 and Bluetooth 4.2/5.0
- 3x Hirose DF40 board-to-board connectors exposing the following interfaces:
- Storage – NOR Flash, SD/microSD, PCIe SSD
- Display I/F – MIPI DSI up to 1080p60
- Camera I/F – 1x 4-lane MIPI CSI-2
- USB – 2x USB 2.0 interfaces
- Serial – 2x RTS/CTS/Rx/Tx interfaces, and 1x Tx/Rx interface
- Audio – 20 channels 32bits @ 384khz, DSD512, SPDIF TX&RX, 8x PDM DMIC channel
- 1x PCIe 2.0
- 2x I2C, 1x SPI, 4x PWM, 75 GPIO
- Debugging – Test point header for JTAG
- Misc – RTC
- Power Supply – Main voltage: 5V, I/O voltage: 3.3V, SOM supply: 3.3V/1A
- Dimensions – 47 x 30 mm
- Temperature Range – Commercial & industrial
Gyrfalcon AI accelerator delivers up to 24 TOPS per watt, and 16.8 TOPS @ 300 MHz with only 700mW power consumption. The chip supports AI frameworks such as TensorFlow, Caffe, PyTorch including pre-trained AI models for ResNet, MobilNet, and ShiftNet.
SolidRun will provide support for both Linux and Android running on the NXP processor, as well as HummingBoard Pulse carrier board which was announced last year with USB 3.0 ports, Dual Ethernet, and more. Some limited documentation (e.g. schematics) is already available on the developer’s website, but more information should be posted soon.
SolidRun i.MX 8M Mini SOM is available through SolidRun or Arrow for $56 and up according to the press release, but it is still listed as “Coming Soon” on the product page. For reference, the HummingBoard Pulse carrier board can be purchased for $182 and up with one of the previous i.MX 8M modules without AI Accelerator.