Over 10 billion Internet of Things (IoT) devices already surround us in our everyday lives. Its extraordinary growth will be further accelerated by the rollout of 5G—Business Insider believes that there will be more than 64 billion IoT devices worldwide by 2025.
Alongside the growth of the IoT, artificial intelligence (AI) has emerged as the next technology phenomenon. Jim Goodnight, often cited as the “godfather of AI” (thanks to his work on a technology to improve crop yield 45 years ago), recently described AI in an interview with Forbes as a “game changer for society,” with the potential to revolutionize our relationship with technology.
Naturally, it was only a matter of time until these technologies come together to form the “artificial intelligence of things,” or “AIoT.” While the IoT connects “dumb” devices to the internet, artificial intelligence gives them a “brain.” Together, they’re capable of changing the world as we know it.
Why we need AIoT
You might question why the world needs AIoT given the huge take up of cloud computing. Can’t we just connect devices to the cloud, like we do with the IoT, and let the cloud do all of the analysis and decision-making? What’s the point of making the devices themselves intelligent?
The simple answer is that the cloud’s compute capabilities simply can’t scale proportionately with the sheer number of connected devices that the world is going to see in the next few years. Moreover, networks that transport data back and forth between devices and the cloud are bandwidth-limited. Even the most modern communications networks won’t be able to support the explosion of data created by devices. This will inevitably cause unacceptable delays in any decision made in the cloud.
Applications such as autonomous cars, where safety is paramount, simply can’t afford to be restricted by unreliable connectivity, high latency, and low bandwidth when they need to make almost instantaneous decisions based on the changing environment around them.
If, for example, someone steps out into the road in front of a car at speed, there simply isn’t enough time for the sensors on the car to detect the hazard, send the data to the cloud (if indeed there is a connection), and wait for the cloud to tell the car to stop. The perception, reasoning, and action must be done within the car itself to save time.
But beside autonomous vehicles, what other applications could AIoT unlock?
Unlocking a new world with AIoT
Manufacturing is another industry that will benefit immensely from the AIoT. Complex far-field voice interaction, for example, will transform almost every part of a manufacturing process. Machine operators will use their voice instead of finding and hitting the “big red button” to shut down equipment in an emergency.
Where hygiene is critical, people will operate machines without the need for physical interaction. With security, manufacturing facilities could restrict access to certain zones using biometrics gathered from numerous sensors. And with data analysis, faults in manufacturing can be detected and machine maintenance can be preempted, ensuring maximum operational efficiency.
Crucially, these applications require ultra-low latency and the highest possible security—on top of the compute intensive nature of voice interfaces. As such, the idea of transporting the voice data off to the cloud and back again is simply not viable. Where latency means inefficiency, and inefficiency costs money, voice interfaces must be built on the AIoT, not the cloud.
Domestic healthcare is another sector that could benefit significantly from AIoT. AIoT-enabled devices, which would be capable of monitoring things like heart rate and breathing patterns, could preemptively flag any health incident before it occurs. Over time, the data from these incidents could be shared with a GP or hospital directly—a living record of your, or your family’s, health that helps healthcare professionals provide you with the right care at the right time.
Turning the AIoT into reality
One of the main barriers to the uptake of AIoT has been the cost of high-performance CPUs, which are required to provide the processing power needed to implement AI. Until now, the naïve approach to intelligent edge devices has been to adapt the architecture of a mobile phone, with all of the cost and complexity it entails. The chip industry is challenged to drive down the cost of these chips while maintaining high performance and versatility. This requires new approaches that deliver the same level of performance, but with greater economy and ease of use.
And finally, and arguably most importantly, versatility is what will best serve the AIoT market. The AIoT is the sum of hundreds of markets, and thousands of market segments—each with different needs. Therefore, while focusing on cost and performance is hugely important, the chip must be able to support a whole host of applications, rather than one or two specific ones. Moreover, the endpoint solution must flexibly and affordably deliver the combination of compute classes (AI, DSP, control, and IO) required by each of these segments.
A small cog powering the world
The industry is beginning to reach a tipping point where these kinds of chips are finally becoming commercially viable, making the marriage between artificial intelligence and the IoT a real possibility at huge scale. But no matter how far you want to take the concept of AIoT, its success depends on some of the most impressive feats of electronics engineering the digital era has ever seen.
About the author:
Mark Lippett is CEO of XMOS – www.xmos.com
This article first appeared on Electronic Design – www.electronicdesign.com