How To Build A Trillion Connected Things

A trillion is a big number.

A broadcaster reads about 1,000 words in a five-minute newscast. At that rate, it would take 6,000 years at that rate to finish a trillion-word newscast.

We’re on a path in the technology industry to build and connect a trillion devices in the coming years. How are we going to do that, given the scale and sheer size of the task? Is it achievable in the time frame allotted?

If we assume 2 square millimeters per IoT sensor device, or 35,000 packed onto a 300mm wafer, a trillion devices require 28 million wafers. That’s three times the annual capacity of the industry’s largest foundry, but just 30% of today’s total annual worldwide production. So in this context, it’s achievable.

The resource challenges
Energy on the other hand, is more of a challenge, given that the vast majority of these devices are expected to require battery power. A 240 mAh coin cell weighs 3g and contains about 109mg of lithium. One trillion cells contain 109 billion grams, which is 109,000 metric tons. The world’s annual lithium production today is one third of that. In this case, it appears we’re going to have to rely heavily on energy harvesting for these IoT devices, which has its own set of challenges I’ll reference below.

Add to that engineering resources. Someone has to design the trillion sensors. What does that entail? There are roughly 5,000 people worldwide who can design a radio for a sensor node. Spread across a trillion sensors, that’s 200 million sensors per designer. That suggests a small number of platforms that would be re-used in many applications. Another approach might be to simplify sensor hardware design so that it was more like coding. There are roughly 20 million people worldwide who can write code. If they were all designing sensors, and made 50,000 copies of each of their designs, we’d still get to a trillion. A long tail of highly customized designs and a small number of standardized platforms are both viable models for the trillion-devices future. Which will win? We don’t know yet.

Regardless, engineers are a resourceful bunch and we’ve overcome the “impossible” countless times in the history of electronics by focusing on essentials. We’re doing the same as the IoT evolves, and IoT needs four crucial elements. It needs to:

• Work separately
• Work together
• Work automatically
• Work resiliently, securely and safely

Separate
Nobody wants to connect a thing if that thing does not perform some useful function in the first place. A temperature sensor needs to sense the temperature and a camera needs to take pictures. If they don’t do these jobs well and cost effectively, nobody will buy them and there will be no need to connect them. Things matter for a successful IoT.

Together
For the promise of IoT scale to be realized, these trillion devices need to communicate with each other and with the cloud to make the most (and most efficient) use of collected data.
Think about connected cars. Automotive capabilities will evolve, and so will their communication needs. Latency becomes a vital issue in such systems. As cars go from being able to park themselves, to knowing where the parking spaces are, to truly autonomous driving, we’ll have varying needs on the communications and latency fronts. To state the obvious, no driver wants a lot of latency in his or her car’s reactions—or the reactions of nearby cars for that matter.

Automatic
Artificial intelligence and machine learning play an enormous role along this road to a trillion devices. A case in point is image recognition, where a system can be trained to determine with a high degree of confidence whether it’s looking at a dog or a cat. Currently, most of these systems rely on data being transmitted to the cloud. But some image recognition is simple enough that it could easily be handled in the camera itself.

This reduces latency and communications bandwidth and increases user privacy in some cases.

Resilient
Lastly, in a world of a trillion connected IoT devices, we need resiliency. Think about safety-critical systems—pumps, robots, locks and so forth—where failure can be catastrophic.
Here is where we need a stepped-up focus—an industry collaboration—to improve security from edge to cloud. Part and parcel of this is to make the user do much less of the work in the security chain, and to make it invisible for him or her as part of daily routines.

We’ll be talking more about this in the coming weeks.

Laying down the gauntlet
At Arm TechCon 2016, SoftBank founder and CEO Masayoshi Son said IoT will enable a “Cambrian explosion” of innovation and that he sees the world getting to a trillion connected devices by 2035.

Are we on a trajectory to achieve that? We can certainly get there tackling some of these near-term challenges. In some cases, it’s going to require reconsidering design. Take energy for instance. Most of our battery-powered systems today are designed essentially to “run fast and stop.” Energy harvesting is fine, but often doesn’t generate enough useful current without storage capacity to be effective. How do we work around that? My colleagues in Arm Research and Development are hard at work investigating possible solutions.

So, can we build and connect a trillion devices in the next 18 years? If you predict 20% growth rates each year, it’s achievable, but we have to consider disposable devices. Right now, the world produces 400 billion soda cans a year, and if we had a reason to connect them, then we’d get to a trillion connected cans easily. If you consider medical and other IoT applications, such as ingestible sensors to check your internal organs or product-tracking sensors on almost everything, then we can do it.

Ultimately, we’re at a tipping point. We need to move from a world where we ask, “Why should this be connected?” to one where we ask, “Why shouldn’t this be connected?” I’ll be describing how we navigate this journey in more detail this week during my keynote at Arm TechCon in Santa Clara, Calif. I hope you can join me.