Are OEMs ready for EdgeAI?

When Industrial EdgeAI does arrive in volume, its impact won’t be felt at the edge, but at the core. And by that I mean the core of the business.

Although technically complex, from an operational viewpoint, EdgeAI is really an extension of IoT, and IoT is an extension of industrial automation. At every point, what really drives uptake of technology is a reduction in operational costs.

Historically, that cost reduction typically comes from moving the human in the loop. That doesn’t mean fewer people for the same process (although it can), but rather more productivity with the same number of people.

Common or garden automation is relatively dumb. It doesn’t need to be smart, it just needs to do the same thing, over and over again, with reliability and accuracy. A skilled operative will monitor the process (or multiple processes, given that automation is intended to enable scaling up). In this scenario, the operative is physically close to the machine.

The IoT moved the operative away from the machine and into an office. But that office doesn’t need to be anywhere near the machine. Very soon, Master Mariners will sit thousands of miles away from their fleets of fully automated boats that have near-zero crew onboard. Manufacturers are making products in factories continents away from their workers. That’s enabled by global, low-latency machine communications, and that comes under the heading of IoT.

When Industrial EdgeAI arrives it will reduce the need for human overseers to constantly monitor operations. Operational decisions will be made by the technology. At first, those decisions will be non-critical and double-checked by humans. Over time, there will be less need for checking the decisions, and the decisions will become more critical.

We are some way from that scenario, but we are definitely moving in that direction. This puts Industrial EdgeAI in a very different category to Generative AI, but they are close cousins. The technology is essentially the same. The fact that we can scale down AI to the point of putting it into small sensors sitting in complex machines illustrates how amazing the semiconductor industry is.

More semiconductor manufacturers are bringing products to market that integrate or embed some form of low power AI acceleration, intended to sit in embedded devices that live at the edge of the network.

I believe AI-enabled semiconductor products will remain a minority in their categories for the next decade or so. Is there a future where all microcontrollers have AI or machine learning hardware integrated alongside the more familiar cores and peripherals? Well, that’s the beauty of the semiconductor industry; when the cost of integrating a few million more gates becomes incidental, then AI acceleration may become a standard feature across the product portfolio.

Adopting anything new into the embedded sector, like an RTOS or Linux, wireless connectivity, internet protocols, and now EdgeAI, impacts product development cycle times. They will be pushed out as engineering skills ramp up. Over time, they tend to return to their stable position of around 24 months. With that in mind, I think it will probably be at least 2 years before we can test today’s estimations about the full impact of Industrial EdgeAI.

There are some examples already in the market and no doubt many projects in development right now, but the real volume is a while away. Some OEMs won’t be ready to let product cycles stretch to support the adoption curve. Others will feel their end-products aren’t ready to benefit from EdgeAI or machine learning. Both make perfect business sense. For those OEMs that feel the time is right on both sides of the equation, the industry is ready to support them.

(Disclaimer: The thoughts and opinions expressed here are my own and are not intended to reflect those of my employer)