At Hardware Pioneers’ event ‘The Industrial IoT Revolution’, in London, Bob Tait, Channel Demand Creation at Xilinx, spoke about how the company plays a part in the Industrial IoT Revolution.
For many years, industrial systems have been very closed and proprietary. “What we are seeing now is the IoT movement penetrating into factories. Now there are still a lot of requirements that factories have that are well beyond what a standard IoT product would need. But there is definitely a very strong trend of investment into IoT in this space, which makes it a good place to look for ideas for startups,” said Tait.
We are starting to see connectivity from sensors all the way to the network and cloud, so from a Xilinx perspective, IIoT plays to its strengths.
In the industrial environment there is a lot of connectivity required between different types of sensors and different types of actuators. There are a lot of different types of sensors, for example there are sensors measuring motor speed or the colour of a product as it goes past; then there is a lot of HMI required, with screens or actuators; and then there is a whole range of different networks.
Tait added: “So because the industrial environment has evolved in a proprietary world and there are a lot of proprietary regulations and networking standards, what we are seeing now is a gravitation towards more global standards in the IoT space.”
Data on the edge
Ethernet is being adopted in factories with an overlay of protocols that provide the time sensitivity that these networks need.
Interesting for Xilinx, because of the amount of data that is generated in these industrial occupations, the processing requirements in the edge is increasing exponentially.
The amount of data collected is requiring an edge processing requirement.
In a factory environment, you have a production system that is running at speed. You cannot afford the latency that it would take to get sensor data through a network, up to the cloud, do some data processing on it, make a decision about what has to happen back at the factory, send that information back through the network, and back to the actuators on the factory floor.
There is just not enough time available to do that. So that’s why there is a primary requirement for edge processing, way above and beyond what is typical at the moment.
Because this area is still very much like the Wild Wild West, there are lots of new standards being developed, and there’s a lot of activity and innovation. So there is a need to have lot of flexibility in the products and the platforms that people are building.
There’s a million different standards of network connectivity requirements out there, and those standards are changing constantly. The Xilinx architecture lends itself to supporting this variety of connectivity.
There are three implicit requirements that Xilinx has in the industrial environment - functional safety, longevity, quality.
An FPGA (Field Programmable Gate Array) is a digital circuit you can customise to deliver a specific function.
It’s different to a processor which you run software on, which constantly is changing, and the processor is doing a different job all the time. With an FPGA you decide what function you want it to do, and it just does that. The advantage it has over a processor is it does it very well, very fast and at a very low error rate.
So the performance of an FPGA is far superior to what you get from a processor. The trade-off is, you don’t get the flexibility you get with a processor, where you constantly change the task that it is running at any point in time.
Xilinx’s, Bob Tait said: “Ten years ago, Xilinx started to combine FPGAs with processors, which has been a very successful product for us. It is being used across a very wide area of applications.
“Xilinx has been developing the family of products over the years - we now have a product that has five different ARM processors, as well as the FPGA fabric that’s built into it. It affords you a huge number of I/Os, so you can connect to many different sensors. With the FPGA fabric you have a lot of flexibility in the processing that you can implement and the types of sensors it can support.
On the network side, you can implement a large variety of network protocols. Then as the network standards evolve, you can change those even after you’ve shipped your products into the field. They are ‘Field Programmable’.
Tait contined: "The other thing that we built into the product is security. We have eight different layers of security implementation right down into the silicon. This allows you to build very secure products and connect securely from sensor through to network."
What Xilinx see in industrial environments is lots of different sensors and lots of different types of data being measured. For example, you might be monitoring a motor system; the speed of it, the current, the temperature of different parts of the system; and all of those different sensors will be able to feed into one chip from Xilinx that will be able to do all the processing.
Tait added: "There is a huge portfolio of products that we offer in this space, covering wide ranges of requirements from very small products to complex and sophisticated, high performance products."
Across all of these different areas, one trend that we see dominating discussion with customers is machine learning.
It is being implemented in lots of different parts of the IIoT space.
Machine learning collects data and creates an analysis that you can use to create a model that can be put into the edge. So you can effectively implement intelligence on a macro scale but within a very compact format.
It’s an area of a lot of innovation. It’s an area I’d suggest you look into if you’re looking for opportunities.
Tait concluded: "Another big driver in industrial is vision, which is cutting across a whole range of applications. So we see it in consumer, Virtual Reality (VR), Augmented Reality (AR); we also see it in automotive and aeronautics.
Machine learning is a key part of that, because it is taking these smart vision systems (where you can make some decisions about what is going on in the environment based on the image data you are collecting), to a system that is autonomous; cobots, autonomous cars, etc.
So that’s the trend we are seeing at moment in the vision space. So going from smart vision (e.g., object attention), to autonomous systems (e.g., drone collision avoidance based on camera input).