During its brief flight, the outer skin of a missile can reach 300° Fahrenheit, creating an ideal source for thermophotovoltaic conversion cells that can be used to power guidance electronics. Piezoelectric materials can be built into smart ammo shells, so that recoil forces become a power source. And electronics in a foot soldier's clothing can be powered by "heel strike" electromechanical systems built into the boot.

Those are some of the far-out concepts military engineers are exploring in the emerging field of energy harvesting, in which extreme environments are bent to the needs of electronics. The commercial sector, too, is beginning to eye energy harvesting for many of the same reasons. The principal power source for untethered electronics is batteries, which tend to be bulky and have a limited lifetime. The idea behind energy harvesting is to make electronic systems self-sustaining.

The environment provides ambient energy in a wide variety of forms: vibration, strain and inertial forces, heat and light, wind and magnetic fields. And there are a variety of systems that can tap those energy sources. Piezoeletric materials can convert strain and vibration directly into electric current. Magnets and inductive coils can tap inertial forces; thermo- and photovoltaic cells can harvest the energy in light and heat.

It would be a significant breakthrough to find some way to use these ubiquitous energy sources to eliminate the battery from mobile computing devices and wireless sensors, but doing so is proving to be a formidable task. One problem is economic: Batteries are simply better at providing reliable power at the lowest cost, and it will be difficult to dislodge them. But even when the economic factor is removed, as in military systems, it is still difficult to get electronic systems to run on ambient-energy sources exclusively.

"About a decade ago, it seemed clear to me that there was a technological path to integrating sensing, computation, communications and power into a millimeter-scale package. All of the technology drivers were going in the right direction, following Moore's Law-type exponentials down to zero size, power and cost," said Kris Pister, founder and CTO of Dust Networks Inc. (Hayward, Calif.), a wireless-sensor network company. "I coined the term 'smart dust' to describe where all of that was headed."

While at the University of California, Berkeley, Pister began to push down the power consumption of circuits and has continued to set records. At the recent International Solid-State Circuits Conference, his group at Berkeley presented a paper on a 2.4-GHz radio receiver that burns only 200 microwatts.

The university work explored the lower limits of power consumption in terms of the circuits themselves. The objective of research and development at Dust Networks was to build wireless-sensor networks, and reducing the power used at the individual nodes was just part of the overall energy problem.

"The university work was all about low-power hardware; the company work has been about low-power software," Pister said. "We came up with communication protocols that keep the radios off 99.5 percent of the time. Low-power software running on low-power hardware--at that point the power consumption goes down into the single-digit microamp kind of range."

However, even with the ultralow-power architecture, the nodes still need batteries and Pister doesn't see that changing anytime soon. "Whatever you are going to replace batteries with is going to be more costly," he said, "and most energy-harvesting systems are still in the development stage." Low-power operation greatly extends battery life, and energy harvesting could further amplify that. But one would still have to come up with an economic rationale for adding an energy-harvesting component to extend battery life.

Pister is looking at using photovoltaic cells in that role. Even inside buildings, the ambient-light levels are high enough to gain a distinct benefit from photovoltaics, the most mature energy-harvesting technology, with a long development history. Indeed, photovoltaic technology's struggles over several decades to find a niche in the energy field is a good indicator of the hurdles other, less-mature energy-harvesting approaches will have to overcome.

"When people talk about energy-harvesting systems, they usually overlook photovoltaics," observed Michael Robinson, vice president of marketing at MicroStrain Inc. (Williston, Vt.), which is also developing wireless-sensor networks for industrial applications.

Photovoltaics, while available and relatively low in cost, do not solve some of the problems MicroStrain has encountered, however. The company has developed high-performance piezoelectric materials and is applying them to building strain sensors that can be implanted in structural elements to monitor their integrity. For example, MicroStrain hopes to be able to place such sensor networks in highway bridges, where there is a significant level of vibration. Highway engineers could then read out data on structural integrity using a wireless unit.

The material is also being eyed for harvesting mechanical energy in a variety of applications.

Neither photocells nor batteries are a good match for such applications, but engineers at MicroStrain have found a way to run wireless nets entirely off the energy harvested from vibration. The system uses a cantilevered beam with a weight attached to the end to amplify vibrations. As the beam bends, it generates electric power. This work is still in the research stage.

"At the end of the day, wireless networks will always be hampered by the need to change batteries," said Robinson. "Harvesting energy is the only way to avoid that."

Problem of leakage
Broad-scale energy harvesting can't truly take hold until low-power design becomes entrenched at all levels of electronics, from systems to circuits. And that is beginning to happen, notably in microprocessor design.

One trend in the circuit business that isn't following a Moore's Law curve is current leakage, which has a direct impact on static-power consumption. Increasing circuit density only drives up the leakage. The problem cuts across applications and is particularly troublesome for any system that needs to run off low, intermittent power.

"We are all trending into smaller geometries so we are getting many, many more transistors on a piece of silicon. The bad news is that leakage is growing quite rapidly as a percentage of overall power," said Rick Hetherington, chief architect for Sun Microsystems Inc.'s new Niagara processor, which is optimized for low-power operation. Niagara is going into large servers and data centers, a far remove from the remote applications of energy-harvesting schemes. However, system designers can no longer take an endless power supply for granted.

"Many of the clients we have been talking to have been filling up their data centers--they are running out of space, they are running out of power," Hetherington said. "The next step up for them is quite expensive--it means building new data centers."

Banks on parallelism
The Niagara project targeted that problem by using architectural innovations to multiply performance while cutting power, which would allow data centers to continue expanding their information-pro- cessing capacity without increasing floor space or power consumption.

Built in a 90-nanometer process by Texas Instruments Inc., the Niagara processor has eight cores supporting 32 threads, a type of parallel architecture that is well matched to networking. Instead of pushing processors to run ever faster to get better performance, the design philosophy behind Niagara is to use parallelism at higher system levels to control processor speed and power consumption.

"I think people in the industry are giving up on chasing more gigahertz and are starting to realize the important role of power dissipation," Hetherington said. Not only do servers burn a lot of power, that power is turned into heat, which requires more power to run cooling systems. And the continuing explosive growth of the Internet is becoming a major power issue. "Especially in the New York City area, when you are talking to Wall Street people, they are all very keen on the power consumption issues in their data centers," Hetherington said.

The next frontier for power reduction will be at the circuit level, he believes. Both Intel Corp. and Advanced Micro Devices Inc. are introducing new power-saving strategies in their CPU designs. "Intel has been particularly aggressive in attacking the leakage problem," Hetherington said.

AMD's Opteron processor has a variety of power-saving enhancements. Its PowerNow management system dynamically reduces power based on the workload of the processor. Leakage is attacked by using silicon-on-insulator technology, and efficiencies are wrung from memory access methods and direct I/O operations.

Dig two holes
One company taking an innovative approach to leakage is Transmeta Corp. (Santa Clara, Calif.), which specializes in low-power processor technology. "Suppose you dig two holes in the ground that are a certain distance apart. You pour water into one and it starts to diffuse through the soil toward the other hole. That basically is the leakage problem for semiconductors," said Aashish Patel, director of technical marketing at Transmeta. Reducing feature size only drives the holes closer together, and the properties of the substrate determine the rate of diffusion.

"You could say that some semiconductor processes give a clay soil while others are more porous, like sand," Patel said. "What we can do is allow you to choose what type of soil you have dynamically. You are no longer committed to one type--clay or sand--by the process you choose." By allowing the system designer to set certain transistor characteristics, trade-offs between leakage and performance can be used to optimize processors for a given applications.

It is clear that whether the concept of energy harvesting means windmill farms, photovoltaic arrays at one end of the power scale or power for tiny motelike sensors at the other end, it will heavily depend on complex system-level innovations to move ahead. For example, the military is developing a standard control architecture for managing harvested energy from a variety of sources. One of the factors holding back the acceptance of photovoltaics in residential applications is the difficult technical problem of balancing the power output. Windmill farm builders are turning to an efficient energy-balancing system employing high-critical-temperature superconductors, which is blossoming into a new market area for American Superconductor (Westborough, Mass.). And mainstream processor makers are now scrambling to build energy efficiency into a broad spectrum of processors.

Perhaps all this engineering will converge on some ultimate, energy-independent form of electronics.