At Embedded World this week, companies are showing off embedded chips with more of everything: more memory, more performance, more accelerators, more software options, more cloud integrations, and yes, more security. What used to be a fairly staid and even somewhat dull market has exploded with options, and it’s all thanks to connectivity.
When you pop a radio on a constrained chip you open up a world of potential. And that potential is causing both the humble microcontroller and the embedded software world to look more and more like the traditional CPU and IT market, respectively.
Unfortunately, the core needs of the embedded hardware market haven’t changed. These devices are running our factories, water treatment plants, oil rigs, and life safety equipment. Whereas at one time they were mostly secure by way of obscurity and running highly specialized software, they are now getting connected to the rest of the world and getting additional jobs such as running neural nets used in AI.
I’m not at Embedded World, but I did have a few briefings and conversations with companies that have highlighted the big shift in processing and attendant software ahead of the show. First up, let’s talk basic silicon. While there may be some use cases that still use old-school 8-bit or 16-bit microprocessors, it’s clear that the ability handle more bits are better, despite 32-bit chips costing more. This trend is driven by the demands of adding memory and software to handle the security needs of the new embedded market and new jobs I mentioned.
Carsten Rhod Gregersen, founder and CEO of Nabto, told me that simply providing for encryption on a device requires 32-bit architectures. And most embedded microprocessors require some form of security, especially if they get connected to the Internet. That desire for security and the need to manage connected devices is also driving two other trends in the embedded world.
The first is a reduction in the number of RTOSes. The second is a need for high-quality open source software and software tools. I often talk about the reduction in the number of RTOSes that companies want to support, and it makes complete sense. If you’re developing an application that needs to run across multiple devices (such as a suite of home appliances or different machines on a factory floor), ideally you’re writing software to run on all those devices just once, which requires writing to just one RTOS.
And when it comes to the winning RTOSes Gregson believes open source options such as Zephyr are going to gain ground. He said that after the acquisition of FreeRTOS by Amazon and ThreadX by Microsoft, companies are worried about the potential for lock-in on the part of those behemoths.
Indeed, the desire to move away from proprietary systems extends beyond RTOSes to other areas of embedded software. For example, at Embedded World, Canonical showed off deals designed to bring open source Ubuntu to Mediatek’s hardware. It also boasted that its OS had achieved ARM’s initial PSA Level 1 security certification.
Thanks to greater performance, embedded devices are also performing a host of new jobs. What used to be control-loop functions and dedicated tasks have turned into using the available computing for flexible tasks. These tasks might include running some kind of software in a container, vision processing, or other AI.
In fact AI was everywhere, from the Tiny ML I get so excited about to specialty computing platforms designed for robots and automotive that can fuse camera data with other sensors to handle complex tasks industrial tasks. Renesas showed off demonstrations of an ARM-based microcontroller doing people detection, showing off one of the popular new jobs higher-performance microcontrollers can do for the embedded market.
As we connect everything, it’s clear that the embedded world will look more and more like the traditional computing world. But even as it takes on new tasks and puts higher-level computing in more industries, we must figure out how to ensure that the glitches and best effort nature of today’s computing infrastructure don’t compromise the crucial world that embedded devices have to do.