Rob Aitken is digging a bit deeper into what it would really take to connect a trillion things in his keynote next Thursday at Arm TechCon How to Build and Connect a Trillion Things . What would those things be? What might unit volumes be? How could we power them? Secure them? Manufacture them? He is talking on Thursday morning at 10.10am The keynotes are open to everyone, even an "exhibits only" pass. But I couldn't wait that long, so I went over to Rob's office at Arm's US HQ off First Street, technically on the pastoral sounding Rose Orchard Way, notable for the absence of orchards and roses. It is an example of what my ex-wife noticed about housing and commercial developments, that they get named after whatever was destroyed to build them: "Cherry Glen", "McCarthy Ranch" and so on. How Big is a Trillion? It is 10 12 , of course. But how does that compare with other things we produce? Here's one datapoint to put it in perspective. The world produces about 400B beverage cans per year. So if we put a "thing" chip on each beverage can then it is the right order of magnitude, but it is hard to see why we might do that. Expensive wine, to make sure it has always been kept at the right temperature, sure. But not every can of Coke or Budweiser, maybe each pallet of soda or beer. But there is a message hidden in here: we will only get to a trillion things if some of those things are disposable. We just don't use a a trillion permanent objects. Getting to 1T devices will take, maybe, 30 years. But growth is likely to be exponential with a flat part of the curve at first and getting really steep in the out years. How many is a trillion things? It's a lot, but not unmanageable. The chips would be about 30 percent of today's worldwide semiconductor production. Old Standards One challenge for most interesting markets for "things" is that there are lots of pre-IoT standards about what a device should look like, what is its physical size, how does it get its power, and so on. The IoT device has to fit with whatever is there. For a historical example, in his office Rob had the old phone in the picture on the right. This was a rotary dial phone. If you have a landline, you can still connect a rotary dial phone to it and it will work. In the modern era, when phone lines can be used for ADSL or other digital services, this is known as POTS. It stands for "plain old telephone service". I am not making this up, that is the name that the industry uses. Of course, when this phone was new, it was a "thing" and it was physically wired into the wall, since the RJ-11 connector was only introduced in the 1970s (and this phone in Rob's office has been retrofitted with an RJ-11 connector so it can actually be used). So this is a good example, an old standard that gradually evolved. Now, upgrading this phone to voice-over-IP might take a bit more effort! So IoT devices are going to have to fit in with existing infrastructure and then gradually convert. Today a smart light bulb fits in the existing sockets, and connects using WiFi or Blueetooth. Longer term, it's not hard to imagine smart light bulb outlets that carry the control signals, too. What Will Getting to a Trillion Devices Take You need to design them for a start. Getting to 1 trillion devices is not going to be a huge number of just a handful of devices, the market is inevitably going to be very fragmented. All (or almost all) IoT devices contain radios. There are about 5,000 RF engineers in the world today. That reminds me of when I was last in Cadence and I gave a keynote at an internal conference of Nokia engineers in Finland. The other keynote was a Nokia VP. He told me that he had a spreadsheet that listed, by name, every competent RF engineer in the world, where each one worked, and contact details if they knew them. However, for a trillion radios, that gives an average volume of 200M things per RF designer. Rob is not trying to do an accurate calculation here, just see the orders of magnitude. There are about 20,000 IC design teams and, if we need to design each device, we need an average annual volume of about 50M, assuming each team designs one device per year. On the other hand, there are 20M people in the world who can write code, so the average volume per chip to get to a trillion devices is only 50,000 per software engineer. The conclusion of all of this is that IoT devices will need to be platforms that can be programmed in software. There aren't enough IC designers and RF designers for it to be any other way. So the same platform (chip) will be used for your toothbrush, your cordless drill, or on the factory floor. Batteries How many batteries will we need? A trillion, of course, if they are all battery powered. A standard CR2032 coin cell ("hearing aid battery") is 240mAh. It weights 3g total, of which about 109mg is lithium (lithum is element number 3 so is pretty light). So 1 trillion contains 109 billion grams which is 109,000 metric tons. The approximate worldwide production of lithium is 32,500 metric tons, so quite a bit mismatch. Conclusion: We will need energy harvesting. But energy harvesting has a big problem. Batteries deliver a lot of power but for a short period of time, and we design chips to work that way. Do all the work, then go to sleep for a long time. Chips running on energy harvesting can run forever, but can only consume a tiny amount of power when running. Plus, the power may go away at any moment. That can be handled a little by adding a tiny battery, but then we are back to the lithium problem. Or we could use a capacitor. Or, perhaps, we can just design systems that can cope with the power going away and coming back at any moment. Obviously this requires non-volatile storage. Security and Trust The whole IoT system is going to require end-to-end security from the edge device to the cloud. Based on what has happened so far, I would say that the biggest issue is likely to be that a lot of people who don't know what they are doing will design the security of their things. So the big issue is likely to be getting most things up to state-of-the-art standards, not advancing the state of the art. Just getting everyone to follow basic security. Systems will also need to be resilient, able to respond when attacked and able to be updated with new firmware. Many standards evolved before we were worried about hacking and security. One I immediately though of is mobile phones and basestations. When the systems were design, the threat of bad guys cloning phones was understood and phones need to authenticate themselves to the network. But the notion that basestations might be cloned seemed silly, they were huge towers costing millions. But see my post Do You Know What Stingray Is? which talks about portable basestations that hijack phones in the area. Supposedly these are used by law enforcement, but there are stories of law enforcement discovering these devices where nobody knows who owns them. Three-letter-agencies? The Chinese? The israelis? The Mafia? Drug dealers? Who knows? It is now clear that the networks should have had to authenticate to the phones (terminals in mobile-phone speak) but it is too late to change that now. Trust and security are opposite properties, and at some level, all security involves trust. We don't ID our kids in the morning to make sure they haven't been replaced in the night by impostors, we just trust. All cryptography trusts something. For example, the RSA public key algorithm that underlies internet security depends on the unproven belief that very large numbers (hundreds of digits) that consist of multiplying two prime numbers together cannot be factorized in less than geological timescales. Trust depends in some way on value. If you buy a ballpoint pen, you don't worry about its history. It is unlikely that Staples stole it, but you don't care if they did. On the other hand, when you buy a house, you pay a title company to look into its history and you don't buy it if it is stolen, and you even take out special insurance to make sure that you are protected even if the history turns out to be incorrect. In the middle, there is the zebra mentality: some zebras will get eaten by lions but most will not, so there is some level of lion attack that can be lived with. Learn More This is basically a trailer for Rob's keynote at Arm TechCon. You can find out more about the keynote on the conference website . It is October 24-26 in the Santa Clara Convention Center. You can also register . Note that all UBM-run conferences use the same login, so you might not need to start from scratch. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧