Santa Clara, Calif. — Wireless sensor networks are moving from demo to production, propelled by the advent of low-cost radios. But it could be several more years before the technology achieves its anticipated status as the next big wave in computing and communications, ushering in a consumer tsunami of wearable, residential and automotive network nodes.

Before that wave can crest, researchers must focus more effectively on the enabling technologies. Chip makers are working now on the integrated silicon needed to lower the costs and power consumption of the tiny nodes. And core technologies and standards must be hammered out for everything from application programming interfaces to quality-of-service.

That was the conclusion from the first IEEE Conference on Sensor and Ad Hoc Communications and Networks (Secon 2004). Researchers and businesspeople used the gathering here last week to share ideas and debate the issues in two overlapping areas: wireless sensor nets, and ad hoc nets of any kind that can organize themselves into meshes (see story, below).

Sensor nets are being fueled by the availability of low-power IEEE 802.15.4 radios from multiple vendors and the imminent launch of the 1.0 version of the Zigbee protocol for them. Applications for the tiny sensor nodes range from anti-terror nets that would detect chemical, biological and nuclear threats to lighting systems that could be linked to motion and daylight detectors to cut energy bills.

Sensor networks "could represent the highest-volume computing marketplace ever," said Michael Horton, president and chief executive of Crossbow Technology Inc. (San Jose, Calif.), a startup that claims to have shipped half a million sensor nodes since its launch in 1995. "This technology has the potential to be much higher volume than PCs and cell phones, but it will take well beyond 2008 to get there," he said during a keynote.

Sensor nets will address such applications as asset tracking, agriculture, defense, and building and industrial automation. The total market will rise to at least 50 million units, or $1 billion, per year by 2008, Horton said.

"We have big expectations for the wireless sensor market," said Geir Forre, president and chief executive of Chipcon AS (Oslo, Norway), which supplies integrated radio chips to Crossbow and others. The company doubled revenue to $16 million last year, with 40 percent of revenue coming from wireless sensor nets. "We see the potential for hundreds of millions of nodes in a few years' time," Forre said.

The wave of consumer applications is expected to follow. "The automotive vision," for example, "is a wireless replacement for the CAN bus that lets all the sensors in a car talk to each other [without] today's expensive wiring," Horton said.

Some are already gaining traction, said Robert Poor, chief technology officer of Ember Corp. (Boston), which sells radio chips for sensor nets with technology licensed from Chipcon. The chips — many supporting the 802.15.4 standard — are going into prototype office lights, air conditioners and home controls from customers including Philips Lighting and Eaton Corp., at least one of which will start volume production this year, he added.

Freescale Semiconductor is also ramping production of the low-power, low-data-rate 802.15.4 chips that form the underpinning of networks using the Zigbee protocol. "We anticipate this will be a nice emerging market," said Brett Black, a manager of commercial wireless components.

"Until recently I thought this was a push area for research, but I have become convinced there's a tremendous pull," said Guru Parulkar, who directs the Networking of Sensor Systems program for the National Science Foundation.

Indeed, researchers from Nortel Networks and Sony Corp. were on hand last week to gather ideas about how their companies might leverage the technology.

One issue for sensor nets is the apparent disconnect between what some government researchers are targeting and what the sector needs. The Department of Homeland Security is taking the wrong approach to funding research, said John Strand, program manager of the SensorNet program at Oak Ridge National Laboratory (www.sensornet.gov/index.html).

"The DHS is spending on sensor development but not on sensor networks and interoperability," Strand said in an interview with EE Times. "They haven't looked at an overall architecture for how to put the pieces together. As a telecom guy, that scares me. We could wind up cobbling systems together in a way that's inefficient."

The two-year-old SensorNet program at Oak Ridge has deployed about 200 sensors in three test beds as part of an effort to establish a national sensor network for first responders to crises such as terrorist attacks. The Department of Defense has bankrolled much of the SensorNet effort. The program has a budget of about $20 million this fiscal year, expanding to about $30 million next year.

For its part, the National Science Foundation, chartered to advance the state of science and technology, is too staid in its approach to funding, said Parulkar of the Networking of Sensor Systems program. The agency has spent about $35 million a year on sensor net projects in the last two years, but it only funds about 10 percent of the proposals it reviews.

"When you take just one in 10 projects you don't want to make a mistake, so there is a tendency in the peer review process to pick 'flawless' proposals that guarantee results that are typically incremental," said Parulkar. "Thus there is a sense NSF is not funding enough high-impact research."

Chip makers are racing to deliver integrated silicon to drive down wireless nodes' cost and power consumption — two of the biggest inhibitors to development of the sector.

Ember expects to sample chips early next year that provide a baseband and RF front end for 802.15.4 radios, as well as an 8-bit XAP controller as part of Ember's acquisition of a design team from Cambridge Consulting Labs. "That could help bring the cost of a module down from $10 to $5," said Poor.

Ember is in a dead heat with Norway's Chipcon in the race to be first to deliver such integrated chips.

"We also believe the market for Zigbee and wireless sensor nets will be served by system-on-chip solutions," said Chipcon's Forre. "We have several solutions on our road map with the radio, controller, flash and RAM — everything you need."

Those chips will have to support a variety of interfaces — many requiring external circuitry — to handle the breadth of sensors people want to deploy. "It's not realistic at this point to define a generic sensor interface," Forre said. "You would be making too many bad compromises."

Startup Crossbow has started its own ASIC, apparently aimed at a follow-on generation of the Ember and Chipcon devices. Horton said the company is working on a design that includes a radio and an alternative processor with a memory subsystem geared to the needs of sensor nets. When the ASIC appears in two years, it could help drive the company's $20 sensor nodes down to single digits, he said.

Freescale sees the opportunity as well, though it is moving more slowly in efforts to capitalize on it. "We are looking at single-chip solutions not just for cost but for power reduction as well," said Black. "We believe that on the radio side you could reduce power consumption by 50 percent, and that's conservative. We will address this in the next couple of years."

Many of the base capabilities of wired networks, including routing, security and quality-of-service, are still in a primitive stage for sensor networks. "In our current round, we had very, very few proposals on privacy and security in sensor networks and only one we funded," NSF's Parulkar said, calling for researchers to put more focus on the area.

The SensorNet group at Oak Ridge Lab is starting work on a policy server for sensor nets that would be managed at the local level in tandem with a security server, said Strand.

With so much work on fundamentals, too little effort is being spent on applications and ease of use, some researchers said. "I'd like to see more work on integrated self-configuration capabilities so you can just take a network to an area and it will work," said Mark Yarvis, a researcher working on sensor nets at Intel Labs.

Rising to the challenge, researchers at the University of California-Riverside demonstrated eBlocks, a set of plug-and-play modules designed for easy configuration of sensor networks for jobs like remotely monitoring a doorbell or a pet. Half the users in the test were able to use the modules to build working nets in less than 10 minutes. The blocks, based on 8-bit Microchip PIC controllers, could cost as little as $4 in high volumes and run for two years on a 9-volt battery.

Oak Ridge's SensorNet group is defining API requirements for sensor networks used in embedded control. It expects to have a prototype API in about nine months.