We’ve written about SigmaStar SSD201/SSD202 processors several times now. The low-cost dual-core Cortex-A7 processor comes with 64MB or 128MB built-in RAM, and so far has been found in modules, smart displays, an SBC, and a gateway only available from China via stores like Taobao.
But with mainline Linux support progressing nicely, it was only a matter of time until others make a product based on the processor, and M5Stack Unitv2 is an “AI” camera combining SSD202D processor with 128MB DDR3, and a GC2145 2MP camera sensor.
M5Stack UnitV2 specifications:
- SoC – SigmaStar SSD202D dual-core Cortex-A7 processor @ 1.2 GHz with 128MB on-chip DDR3
- Storage – 512MB on-chip NAND flash with around 100MB free space, MicroSD card socket
- Camera
- GC2145 1080p color sensor with USB UVC support
- 68° FoV, 60 cm to ∞ depth of field (DoF)
- Audio – Built-in microphone
- Connectivity – 2.4GHz WIFi 4 up to 150 Mbps
- USB – USB-C port connected to Ethernet USB chip
- Expansion – UART connector
- Misc – Button, fan
- Power Supply – 5V/500mA via USB-C port
- Dimensions – 48 x 24 x 18.5 mm
- Weight – 18 grams
M5Stack UnitV2 AI camera is said to run the “Teeny-Tiny” Linux operating system with SSH enabled. The company has developed AI recognition applications such as Face Recognition, Object Tracking, Color Tracker, Shape Detector, Barcode Detector, and others. While some other SigmaStar processors come with an NPU, AFAIK SSD202 does not include any AI accelerator, so this is all done by software possibly using NEON instructions for acceleration. Custom models can be generated through M5Stack’s V-Training service. The UART interface transmits results for AI detection in JSON format so the information can use on the host.
M5Stack UnitV2 is also said to have a built-in USB LAN, so that it can be connected to a PC through the USB-C port and connect to the network that way, or through WiFi. I’m not sure why they didn’t use the USB interface and Ethernet gadget support like it is done in BeagleBone SBC’s, as adding an Ethernet to USB chip only adds to the BoM costs, for no obvious benefit, unless I missed something.
The AI camera ships with a 50cm USB-C cable, a stand, and a “back brick”. Typical applications include access control, home automation, occupancy monitoring, smart retail, machine vision, warehouse management, and so on.
The SSD202D AI camera is available for pre-order now for $75. It could only be the pre-order price though, as I’ve also received error-filled documentation that says the camera sells for $89.90. Regardless of whether the standard price ends up being $75 or $89.9, that’s quite a steep price compared to the earlier $40 WiFi connected M5StickV AI camera based on Kendryte K210 AI RISC-V processor. The company did not provide any performance number, and I’ll update the post as I get more information. It’s probably best to wait before purchasing/pre-ordering the camera, and there will be a webinar in a little over two weeks, where further details should be provided..
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
64 gb or 128 gb.
When I saw the price I was a bit surprised.
I thought maybe it’d be in the 20 or 30 USD range as modules with the same hardware except the camera are less than 10.
They showed a bunch of stuff like realtime shape detection on Twitter that looked neat. I wonder if some of that is licensed and that’s where the cost is. It looks like they needed a fan too.. the crappy vendor kernel doesn’t use the ondie sensor so it probably can’t throttle the CPU frequency to reduce the heat.
Anyhow, it should be possible to make it all run with mainline eventually. I sent the DT bits for it already.
If only it was a bit cheaper.
chip shortage in full effect right here.