Microchip VP Rod Drake: Pandemic accelerates shift to machine learning and edge IoT

Rod Drake, vice president at Microchip Technology

The Covid-19 pandemic has shaken up the embedded industry, not least with a dramatic shift to more automation, sensing and edge processing with machine learning.
While customers have accelerated their digitalisation plans, the semiconductor industry is running hard to keep up with demand. This means flexibility is key, from the design of a board or system all the way through the supply chain. This will be more apparent through the challenges the industry will face throughout 2022.

Meeting these challenges will not only require flexibility in how the edge processing is implemented, the types of devices and techniques that are used, but also the way edge systems are rolled out across the IoT. Microchip’s wide range of technologies with its own manufacturing capability provides an advantage in being able to deliver the technology that the end customers need.

One example is smart image sensors with local machine learning on programmable FPGAs. This puts the key machine learning algorithms for image processing at the edge by the cameras, rather than sending streams of data to the cloud. This reduces the latency of response, the power consumption and the data requirements across the IoT network.

Using FPGAs also allows engineers to develop their own machine learning or tweak existing ones for their particular application. That can provide more efficiency for specific applications at the edge, from defect detection for quality control on a production line to control systems.
For example, the latest Microchip FPGA development platform adds new sensors with an interface that links industrial cameras with 1 Gbps per lane while receiving up to 1.5 Gbps per lane. There is increasing demand for higher performance interfaces as developers use cameras with higher resolution to capture more detail and need to perform the processing locally. This avoids overwhelming the local network and allows a boost in productivity with the existing infrastructure.

While high speed Ethernet networks are increasingly common in industrial edge applications, an FPGA solution at the edge also allows developers to configure the system to specific industrial networks such as Profibus and Hart. The flexibility of adding network protocols to an FPGA reduces the size and complexity of network nodes and gateways, and this is a key trend for 2022 and beyond.

Machine learning can also be implemented on a wide range of microcontrollers for applications such as predictive maintenance at the edge. Microchip works closely with the developers of software algorithms to run sophisticated pattern recognition code on microcontrollers close to the sensors. These allow for more local monitoring of equipment, identifying patterns in the data that can indicate that equipment is starting to fail, and even the location of the problem.

This allows machines to be taken off-line in an organised way as part of scheduled maintenance with time to order replacement parts. This avoids unexpected failures that can bring a production line to a halt and can cost millions of dollars in lost production, hitting delivery schedules to customers.
These machine learning frameworks are constantly improving, using data from the suppliers or from the application itself, boosting the accuracy of signal detection and classification and improving the performance of the overall system.

The high level data from these edge systems is also fed back to cloud services that are growing in importance through 2022, and as a result security is also key for these systems.

With more and more devices connected across the IoT, designers are realising that applications are vulnerable to being taken hostage and edge nodes are generally very susceptible to security hacks. This is driving demand for over the air (OTA) updates to keep edge IoT device security up to date.

OTA updates are basically a required feature now and that needs security, otherwise the network is open to someone inserting unqualified code into an edge node. Microchip has relationships with all the major cloud suppliers and adheres to the latest security standards. This is an area that is growing fast and will continue to do so through 2022.
Functional Safety is also moving into industrial applications, taking the design methodologies used for driver assistance technologies and driverless cars and applying them to the factory floor.

A lot of industries followed automotive with adopting the ISO9000 quality standard and Microchip sees the same thing happening with the ISO26262 standard moving to industrial designs. Having the ability to understand in the system how something is going to fail and what will happen and that it fails in a safe manner – this is the most mature and most widespread in automotive and a key technology for edge IoT in 2022 and beyond.

Through 2022 there will continue to be a constrained environment for the availability of components, and this means communication with suppliers is key. Microchip has always worked closely with customers on their designs and this will be a key area of focus over the next year. As Microchip has its own production capacity, it has more control over availability of parts, and the programmability of microcontrollers and FPGAs gives developers more flexibility in avoiding supply constraints.
What is clear is that digitalisation is a vital part of industrial designs, particularly at the edge, and that will continue to be a significant driver of growth. From smart sensors at the edge with high speed interfaces and FPGA processing to machine learning algorithms running on microcontrollers, there are many different options for customers to consider.

Working closely with semiconductor suppliers such as Microchip will be key to ensuring the successful roll out of edge IoT systems. Early communication and detailed planning will help the supply chain deliver the technologies that are needed to meet the drive to edge IoT that is happening all around the world.