ARM tries to spread its chips to forests, fields, and factories (Q&A)
Gary Atkinson's job is to make sure the Internet of Things becomes the next big thing for chip designer ARM.
Forest fire on the way? Building stress getting too high? Farmland too moist? Gary Atkinson's job is to make sure a computer knows it.
As director of emerging technologies at ARM, Atkinson's wants to see the company's processor designs spread from their current stronghold of mobile phones to hundreds, thousands, or even millions of small networked sensors and actuators. It's a core part of what's called the Internet of Things.
In his view, computing will become laced into the fabric of home life and business operations. Today there are pockets of computing resources -- PCs, servers, smartphones, car navigation systems, surveillance cameras. Tomorrow, as the Internet of Things arrives, the pockets will instead be the places where there's an absence of computing power.
It's a bit sci-fi, but Moore's Law has made it economical to put more and more computing power onto small processors; smartphones are more powerful than PCs once were or than supercomputers were before that.
It's a potentially new market for ARM, which designs power-efficient chips and licenses the technology to others -- such as Apple and Samsung -- that actually build the products. This new market of computing devices will be tiny and relatively primitive, the kind of thing for designs like ARM's Cortex-M0 processor. So to make it profitable, ARM needs its licensees to make an awful lot devices.
Atkinson talked to CNET News' Stephen Shankland. The following is an edited transcript.
Q: Your work at ARM is to push the Internet of Things.
Atkinson: I joined ARM about 3-and-a-half years ago to look after the embedded marketing group. That covered things like automotive, industrialization, and medical markets. Our business model is licensing and royalties. In the embedded space, microcontrollers [small, basic processors] are headed south of 50 cents each. We only get a few cents of royalties on those, so we need to see huge volumes for it to have any significant impact on our revenue stream. When I looked at the next big growth opportunity, the Internet of Things was high on my list.
What could make it kick up? There are three core drivers. I'm looking at technologies that are disruptive, mostly coming out of new startups and small companies, because the larger companies aren't going to disrupt themselves out of business. Second, I'm looking at scalability. I'm not looking for vertical, niche applications or geographically restricted applications. I'm looking for something that can scale globally. Third, can we provide some impact with integration of these technologies, so it's not just a pure revenue-generating opportunity? There's nothing wrong with that, but can we use the Internet of Things technology to create some sort of sustainable impact on the globe. We have big global challenges: food, water, energy efficiency, health care, and transport.
What is your definition of Internet of Things? It's a very squishy term.
Atkinson: I look at it as a device that's able to transmit data without any human intervention. As the cost and size of these devices come down, as connectivity becomes more power-efficient so you can get batteries that last 5 or 10 years rather than months, then you can have a chip that provides not just identity but other information. They may be in sleep mode most of the time, but when something happens in their vicinity, they can wake up and transmit that data. The vast majority of devices in the Internet of Things are sensors and actuators. Industrial applications will dwarf the market for wearable computing and the kinds of devices that are connected over Bluetooth.
Which is more interesting to you, industrial or consumer? Industrial markets often end up more narrow and nichey, and you're interested in apps that scale globally. Often consumer markets are that.
Atkinson: Home automation and industrial automation are two very large markets. But by far the biggest volume of devices is going to occur with instrumenting large-scale assets -- things like water and agricultural monitoring, erosion monitoring for shellfish beds, forest fire detection grids. It's about putting sensors out in the real world and gathering data to make better decisions about how we use our natural resources, to be more efficient, to increase the well-being of people's lives. Once you can build a device for a few dollars and connect it to the Internet over a 10-20km range, and it only costs a dollar per year to get data from it, then the amount of innovation that will happen -- we can't even imagine that yet.
When I was asking about industrial, I was talking more broadly about commercial use -- what your farm is doing or where your airplane fleet is located, as opposed to consumer stuff like what's going on in my house or what's my heart rate.
Atkinson: There's going to be some significant bleedover for things like energy efficiency. The power generators, distributors, and retailers that sell us our electricity and gas will be putting all sorts of monitoring equipment in to make themselves more efficient. Regulation will encourage them to reduce their consumers' energy consumption. They'll put devices and controllers in our houses in exchange for lower tariffs to help with their targets of energy reduction. Those will appear inside consumers' homes, but they're part of the industrial Internet of Things. You could have remote health care for the elderly or infirm. That could be driven out a health service's efficiency needs, but it bleeds into the home. It's stimulated by industry rather than by the consumer.
What sorts of sensors are going to be big? Where will we see people wiring things up to report status? Temperature, strain, moisture levels?
Atkinson: In the short term, the usual suspects will be temperature, pressure, light level, noise level, humidity, or water level. The mobile phone world is shrinking the size and cost of gyroscopes and accelerometers. An accelerometer is almost by default put down on any development board to build a prototype because it's so cheap now. They're good for triggering movement or activity, and that could wake a system up to start measuring other things. Phones also have biometric fingerprint sensors, digital compasses, temperature sensors. Some of those will just happen by default. If you're building a device and a temperature sensor costs 20-30 cents for a device, the biggest cost for that sensor is putting a guy in a truck to deploy it somewhere.
Can you give some examples of things that will be measured using an Internet of Things network of sensors?
Atkinson: Agriculture is an example. Most water usage by the human race is in irrigation. India is the biggest consumer of water on planet. It uses twice as much water as China. China uses about a third more than the United States. The reason is almost all of India is used for growing crops. It only rains about a month every year. It fills the rivers and reservoirs, then they deplete that over the year. They potentially waste two thirds of that water because they use old-school irrigation methods like pivot irrigation. We'd like to move to drip irrigation, which puts the water down in ground much more efficiently. Most agricultural analysts predict that drip irrigation could reduce our agricultural water consumption by two thirds. That helps with watering crops today and with the 2 billion more people joining us by 2050. We need to put these systems in place or we won't be able to feed them.
The sensors themselves will be a spike you drive into the ground. It'll measure moisture levels at different depths in the soil, ground temperature, and potentially salinity. If you wanted to put a more expensive sensor in, you might measure nitrogen and phosphorus, which change according to how much fertilizer is in the ground. Moisture and temperature sensors will be deployed in hundreds of thousands across hectares of land to be very specific about where you water and where you don't. You build a good picture over time of which parts of a field drain quicker. This area could take a week to dry out, but another area dries out much quicker. You can tailor where you water and how much. We could significantly reduce water consumption and increase yields with precision agriculture. You get more food out of that piece of land.
The forest fire detection grid is being done today already in very high-risk places using Wi-Fi, cellular, or satellite connections and small chips. But they have a lifetime of a few months. They're deployed in the middle of summer when the risk is very high. They're kind of disposable. If there's a fire, the whole lot get destroyed. But the cost still way too high. If you have a low-cost radio with a range of 10 to 20 kilometers, a temperature sensor or temperature plus smoke sensor, you can build a small package with little hooks sticking out. You fly over the region you want to monitor and just throw them out of the plane or helicopter. They'll hook on to the branches of the trees and dig into foliage on the ground. Some might be a bit inaccurate, some may not work, but you drop them in numbers. You get a message that says this part of the forest is hot and detecting smoke, then you get another beep 30 meters further on, and now you've got the direction a fire's moving in. Rather than trying to infer forest fire behavior based on wind patterns and meteorological data, you get real information about where the fire is to within 10 or 20 or 50 meters.
How many of these would you spread over 1 hectare (about 2.5 acres)? Forests are awfully big.|
Atkinson: In the forest in the middle of nowhere, you might put only 1 every hectare or 10 hectares. When it starts to push on the boundaries of population or property, you increase the numbers. But these things could be manufactured for less than $5. Now you could deploy twenty for $100, whereas today you probably only get one for $100, and it has to be within a few hundred meters of an access point to get the data back. Insurance companies are very interested in supporting this.
From a consumer perspective, are people going to gradually become aware of uses of the Internet of Things? Or will there be some big aha moment as there was for smartphones in the last decade?
Atkinson: A few things could happen. With energy monitoring and energy efficiency, you'll see some early adopters. I already track how much energy I use, but it doesn't really help me, because I can't really reduce the cost of my energy consumption because there's no time-based tariffing. Until energy companies change to a model where I get rewarded for using electricity in middle of night when it's cheap to produce, then my behavior is not going to significantly change.
I think the concept of wearables is interesting. I've got a Fitbit and a Nike Fuelband. I don't wear it every day, because apart from knowing I'm increasing my activity level, there's not a lot to be gained from that for me. But imagine I could get a lower [health insurance] premium from Kaiser Permanente if I could show I'm doing more exercise than I did last year, or that I walk up the stairs rather than taking the lift. That could mean hundreds of dollars a year for me, and that could stimulate changes.
What about in people's cars?
Atkinson: You're already seeing some trends. In the UK, car insurance for 18-to-25-year-olds is ridiculous -- more than $5,000 a year. It's a very high-risk group, and they're all treated as equal even though they're not. There's a trend now: if you choose to have a black box put into your car, it will gather data about how quickly you accelerate, how hard you brake, whether you speed in speed-restricted areas, what time of day or night you drive. These are known by insurance companies to be significant factors in whether you're high-risk or not. I'll get significantly lower premiums if I drive carefully. That's the first wave of telematics data impacting me as a consumer.
The manufacturers would love to get ahold of this data. If I drive my car carefully and service it when my computer says to, why shouldn't I get a five-year warranty instead of a three-year warranty?
Things like using my phone to lock and unlock my car, or turning the air conditioning on in summer without leaving the house -- and we'll see those as cards get connectivity built into them. But if history has told us anything, it's that if you want to change consumers' behavior, you have to hit them in the wallet.
ARM makes money from licensing, and unit volume is important to you. What volume will you see in the Internet of Things?
Atkinson: It's certainly in the billions rather than the hundreds of millions. The key pieces of a sensor are the microcontroller [a basic processor] that does something with the data, the sensor itself, the power management system so it can last 5 or 10 years in field, and the radio itself. Three of those pieces are already well established in very low-cost model that can run 20 years on a battery and cost you a few dollars. Those bits are done. When the connectivity piece catches up, that's when we get billions and tens of billions of devices deployed. That's when the data we get back from them makes it absolutely worth doing. There are analyst firms that will tell you 20 billion, Cisco and Ericsson talk about 50 billion. In recent months, I've seen people talking about trillions of devices. That seems to be quite a ballsy number.
Trillions seems like a stretch to me. Maybe if you have smart dust motes floating in the atmosphere?
Atkinson: We've done a research project with a university for a sensor that's less than 1 millimeter cubed. It goes in the eye to measure pressure for glaucoma. It's got a battery, a microcontroller, a solar cell, pressure sensor. It's less than 1x1x1mm -- the size of a grain of sand. That's one application.
I was talking recently with a large construction company in Europe. They can lose between 7 to 8 percent of their concrete to theft -- a trucks takes a different turn, sells its concrete for cash, goes back and reloads, and nobody is the wiser. It's very difficult to track. And when they cast foundations, the [building] integrity is based on mathematical modeling and destructive testing after the fact. Concrete setting is exothermic, so if you replace the solar cell with heat-harvesting power generator, replace the pressure sensor with a temperature sensor and a strain gauge, then you could pour 150 of these into a mix of concrete. Maybe it has ID so you can track every load to reduce the concrete theft. You could also pull back really good data as the concrete is setting to get a really good model of how good the foundations are. You could reduce amount of material because you don't have to overengineer it like you do today. Five years later, you could warm the concrete by applying power to the steel reinforcement in the concrete, enough to generate energy that wakes the sensor, and transmit a bit of data. That's very interesting for the construction industry.
With the amount of concrete that gets deployed, there are tens of billions of devices just in that one one niche application alone.
You touched on energy harvesting to power devices. Heat is one possibility, solar is another, though I'm not sure the efficiency of solar cells is good enough for a tiny little device that's broadcasting a lot. There's also kinetic energy conversion from motion.
Atkinson: The vast majority of these devices are sensors gathering data and sending it somewhere. They're low-cost, small devices with microcontrollers to analyze data. They're in the 3-volt domain. The radios are either 3 or 5 volts. But if you could have that entire system working in a 1-volt domain, which is the natural voltage of a solar cell, then you have much less wastage in boosting energy up and down to power different elements in the system. When the device wakes up the radio and sends the data, if that takes half a second rather than a tenth of a second, does that make much difference? You could clock that microcontroller down to kilohertz rather than megahertz. You can scale things down to where it's using hardly any energy. If you have some sort of capacitor storage, you can harvest very small amounts of data around you. This is waking up once an hour, if it's changed in the last hour, I transmit it, if it hasn't I go right back to sleep.
If you can use these very low-power domains, you can gather energy from your surroundings and keep that capacitor topped up. That can be from kinetic energy. It can be from heat, where the difference between one surface and another is enough to generate power. Another big one being worked on with some success is RF [radio-frequency] harvesting -- the energy in the airwaves from TV, radio, and other wireless signals, the proper antenna can capture that and convert it into electricity. Another is materials -- if you apply pressure to a piezoelectric crystal, that will generate power. All those technologies are coming down in size and cost. Once they get to parity with the cost of a CR232 coin-cell battery, we'll see them deployed much more widely. That'll have an impact on deploying devices in places where we really don't want to go back and touch it again.
You're one of the leaders of the Weightless consortium for wireless communication technology. That supports low data rates but can link up lots of devices and over a long range, right?
Atkinson: I helped get Weightless started because of this connectivity gap. We want something at the price and power consumption of Bluetooth Low Energy -- something very cheap and very power efficient -- but with the range of cellular. In any wireless device, the thing that consumes most of the power is the radio. Internet of Things devices are very price elastic: you reduce the cost, and deployment increases by orders of magnitude. If it costs $100, I might deploy 10. If it only costs $10, I might deploy 10,000 because the data benefit far outweighs the cost of deployment.
We needed a radio that could instrument these large-scale assets like forests or near-coastal ocean waters, water treatment plants, oil refineries. If I only get less than a kilometer of range from an access point or base station, I have to deploy quite a lot of infrastructure. Whereas if I can fasten a base station to tree above the canopy, I can get a line of sight at a range of 30 or 40 kilometers, I can get hundreds of thousands of sensors in a forest for only one base station. That changes the game, and that's what Weightless was designed to do.
The world has changed since we set up Weightless. There are now a number of players offering the sort connectivity price points we were trying to enable with Weightless. People like Sigfox, Semtech LoRa, Onramp Wireless in the US. These are all examples of low-cost, low-power radio technologies that use the ISM [industrial, scientific and medical] bands to collect data from small, cheap devices -- somewhere between 100 bits per second to 1 kilobit per second. It's a very low data rate. It's about sending messages -- "I'm hot, I'm cold, I'm light, I'm dark, it's high pressure, it's low pressure." I may only send it once a day, once an hour, or maybe I won't send anything apart from a keepalive note once a day. As long as it only costs me a dollar or two to connect it, that would be in the hundreds of dollars. If I deploy 5 million of these, you could probably get down to 10 or 20 cents. So the operational cost is trending toward zero, the capital cost of the sensors is very low, the amount of infrastructure you need to put in to get the data out is an order of magnitude lower than cellular, and now I'm gathering 10, 100, or 1,000 times more data than I could -- that's where we see the efficiency.
Do we need five or seven different standards? Is this going to consolidate? Is Weightless head and shoulders above the alternatives? If there are seven different alternatives, the manufacturing volumes aren't going to be high, you won't get economies of scale, and the prices will be higher.
Atkinson: If people who want to deploy in volume have to choose between six different options, we're never going to get the critical mass we really need. We'll see some early leaders. There's some good technology there. Sigfox is deployed in France, and will be deployed in Spain by September. Arqiva are working on deploying in the UK. There are deployments in San Francisco Bay Area. Once there's ubiquitous coverage in a region, quite a lot of people can take advantage of that.
I love standards, but if I can by a radio off the shelf from Texas Instruments, Atmel, or Freescale that's Sigfox-certified and I can drop it into my device, and I'll only be charged less than $1 year to connect it, maybe that works for me and it doesn't have to be a standard at all.
I don't think it's useful to have six or seven solutions competing with each other. That's the problem we have with indoor networking. You have ZigBee, Z-Wave, Bluetooth Low Energy, 6LowPan. A lot of people haven't deployed because they're waiting to see how it all shakes out. We'd be keen to see critical mass to really grow to the volumes we'd like to see.
So is Weightless one of the contenders?
Atkinson: Weightless was designed to use the TV white space -- TV's unused radio spectrum. It's one of the most valuable pieces of spectrum on the planet. We need regulators to move more quickly on how they open that up for a wider range of applications.
We're looking at how can we encourage the ecosystem to take the specification and build something that means we can get some deployment. There needs to be movement in the next six months for it to stay relevant when you've got very strong competition from people that are already deploying.