This blog originally ran on the Antmicro website. For more blogs and articles like this, click here.
With the rise of ever more efficient and cost-optimized edge AI technologies, the possibilities for connecting the multitude of smart sensor devices into intelligent networks via gateways are now more abundant than ever. The way we at Antmicro see things, it is only natural to put effort now into opening up meaningful solutions not just in industrial high-tech (smart factories / Industry 4.0), but for the broader consumer-targeted applications of smart home and smart offices – indeed, smart buildings in general.
One of these recent efforts includes a smart hardware module in the widely adopted, unified M.2 form factor that combines a multi-protocol radio transceiver from Nordic Semiconductor with an efficient edge AI accelerator, the Edge TPU from Google’s Coral team. This combination seems to be unique in embedded systems as it adds features that are usually developed and maintained separately on the system integration level.
By releasing an open, AI-boosted, versatile radio module, we can now offer our customers a thorough building block for their products, and address the need for communication between several smart edge AI units. In many such product development projects, the initial proof-of-concept phase can be rather tricky and lengthy, so having the advantage of a complete building block can speed up the design cycle significantly.
Moreover, the Open M.2 Smart IoT Module is a great way for adding new capabilities to our growing array of open hardware baseboards such as the Jetson Nano Baseboard and the Google Coral Baseboard, turning them into smart, open gateways; or Antmicro’s Scalenode to form a smart server room monitoring platform. And with a fully open source and open hardware solution, the products we develop give our customers and their end users true confidence of data privacy and security.
Edge AI in smart homes, offices & cities
While many of our customers require our services to build advanced and AI-enabled machine vision systems for situation awareness, obstacle avoidance or sophisticated surveillance with object detection and tracking (deployed in drones, mobile robots, agricultural machines, fabrication plants, etc.), the same technologies can be successfully used in home and office automation.
In each of these applications, the smart camera needs to communicate with the rest of the IoT system and report its status to the coordinating unit. In most cases nowadays, this is usually implemented with either WiFi or Ethernet. Adding a configurable radio link to the device allows to distribute the video system across multiple agents and synchronize them under an optimized latency. This way, a device equipped with our Open M.2 Smart IoT Module will not only be able to collect data from literally any other IoT-enabled sensor installed on site, but the extra ML capabilities offered by Coral’s edgeTPU can be used to process data locally before sending it out to the cloud. And with more data available, the AI algorithms programmed will be doing a far better job than with each device separately.
It is easy to imagine this approach as valid for even larger configurations, such as entire smart cities with distributed networks of smart, AI-capable cameras and other sensor systems. Indeed, in the case of smart city 24/7 surveillance, anomaly detection becomes even more important than simply recording the video footage itself. Having a permanent, long range and low latency radio connection allows for advanced diagnostics and for passing the triggering information to any modules residing nearby, making them focus on the region of interest to optimize data collection.
In domestic applications, our Open M.2 Smart IoT Module can be used to provide a gateway functionality to the host board in order to collect data from sensors installed in different locations around your house or office. It also allows the smart unit to integrate with already existing infrastructure. So if you are thinking of building a new next-gen, video-enhanced, AI-driven robotic vacuum cleaner, workout assistant, lawn mower or intelligent doorbell – this module, being an off-the-shelf PoC, is a good starting point for a customized product design.
Open M.2 Smart IoT Module at a glance
The Open M.2 Smart IoT Module utilizes the B+M key of the M.2 standard, which is present in most of the modern processing platforms, and exposes a generic PCIe interface.
At a glance, you will see this in the module’s block diagram:
The module is based on a Renesas UPD720202 USB controller connected to the 1x PCIe on the host side. The USB controller handles a Google Coral Accelerator Module and a Nordic nRF52840 multiprotocol radio SoC. This allows for running accelerated calculations within the open source TensorFlow Lite framework, and communicating over Bluetooth, LoRa or Thread at the same time.
Both modules can be configured and controlled from the host platform over USB. Since the Nordic chip supports multiple IO interfaces such as I2C, I2S, SPI, PDM, it is possible to expand the Open M.2 Smart IoT Module with a variety of sensors including environmental sensors or microphone arrays. To start with, we’ve equipped the current design with an on-board temperature/humidity sensor.
The PCB design files for the initial release have been published on GitHub as open source hardware. We are now in the process of prototyping and bringup.
Furthermore, the nRF52840 is a platform already supported by Zephyr, so it is possible for the module to operate Zephyr RTOS for handling sensor data acquisition and synchronization, while the host SoM running Linux takes care of high-level communication and processing.
Open and transparent designs for privacy and security
Interesting to know, as part of our collaboration with Google’s TF Lite Micro team, Nordic’s nRF52840 is also supported in our open Renode simulation framework – and on top of that, we’ve recently added initial BLE support for it. This allows for seamless integration and debugging of the radio protocols in simulated environments assuming multi-node configuration with nodes controlled by different pieces of software. Renode allows to prototype and test even complex IoT systems with gateways, sensors and other auxiliary devices before the target hardware is ready, which not only greatly reduces the time-to-market factor, but perhaps most importantly – allows for thorough testing against security threats.
Antmicro’s Open M.2 Smart IoT Module makes for a versatile booster for some of the top-tier edge platforms on the market, which opens them for new applications by providing a simple and thorough way to add significant AI processing power. This leads to smart gateways with open hardware and an open software stack, giving our customers full control over system architecture, performance and security in a transparent manner.
If you are interested in that kind of integration and would like to fully benefit from our open design approach to secure and scalable AI-enabled IoT products, contact us at contact@antmicro.com.