At the end of 2018, I have been invited to take part in beta testing of the new IoT + AI solution by AVNET and Octonion, organized by Element14. I was very interested in this new product and its ecosystem, as quite a lot of my latest work has revolved around AI, one way or another. Once the beta tested completed, I thought it could be helpful to take part to the RoadTest that followed, so I have offered to participate, as I already had the device.

I am writing a series of blogs that will be published alongside the RoadTest. The aim is to dig a bit deeper on some of the interesting features of this solution, trying to explore it and find its strengths and weaknesses, while at the same time better organise the RoadTest, breaking it down in smaller and more manageable parts.

In this first article, I will give some information about this new solution. While most of the information can be found on the marketing material published on AVNET and Farnell websites, whenever possible, I will try to enrich the information using the experience gained using the product.

What is all about

In the last few years, Artificial Intelligence and its derivatives, have gained traction and became mainstream on the technology media circus, and they are now amongst the most used buzzwords in papers and all sort of marketing material around (the choice of the blog title has not been made randomly! ). As such, it is hardly a surprise to find it all over the advertising brochures of AVNET SmartEdge Agile device  (well, I have to admit: whoever came up with the name for the device has done a pretty good job cramming in as many buzzwords as possible: Smart, Edge and Agile...3 for the price of 1!).

But is this solution just a marketing exercise, or is it bringing something new in the arena? Before even attempting to answer, we need to understand what the product is, and which needs (and market) it is trying to address.

From AVNET website:

Avnet’s turnkey SmartEdge Agile is a certified hardware solution with a full software stack that helps you avoid the complexities of deep learning tools and development processes. This offering allows you to build artificial intelligence models and push them to the edge with the support of Brainium’s zero-coding platform hosted on Microsoft Azure.

With SmartEdge Agile, you can use the same solution for all stages of your IoT project from proof of concept and value to pilot and production—with confidence that constant software and firmware updates along with reference designs and related BOMs keep you on the edge of the cloud and the cutting edge of development.

SmartEdge Agile isn’t just a new product, though. It also establishes a brand new category in the Internet of Things: meta-sensing. This CE and FCC-certified product is the first in a suite of hardware-software solutions that allow engineers at any level to develop and deploy AI-backed IoT solutions in as little as six months.

It is quite an interesting proposition. The solution has an hardware component and a software one. The hardware is the SmartEdge Agile device, a powerful IoT platform, equipped with a bunch of sensors and some processing power. The software is Octonion's Brainium software stack, which provides the backbone for management of IoT and AI, and securely connects the device to the cloud.

The product is aimed to manufacturers and engineers looking for rapid design and prototyping of IoT and/or AI solutions. Leveraging both the hardware platform offered by the SmartEdge Agile device, and the zero-code approach offered by Brainium and the AI Studio, AVNET reckons the development cycle can be significantly accelerated, ensuring a shorter time to market.

In this first part, I will explore in more detail the SmartEdge Agile device.

SmartEdge Agile

The first thing you notice when you get the SmartEdge in your hands is that the device it is rather compact. The enclosure is made of a sober black and white plastic, and it is not much bigger than a cigarette lighter.

From the outside, beside the on/off switch and the LED indicator, it is noticeable the presence of a particular kind of USB-C port. On closer inspection, on either ends of the device there are some rubber gaskets, which is a strong indication the device could have some sort of waterproofing.

Checking the USB-C port, it is rather evident it is really a special port: the photo on the right shows there is some rubber sealing around it, which suggests it may indeed be a waterproof USB-C port (similar to this) , possibly rated as IPX7 or even IPX8 (the "X" is due to the port not being classified for the protection from ingress of solids).

My bet is that the device is waterproof, resistant to accidental splashing of water, and possibly to some very short-time submersion in shallow water (<1m), but since I could not find any reference about IP rating in the official device literature, I would not try dipping the SmartEdge in water as yet!

UPDATE: I found out the device is IP67 certified. It is stated on the back of the packaging the SmartEdge Agile is packed and sold with.

Before taking apart the device, it is useful to understand the modular nature of the SmartEdge Agile, by taking a look at its architecture, as depicted below (source: AVNET):

The main board features two board-to-board connectors, used to expand the system. One of the connectors is used for the Sensor Board V1, and the other is available for adding an alternative connectivity board. AVNET has clearly made an effort to make the device as flexible as possible, to accommodate customers' specific requirements with minimal effort.

I can see the value in such offering, especially for those manufacturers that need to add IoT (or AI) capability to their products but don't have core competencies in the IoT area (or AI), and have very short time-to-market constraints. Using AVNET device and services could help, but... is there a catch?

Time to look inside the SmartEdge and check the specifications (full specifications can be found on the published datasheet).

The device I have been given, has the main board and the sensor board V1 configuration. The main board (BRAIN module) is built around the STM32 32-bit microcontroller, and uses a Nordic nRF52840 Bluetooth 5 SoC for low power connectivity.

While it was easy to spot the nRF52840 on the board, the STM32 microcontroller was covered under the QR code. Gently lifting up the sticker revealed the MCU is indeed the STM32L46QG, an ultra-low power ARM Cortex-M4F. Looking at the specs for the Nordic SoC , I cannot help but wonder how taxing is, in terms of processing power needs, the "AI Engine" hosted on the board: in addition to all the Bluetooth 5 LE and the communication facilities (like encryption, etc...), the nRF52840 is also equipped with a rather powerful 32-bit ARM Cortex-M4F core. Surely, judging from the datasheet alone, the Nordic Soc would be more than capable of managing the sensors and the communication, making the STM32 redundant. Obviously mine is merely a speculation as unfortunately, for the AI side of the product, there is very little information available.

The slideshow above illustrates some of the parts of the device. As said before, the device came with the Sensor Board V1, shown below.

The little board accommodates quite a number of sensors, and looking at the markings, they are all connected via I2C (from the photo on the right, amongst all the test points, we find the familiar SDA and SCL pads, which is probably the test access to the I2C bus where all the sensors are connected). Unfortunately I cannot verify this taking actual readings from the test points, as I don't have anything I can use to make a stable contact with the pads, and trying to solder anything on such minuscule 30MIL test points is way beyond my capabilities (and tools).

On the board there are the following sensors:

1. Light Sensor: AMS TSL2540
2. Temperature/Humidity Sensor: STMicroelectronics HTS221TRl
3. MEMS Microphone: STMicroelectronics MP34DT01-M
4. Time-of-flight (Proximity) Sensor: STMicroelectronics VL53L1CXV0FY/1
5. Pressure Sensor: STMicroelectronics LPS22HB
6. Inertial Measurement Unit (Accelero/Gyro) Sensor: STMicroelectronics LSM6DSLTR
7. Magnetic Sensor: STMicroelectronics LIS2MDL

The task to supply power to the board is assigned to the Li-Poly 260mAh rechargeable battery.

This small battery ensures the device can work from a minimum of about 18 hours, with all the sensors enabled and collecting data at fastest sampling rate, to a maximum of 22 hours, with only one sensor (temperature) enabled and collecting data at the slowest sampling rate. The battery can be recharged using the USB-C connector, and takes about 3.5 hours to fully recharge from flat.

The USB-C connector can also be used to extend the device further, by adding other peripherals/sensors. This functionality has not been completely developed yet, and it is not yet available. Underneath the battery, there is another interesting IC, the Microchip USB2514 - USB-2.0 High-Speed 4-Port Hub Controller, which supports the idea the SmartEdge can be expanded also via USB connection.

This pretty much covers all the hardware specifications for the SmartEdge Agile, so let's talk about how the device is used. Basically, on its own, the device doesn't offer any user interface to interact with: all you can do with it is switch it On/Off and watch the LED blinking. If you want to manage the sensors, you have to use the Brainium Portal.

Essentially, the device is sold as part of a development kit, and the cost covers both the hardware, and the licence to use the online Brainium Portal (and the Microsoft Azure infrastructure) for 6 months (there is a data usage limit as well).

I will go in further details about how the "software side" of the solution works in my next blog, for now it is enough to know that without it, the device on its own is useless (I was hoping there was a way to leverage the USB to connect the SmartEdge to my laptop, and use its sensors "raw", bypassing all the software stack, but it is not possible)

Once again, the choice seems clear: AVNET is providing a Development Platform for IoT and AI, for everyone to use as prototyping tool. Then, if you think the platform is just what you need/want and would like to use it with your product/solution, you have to engage AVNET and basically use their services to finalise your product. While I think that the SmartEdge Agile boards are "reference designs", I'm not sure if the customers would be allowed to use the designs freely in production (to add to their own boards) or if they need to purchase from AVNET. Either way, the software side is not going to be open, so its cost needs to be factored in regardless. I have not enquired about the costs involved with all this, but I suspect it won't be cheap.

Now, as hinting at the beginning, there is a value in this proposition. You just offload to AVNET/Octonion all the burden of designing the IoT/AI layer, so you can focus on your product's functionalities. Not knowing how much this would cost, I cannot say if it is good value for money.

For sure, it is a different kind of development board, one that, after 6 months, becomes useless.

In my next blog I will introduce the Branium Portal, and will discuss the AI capabilities of the tool.

From the same series:

AI to the Edge - Part 2 : Introducing the Brainium Platform

AI to the Edge - Part 3 : AI Studio