Artificial intelligence and machine learning will transform the IoT
One of the biggest transformations happening in wireless IoT connectivity right now is the introduction of artificial intelligence (AI) and machine learning (ML).
This isn't coming 10 years from now. Or even five years from now. This is where everything is heading behind closed doors on countless product roadmaps that will come to market within the next couple of years.
In parallel, AI itself is also generating a lot of public interest and thus mainstream press coverage. This ranges from AI being described as a solution to all of humanity's problems to AI being its biggest existential threat.
I personally see AI becoming massively beneficial, but probably needing to be subject to some kind of legalized or certified control.
But in the IoT, there is no question that AI and machine learning is going to push everything towards being much more intelligent, useful, and powerful. And enable new categories and classes of application that were previously unthinkable.
As such, if your next generation product or application isn’t incorporating AI and ML, then it risks being completely outclassed—even invalidated—by your competitors’ products that do employ these technologies.
And this thinking is what was behind Nordic’s announcement earlier this year to acquire US-based artificial intelligence and machine learning company Atlazo. As well as Nordic’s ongoing partnership with Edge Impulse in the AI and machine learning space.
Here I'd like to briefly predict what I think are going to be some of the key developments to expect from AI and ML in the IoT.
If you're not thinking about AI and ML, you really should be
In the IoT, AI and ML are going to enable whole new classes of applications and markets that were completely impossible or extremely challenging to develop before now.
Technologically, using AI and ML in the IoT was traditionally way too complex, slow, and insecure. And commercially, the use of AI in IoT was simply too expensive: you needed costly high end processors and a team of data scientists just for starters.
What’s transformed this is the development and advancement of 'tiny machine learning' or 'TinyML'. This enables ML to be performed on constrained-computation devices in a highly simplified way. The industry gold standard for running advanced AI and ML algorithms on resource-constrained devices typical of the IoT is TensorFlow Lite.
In parallel, Nordic Semiconductor has constantly expanded the limits of what “resource constrained” means in battery-powered IoT with its nRF52, nRF53, and nRF54 Series Systems-on-Chips (SoCs).
And when it comes to marrying high computing power to low power consumption. AI traditionally required a lot of computational processing that in turn burned a lot of power: more than could be supplied by batteries.
Yet battery power is required to do anything useful in IoT. And this is where Nordic Semiconductor’s expertise in ultra-low power wireless with the use of high-end processors is now going to firmly come into play. In fact, Nordic is at the forefront of AI and ML in IoT and ready to enable any of its customers to take advantage of it.
It’s now possible to develop AI and ML-powered applications that use battery-powered resource-constrained devices. Applications that multiple Nordic chips are capable of supporting – including both the nRF53 Series (nRF5340), and nRF52 Series nRF52840 and nRF52833. But without question, Nordic's latest nRF54 Series (both the nRF54H20 and nRF54L15) are exceptionally well matched for AI and ML due to the way they combine optimal-computational power with -power efficiency.
AIoT turns from ‘nice have’ to ‘must have’
AI and ML in the IoT, sometimes termed ‘AIoT’, can turn an IoT application from useful to incredibly valuable if not invaluable. By that, I mean indispensable in a vitally necessary way.
As an example, consider a health and fitness wearable. This in the past might have been used to measure your steps, calories burned, heart rate or ECG, and sleep patterns. This kind of data has never previously been freely available to consumers and certainly not from a simple-to-use wearable.
Although this wearable would have used advanced algorithms to process sensor data, it wouldn’t have employed AI or ML. But add AI to the mix and that same wearable with the same sensors can suddenly elevate to an entirely new league of capabilities.
Now you have the foundations for a medical wearable that can continuously monitor several vital signs simultaneously. And if they all suddenly change, and especially together in a specific ‘red light’ combination, could be used to immediately signal and even pre-diagnose a serious medical emergency. An example might be a sudden change in blood oxygen level, heart rate, blood pressure, and breathing that indicate an imminent cardiac problem.
You suddenly have a wearable that becomes a 24x7 first-responder care device that never gets tired and rarely if ever makes mistakes or triggers false alarms. And enables the first responders to arrive warned of what might be wrong - slashing vital seconds or even minutes off delivering the critical care required.
That’s a transformative continuous medical monitoring product that could optimize stretched healthcare budgets and above all save lives. It’s just one example of the raw potential of battery-powered AI and ML in wireless IoT.
Everything happens at the edge
In the early days of the IoT there was this notion that everything would revolve around big data: both in its collection from millions of sensors, and its analysis in the Cloud. But what became very quickly clear was that this approach was a technological and commercial non-starter.
Sending vast amounts of raw unfiltered data up the Cloud was time-consuming and costly, exposed that data to unnecessary privacy risks, and could not support battery-powered operation for extended periods of time.
It soon became clear that the solution is to filter out the key, relevant data and make any data-driven decisions as close as possible to the sensors themselves. This means locally or what's termed ‘at the edge’.
Such ‘edge computing’ will allow all kinds of advanced and exciting battery-powered IoT applications. One example is predictive maintenance and monitoring in fields such as electricity power grid monitoring where utilities need to know as quickly as possible about critical events such as trees falling on transmission lines or fires. Even consumer washing machines could be monitored for impending failure from simply listening to the tell-tale sounds a machine makes when operating abnormally.
Why more is more in AI and the nRF54H Series
Running AI and ML applications successfully on battery-powered wireless chips requires powerful computational capabilities at modest power consumption levels. Nordic's nRF54H Series offer the best of all worlds. You get multiple Arm Cortex-M33 processors that offer best-in-the-world levels of mobile computing power at the lowest possible power.
This means you can do 10x as much data processing as you could in the past. Or do the same amount of data processing 10x faster and have the device rapidly go back to into ultra low energy sleep mode, and so use 10x less power. That’s the key.
- Read more: Massive hype to massive IoT
You then have the latent ability to support the widest range of applications in the lowest power way possible.
The game-changer here is going to be battery-powered AI and ML and that is what the Nordic nRF54H Series is all about and where it really shines.
The nRF54H Series breaks the traditional tradeoff link between choosing either high computational power and high-power consumption, or low computational power and low power consumption. Now you can have high processing power and low power consumption: the lowest power on the market for AI and ML.