June 27, 2003

Sensors of the World, Unite!

Q&A: Ember CTO Robert Poor on turning wirelessly interconnected networks of sensors into a ubiquitous reality.

By Technology Review

Imagine sprinkling tiny sensors on road and fields for surveillance, putting them in buildings and bridges to monitor structural health, and installing them in industrial facilities to manage energy, inventory and manufacturing processes.That's the idea behind the emerging technology of wireless sensor networks (see "Casting the Wireless Sensor Net").

Boston-based Ember is at the epicenter of this field. The MIT spinoff sells radio chips with embedded processors that can organize themselves into networks to manage real-world data from sensors. Ember CTO Robert Poor-whose past life includes stints as a programmer in the computer graphics group that became Pixar and as a guitar technician for the Grateful Dead's Jerry Garcia-spoke with Technology Review staff writer Gregory T. Huang about his visions of a world filled with wirelessly networked devices.

TR: How did you first get involved with self-organizing, wireless sensor networks?
Poor: When Andy Wheeler and I were students at MIT, there was a DARPA [U.S. Defense Advanced Research Projects Agency] program called SensIT, which was looking at sensor networks. One application was battlefield-awareness systems, which really needed an easily deployed, reliable mesh network architecture. It was funding from that that allowed Andy and me to go from software simulations to hardware prototypes. We built hardware systems while we were still at MIT.

TR: How did Ember get off the ground?
Poor: Ember was started just over two years ago. At first, if we talked about wireless device networks, people would look at us strangely and say, what's that all about? So we spent the first part of our time proving the technology to people and getting our very first platforms out there. Since that time, some things have changed-people no longer look at us strangely when we say wireless device networks. By the way, I do have a small issue with the term "sensor networks"-the term suggests just one-way communication, from sensor to collection point. Ember's networks  provide two- way communications. This is important for calibrating sensors, monitoring network conditions, and controlling devices. Our customers have needs that go beyond just sensors, and they actually have to control stuff.

TR: Ember's customers are primarily in supply chain management, automated meter reading, and industrial and building automation. What are the most exciting applications?
Poor: Well, money is exciting, even if the applications may be mundane! One of our customers is Tyco Thermal Control. They make heating tape that you wrap around pipes. Not a big deal except when you're in a petroleum field in northern countries where it gets cold, and you have to keep the stuff in the pipes warm so it doesn't get too thick. Their problem is that they have to have temperature sensors that are wired to a controller, which communicates back to the pipe heating tape. In an oil separation plant, it would cost them between $3 and $10 a foot to run the wire, and there's on the average of 1,000 feet per temperature sensor/tape combo. So that's $3,000 to $10,000 per temperature sensor-for just the wiring. And in their case, they don't own that part of the process. That's actually done by a field installer, so it's subject to errors. For example, if an installer installed a temperature sensor near a steam outlet, it would read that it was too warm, and so it wouldn't heat up the pipe. The stuff in the pipe would get too thick, and a tank might burst. The plant could be down for six days at a cost of $100,000 an hour, just because the technician installed a temperature sensor in the wrong place. So what Tyco's finding with our technology is that they can put down redundant temperature sensors, use two out of three votes to get a reliable temperature reading, and the installed cost is still less than the wired equivalent.

TR: Do you see these systems pervading our everyday lives?
Poor: Yes-much in the same way bar codes pervade our everyday lives. I bet 30 percent of the things on your desk have bar codes in them right now, but you don't think about them. They're just there. In your office right now, you look around, you see lights and switches. Those would be wirelessly networked. You see smoke detectors and heat detectors. Those would be networked. You see air conditioning systems and thermostats. Those would be networked. These are all existing infrastructure. And the interesting thing that will happen is that things that are not currently part of any kind of network will also be networked.

TR: Like what, for example?
Poor:  There are microcontroller chips everywhere. During 2001, DARPA figures that there were 150 million CPU class chips sold.   But during the same period of time, 7.5 billion embedded microcontrollers were sold.  There are about 50 in your car. You probably have 100 around your house right now, just in little things like your toaster and your VCR controllers. Those are all candidates for being networked, because if the last two decades have taught us anything, it's that connectivity, not computation, makes something valuable. I could give you the fastest computer in the world with no Internet connection. How useful is that?

TR: How might these embedded networks have an impact?