Sipeed Maix-III devkit features AXERA AX620A SoC for AI vision applications

Sipeed Maix-III development kit is composed of a core board equipped with AXERA AX620A AI processor and a carrier board exposing several interfaces such as Ethernet, USB, and camera connectors. resources.

As noted in our earlier article about the AXERA AX620A chip, it is a quad-core Arm Cortex-A7 chip with an NPU offering a maximum of 14.4 TOPS @ INT4 or 3.6TOPSs @ INT8, and as such, this development board mainly targets AI vision applications. The board is an upgrade to the earlier Speed Maix II with an Allwinner V831 processor.

Thanks to the low power consumption of the AXERA AX620A SoC, the Sipeed Maix-III board can be powered directly through its USB port as the system consumers 5W (5V@1A) with the carrier board, the system-on-module, WiFi, Ethernet, a camera, and a 5-inch display. Using the display does not require am additional heatsink or fan, nor does it require an external 2A power supply like some other platforms.

Sipeed Maix III specifications

Core Board
- CPU -AXERA AX620A with quad-core Arm Cortex-A7 @ 1.0 GHz, 32KB L1 I-cache + 32KB L1 D-cache per core, 256KB L2 cache, FPU and NEON
- NPU – 14.4 TOPS @ INT4, 3.6 TOPS @ INT8 with support for Imagenet, AlexNet, VGG, ResNet, GoogLeNet, Faster R-CNN, SSD, FPN, Yolo V3, and other neural networks.
- ISP – Proton AI-ISP up to 4Kp30
- System Memory – 2GB LPDDR4X, up to 3733Mbps
- Storage – Optional 16GB eMMC flash (default is MicroSD card boot)
- Video Encoding
  - H.264/H.265 video encoder 4Kp30, multi-stream real-time encoding capability 4K @ 30fps + 1080p @ 30fps + 720p @ 30fps
  - JPEG Baseline encoder 4Kp30
- Video Decoding
  - H.264 video decoder 1080p60
  - JPEG Baseline decoder 1080p60
- Networking – Ethernet supporting dual RGMII/RMII interface mode
- Host connection – 260-pin SODIMM edge connector
Carrier board:
- Storage – MicroSD card slot for OS.
- Display I/F – 4-lane MIPI DSI
- Camera Input – Support for up to 3 cameras via 1x 4-lane MIPI CSI + 2x 2-lane MIPI CSI
- Audio
  - 3.5mm headphone jack with stereo input and MIC input
  - 2x on-board MEMS microphone supporting noise cancellation
- Networking – Gigabit Ethernet port and 2.4GHz WiFi with onboard antenna
- USB – 1x USB 2.0 OTG/host Type-C port, 1x USB-UART Type-C port for serial port
- Misc – 1x User Button, 3x System Button, 1x RTC Clock, 2x LEDs

As can be seen from the official documentation, the Sipeed Maix III supports most of the common AI models. Model resources are preset in the development board system, and ipeed and AXERA provide a large number of deep learning open-source algorithm models that can be used directly on the development board. These models can be found in the MaixHub model library and AXERA-TECH’s Github repository.

Sipeed provides a Debian 11-based Linux operating system for the development board booting from the MicroSD card. A Python API can be used to develop AI applications, but it can only be retreived through a link on Baidu at this time. Sipeed also offers the source code for the BSP, the API documentation, a developer guide for the SDK, and other instructions, all of which can be found on a Sipeed`s GitHub repository. Some other documents such as the AXERA datasheet and the schematics can be downloaded from Sipeed’s website.

There are three different bundles that include the Maix-III core board available on Aliexpress. I suggest you buy the full-featured “Full Board Package” package to fully make use of the board’s capabilities. The package includes the Maxi-III core board, the AxPi carrier board, a GC4653 cameras sensor, a 3D printed frame, a 5-inch MIPI DSI display, a 32GB MicroSD card, a mini tripod, a USB-C OTG adapter, and a USB data cable for $95 excluding shipping. It’s also possible to purchase the OS04A10 instead of the GC4653 camera sensor for better low-light/night vision performance.

Leo Zachary

Leo became a part-time writer for CNX Software in 2022 and works as a software engineer in an embedded software company in China. He’s been passionate about embedded development and the Internet of Things for many years, with extensive industry experience, and a new and unique vision!