Quantcast
Channel: Articles and Op-Eds
Viewing all articles
Browse latest Browse all 10

Intel's Tiny Hope for the Future

0
0
December 31, 2003
The microprocessor giant is thinking even smaller: tiny sensor chips that network with each other --
December 31, 2003

As a department head at the Defense Advanced Research Projects Agency, the Pentagon's R&D arm, David Tennenhouse spent the late 1990s approving or denying funding for hundreds of far-out military programs. One proposal he reviewed, from a research team at UC Berkeley, outlined a concept called smart dust -- fleck-sized wireless sensors intelligent enough to organize themselves into autonomous networks. Dropped from a passing helicopter, the sensors could spy on enemy movements or detect a hidden stash of mustard gas. Tennenhouse was intrigued enough to authorize several hundred thousand dollars in funding. Then he moved on to the next bizarre proposal.

Tennenhouse left Darpa in 1999 to found Intel Research, the semiconductor giant's stab at offbeat R&D. Charged with finding up-and-coming growth technologies, he gave little thought to smart dust. A neat plaything for the Pentagon folks maybe, "but not all that relevant to my new role at Intel," he recalls thinking. A network of minuscule sensors, each containing only a few dollars' worth of circuitry, just didn't seem like a moneymaker.

That is, until August 2000, when Tennenhouse was invited to Berkeley to check out a student-designed mote -- the housing that contains a sensor assembly and a radio antenna to allow it to communicate with other motes. While examining the circuit board, something clicked. If motes could get significantly smaller -- say, small enough to fit inside pill-bottle caps -- they'd be unobtrusive enough to go anywhere. And that, Tennenhouse thought, would mean a windfall to the company providing the processors. Intel could dominate the high-volume sensor market just as it has ruled the high-powered chip scene with Pentiums.

This time, Tennenhouse threw millions at sensor research. He set up an Intel facility at UC Berkeley and told his researchers to use their imagination to develop new configurations and applications. Three years after that landmark trip to the East Bay, Intel's sensor investment remains trivial compared to the billions it has plunged into, say, the new Centrino wireless chip or its Flash memory business. The entire annual Berkeley lab budget is about $5 million. But in this case, Intel is acting more like an angel investor -- one with $15 billion in the bank -- feeling around in the dark for something that's not only different, but transformative. "It's not a high-end computer play or a low-end mote play," says Intel CEO Craig Barrett. "We're looking for growth opportunities outside our core competency."

Intel is hoping for a two-tiered payout. If sensor networks take off, that will create a need for more silicon. But the networks will also generate a huge amount of data, driving demand for more high-end PCs to process it all. The company now foresees networks consisting of thousands of motes, located wherever there's a need for data collection, streaming real-time data to one another and to central servers. Intel imagines the day when every assembly line, soybean field, and nursing home on the planet will be peppered with motes, prodding factory foremen to replace faulty machines, farmers to water fields, and nurses to check on something unusual in room E214.

Barrett's CTO, the perpetually caffeinated Patrick Gelsinger, can't contain his excitement about the sensor play. Like Tennenhouse, he was converted from skeptic to devotee after a single demo. "Intellectually, it probably felt like Gordon Moore's experience after he first saw the microprocessor and started envisioning all sorts of uses for it," says Gelsinger. "I was a believer." Now he hopes to produce "tens of billions of units" by the end of the decade.

Listen to technologists like Gelsinger and Tennenhouse long enough and the outlines of a strategy emerge: By thinking small, Intel could be on the cusp of something huge. Which is exactly where Intel likes to be.

Intel Research Berkeley is just one of four university-based research "lablets," so-dubbed for their relative size and independence, set up by Tennenhouse beginning in 2001. The others -- at the University of Washington, Carnegie Mellon, and the University of Cambridge -- are likewise dedicated to developing disruptive technologies. Though wholly funded by Intel, the lablets aren't expected to push products down a pre-determined pipeline. As long as their work is tangentially related to silicon, the research centers can pursue whatever they wish. In England, the projects include optical switches and new programming languages; in Pittsburgh, it's data mining tools for webcam networks; in Seattle, XML-based messaging and wireless personal area networks. In Berkeley, sensor networks rule.

The Berkeley lab occupies a plum penthouse space adjacent to the university, with a gorgeous view of San Francisco Bay. Joe Hellerstein, the director, leans back in his chair and looks toward Sausalito. The Golden Gate Bridge, he notes proudly, will soon play host to an experimental sensor network designed by professor David Culler and his students. "It's hard to tell just by looking at it, but the bridge actually sways a few feet in high winds," he says. "The motes, they'll measure how far it moves to either side."

By this June, a stretch of the bridge will be lined with about 200 motes. Each will contain chips running at 8 megahertz and accelerometers designed to measure movement caused by strong gusts. Those readings will be radioed to nearby supermotes -- data way stations that are 25 times more powerful than a regular mote -- before being relayed to the central server. If one reading seems out of whack, it could be a sign of a structural weakness that needs addressing before the next big earthquake.

The Golden Gate project will join several other trials already under way. Berkeley researchers installed 80 motes in the redwood groves of Sonoma County, measuring the temperature and humidity around the mammoth trees. An agriculture project placed 65 motes in a British Columbian vineyard, ready to buzz the manager's laptop whenever an early-morning frost looms. The health research team has mocked up a home of the future at Intel's Hillsboro, Oregon, campus, which uses wireless sensors to remind Alzheimer's patients how to make tea or take their medication. And about 100 UC Berkeley motes help monitor the nesting burrows of Leach's storm petrels, elusive seabirds that breed off the Maine coast.

What's Intel's interest in all this? Think of Maytag bankrolling research into cheaper prefab houses: More homeowners means more dishwashers. "Do I care about ducks in Maine? Absolutely not," says Gelsinger. "Do I care about exploring applications so we can start to demonstrate the value of these networks? Absolutely. What I'm really interested in is selling a lot of silicon for these little devices here."

UC Berkeley's relationship with Intel is governed by an open collaborative research agreement, which requires that nearly all of its work be nonexclusive. Sensor net software, in particular, is flourishing in this open environment. The motes are powered by TinyOS, a Berkeley-designed, stripped-down operating system built to run on 128 kilobytes of memory. Hellerstein estimates that only a few dozen programmers worldwide have mastered TinyOS. But that number is growing rapidly as curious geeks download the code from open source software development site SourceForge.net. Hundreds of downloaders are experimenting with TinyOS, and Hellerstein credits non-Intel developers with honing the motes' ability to coordinate their relative positions to one another. He hopes a compatible database program, TinyDB, will get similar treatment.

The lablet has also farmed out its hardware needs, striking a deal with San Jose's Crossbow Technology. Crossbow agreed to make a few hundred motes for researchers at a cut rate; in return, it received an undisclosed investment from Intel Capital.

Sensors can't become the next big thing until a host of mundane technical issues are resolved: How to get the chipset radios off the crowded 900-MHz spectrum? How to program the networks to not just spew reams of information but be intelligent enough to figure out which measurements are vital and which are junk? "The challenge in tiny sensors is doing some computation at the level of the motes," says Hellerstein. "It's just too expensive to ship out all the data."

Then there's the power issue. Despite TinyOS's relatively scant memory requirements, the motes still burn through batteries far too quickly. And even if they didn't, there would be no room for alkaline AAs in a sensor-filled future. There's talk among researchers of using solar cells, or even MEMS devices that harvest imperceptible vibrations in the environment, but no ready solution.

The toughest challenge of all is creating motes clever enough to function without constant tending. "How do you ensure that people build motes that can be deployed by a local farmer?" asks Richard Beckwith, who heads Intel's sensor experiments in vineyard maintenance. "That's a really hard problem. I'm sure it's going to happen, but I'm not sure how."

"We used to joke inside the company that Intel is the largest single-cell organism on the planet," says Gelsinger. The joke, like many engineering cracks, isn't exactly ha-ha funny, but it is revealing. For all of its runaway success with processors, Intel has had a tough time diversifying.

Despite a triumphant third quarter, the firm realizes that the desktop processor market is starting to plateau. In the not-too-distant future, Moore's law could become too pricey to maintain; there's no point in building yet another multibillion-dollar fab if its chips are too expensive for the lion's share of users. The company needs new industries to colonize, and sensors are a logical first step. Intel won't say exactly how much it has put toward sensor nets so far, but it's certainly less than $50 million. "That's not even Intel's latte budget," says G. Dan Hutcheson, president of VLSI Research, which tracks the semiconductor industry.

The good news about sensors as a diversification play is there's no competition to fend off; other companies have shown little initiative in developing the technology. And sensor nets certainly make for impressive demos. At an Intel Developer Forum early last year, Gelsinger wowed the audience by releasing a swarm of sensor-studded beach balls. As they bounced around, the balls beamed their ever-shifting coordinates to a base station, which in turn charted each orb's movement on a large display. It was proof, Gelsinger announced, that sensor networks can self-configure on the fly -- the key to fulfilling his vision of motes that can be deployed like pixie dust.

Right now, the motes are a lot closer in size to golf balls. Which means that the company is looking for short-term uses that don't require the sensors to be ultrasmall -- or ultracheap. The example that Gelsinger loves to bandy about is an assembly line outfitted with $200 motes designed to gauge machine vibration. If a robot's rumble feels out of sync, the motes relay a warning to a supervisor's laptop: Machine number 27 will break down in the next 30 days. "Now, this piece of equipment costs a million dollars, and it's a disaster if it breaks," he says. "In that case, a $200 sensor is a very cheap device. I mean, it's trivial if you can actually solve this nightmare problem."

But Gelsinger knows that assembly-line motes will do little to burnish Intel's rep as a trailblazer. The hoopla, in large part generated by Gelsinger's fervor, is over smart dust, sensors shrunk down literally to the size of specks. The goal is to halve the size and price of a mote every 18 months, which would have each unit about the size of a grain of rice and selling for around $5 by 2011. That would enable the company to scatter tiny self-configuring hardware as casually as corn seeds. "I want to get to where I'm cranking up huge fabs to deliver, you know, Intel Inside Band-Aids," says Gelsinger. "Let's say a Band-Aid today costs a penny. In the future, they'll sell 10-cent Band-Aids that are also interactive heart monitors. We'll get to a price point where everything starts to have sensors. And when we get there, Intel will be building enormous fabs for consumables as well as ones for maintainable devices."

Has Gelsinger been reading too much Neal Stephenson? "Intel's whole business is built around selling $200 processors," says Linley Gwennap, president of the Linley Group and former editor in chief of Microprocessor Report. "You have to sell an awful lot of $2 processors to make the same amount of money. Sure, if you put them in Band-Aids the volume is going to be tremendously high. But it's not a slam dunk that the mathematics will make it profitable."

If the prospect of ubiquitous silicon seems frightening because you're already flummoxed by the constant beeping from your Outlook inbox, BlackBerry, and cell phone, not to mention the tangle of noncompatible software and breakdown-prone accessories, don't sweat it. Deployed the way Tennenhouse envisions, the networks will require zero human input. We'll reap the benefits without having to interact with the networks, and Intel will eliminate a long-standing obstacle to its growth: the feebleness of the human brain. "Sensor nets let us relieve the human being of the responsibility of drawing information out of the physical world," says Tennenhouse. "We need to have computers anticipating our needs and sometimes taking action on our behalf."

Once sensor nets liberate humans from the burden of computer interaction, there's no limit to how much silicon Intel can push. "I'm not going to have 10 computers per person," Gelsinger froths. "I'm going to have thousands of computers per person."

That vision represents a departure from business as usual in Santa Clara, California. Intel grew into the bluest of blue chips by hewing to Gordon Moore's famous observation about transistors, an axiom that has become a corporate strategy: Keep on building a chip that's twice as fast every 18 months, and everything will be all right.

The sensor game plan, on the other hand, has little to do with speed and a lot to do with size. Yes, the silicon that goes into each mote may eventually be dirt cheap. But when those chips are small enough to be blended into house paint or sewn into blue jeans, Intel will sell them by the billions. Smart dust?

It could be ingenious.

URL: 
http://newamerica.net/node/6306

Viewing all articles
Browse latest Browse all 10

Latest Images

Trending Articles





Latest Images