Issue No.01 - January-March (2005 vol.4)
Published by the IEEE Computer Society
Roy Want , Intel Research
Keith I. Farkas , Hewlett-Packard Labs
Chandra Narayanaswami , IBM T.J. Watson Research Center
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/MPRV.2005.12
Pervasive computing aims to integrate computation into our daily work practice to enhance our activities without being noticed. In other words, computing becomes truly invisible. Yet at the heart of every pervasive computing system are electronic components that consume energy. Managing the energy needs of mobile systems, or systems for which reliable power isn?t guaranteed, can be a significant distraction for users. How can we minimize user involvement in the energy management process to make pervasive computing devices more pervasive?
Pervasive computing aims to integrate computation into our daily work practice to enhance our activities without being noticed. In other words, computing becomes truly invisible. Yet at the heart of every pervasive computing system are electronic components that consume energy. Managing the energy needs of mobile systems, or systems for which reliable power isn't guaranteed, can be a significant distraction for users. How can we minimize user involvement in the energy management process to make pervasive computing devices more pervasive?
POWER MANAGEMENT TECHNIQUES
Most users can attest—from their experience with mobile phones, portable media recorders, and portable media players—that power management can be distracting. Of course, the magnitude of distraction depends on the specific application, the hardware, and the energy source's characteristics. Researchers are trying to address this issue predominantly by reducing power consumption and developing improved energy sources. Unfortunately, both approaches have limitations.
Reducing power consumption
Attempts to reduce power consumption have focused on hardware that consumes less power and software that judiciously manages power consumption. Advances in VLSI fabrication and the reduction of a transistor's dynamic power consumption have led to dramatic decreases in the power that electronic circuits consume. This in turn has allowed pervasive systems to deliver greater capabilities on a fixed power budget, or similar capabilities on a smaller power budget. However, in the future, it remains unclear to what extent this trend will continue. Static power consumption is becoming a large fraction of total power consumption, because chips will consume more power even when they are idle, yet we know of no long-term technical solution. In addition, as users become more accustomed to the convenience of wireless capabilities and transfer large amounts of data over wireless interfaces, physical laws will dictate the minimum amount of energy that needs to be transmitted. And VLSI circuitry can't overcome these limitations.
Energy-aware software is a compatible technique in which the software is designed to conserve energy where possible. For example, the software might judiciously power-down components not in use, or engage power-saving operating modes when higher-power modes aren't warranted to deliver a lower quality of service. The challenge for such software is to limit the degree to which these optimizations distract users and to limit the amount of user attention and configuration that's needed. This problem is further exacerbated when mobile devices evolve from single-purpose devices to multipurpose computers with operational characteristics similar to those of desktop computers.
Improving energy sources
Today, batteries represent the dominant energy source for systems that can't be readily powered from a household electricity supply or for which power cables haven't been laid—at a remote weather station, for example. Although exponential improvements have occurred in hardware components (for example, in processor speed, memory density, and network bandwidth), such improvements haven't occurred in battery technology, and we don't anticipate any significant changes in the future.
At the same time, owing to shrinking component sizes and the consumer electronics industry's persistent integration of more features and capabilities, our expectations for mobile systems continue to outpace conventional batteries' energy storage capabilities. This trend has given rise to the development of alternative energy-storage solutions, such as direct methanol fuel cells. However, these cells are no panacea because, like batteries, they require user management and hence distract from the application at hand. Moreover, the energy density of batteries in some mobile devices is already very high, and safety concerns arising from rapid accidental discharge would become more serious with significant increases in energy density.
ENERGY HARVESTING: A PROMISING NEW APPROACH
Fortunately, techniques exist that, given the appropriate applications and modes of use, can significantly reduce or simplify user involvement in the energy management process. One such technique is energy harvesting (or energy scavenging), which aims to collect ambient energy to help power systems, possibly storing energy when it isn't required—typically in batteries, capacitors, or springs.
Many energy transducers exist, each offering differing degrees of usefulness depending on the application. Perhaps the best known transducers are solar cells, which have long been used to power simple hardware components such as calculators, emergency telephones along the highway, and low-grade lighting for pathways at night (charged during the day). Another type of transducer converts the energy contained in a vibrating object into electrical energy—researchers have used these to harvest energy from floors, stairs, and equipment housings.
A third type harvests mechanical energy, such as that produced by a person walking and an object's movement—for example, self-winding watches, which wind themselves from the swing of a person's arm. Hybrid cars that transfer energy from the engine to the battery during braking are a more modern example of this technique, which has been used in commercial vehicles. Yet another type of transducer is one that converts the momentum generated by radioactive reactions into electrical energy. Some transducers can also exploit temperature gradients, or pressure gradients, to produce electrical energy.
Energy harvesting is most applicable to applications that demand small amounts of continuous power or that have short periods of high-power use, which previously harvested and stored energy can provide for. Energy harvesting can also supplement more conventional energy sources—for example, when a mobile computer is in a low-power sleep state or charging its battery. These supplements might offer significant contributions to a computer's energy needs, given a suitable usage pattern. For instance, if a mobile user extensively uses a computer for short periods of time, an energy harvesting system might be able to ensure the battery is always topped up during the standby period when the computer isn't in use.
Although energy harvesting techniques can significantly benefit some systems and their associated applications, they're limited by the efficiency of the transducers and the raw energy available. As we discussed earlier, a complementary technique for extending battery lifetime is to design software that judiciously manages power consumption. Energy conservation summarizes the numerous techniques in this area—such as shutting down disk drives after a time-out period or slowing down a processor's clock. Given that the energy contributions of energy harvesting will be modest and will likely be serendipitous depending on the location of use, conservation techniques must be used hand-in-hand with energy harvesting to make best use of its contribution.
The semiconductor industry believes feature sizes for semiconductor devices will continue to decrease for several years, although the rate at which we can build them, owing to significant manufacturing complexity, is slowing. The result will be continued benefits in switching power, but with current leakage also beginning to play a significant role. This means that power dissipation relates not only to the switching frequency but also to the area of a chip that's actively connected to supply lines. Consequently, we might have to balance the need for more capable electronic circuits that match user expectations against the power cost of delivering the associated increased integration.
As more VLSI designs are built with power characteristics that lend themselves to low-power operation, the commercial opportunity for energy harvesting will become apparent. This will drive interest in improving existing energy transducers, perhaps with more efficient components, or in searching for fundamentally new materials with improved energy conversion properties. Examples include solar cells with greater than 20 to 30 percent efficiency, perhaps based on nonsilicon solutions, or a better piezoelectric material than the commonly used PZT (lead zirconate titanate).
Better understanding of solid-state physics and refined manufacturing techniques will also play a role in improving efficiency. As demand for the personal computer's processing power continues to increase, demand for improving the quality of the manufacturing process also increases to guarantee the required high yields of silicon chips containing tens of millions of transistors. These improved processes have a secondary beneficial effect for other device types, including those used as energy transducers.
Although energy harvesting isn't a solution for all energy needs, it's a compelling concept that will increasingly enhance or even replace batteries for lower-power applications as low-power circuit design techniques and transducer energy efficiencies improve. (Compare, for example, early battery-powered electronic calculators and the solar-powered products sold now.) If the economics are right, for some types of transducers it might just be a question of applying known material characteristics in novel combinations to improve existing energy harvesting techniques.
Energy harvesting is a nascent technology—currently, we can harvest and use only modest amounts of energy. However, the increasing use of pervasive electronics will give this area more attention as researchers try to overcome the difficulty that arises in managing power for numerous devices. Then, as we build new devices with improved efficiency, the amount of energy we can harvest will increase, and eventually we should be able to deploy these technologies in a wide variety of settings.
We hope you find the articles in this issue (see the related sidebar) interesting and contribute to further advances that lead to widespread use of this technology.
Roy Want is a principal engineer at Intel Research, where he leads the Personal Server project. His research interests include ubiquitous computing, wireless protocols, hardware design, embedded systems, distributed systems, automatic identification and microelectromechanical systems. He received his PhD in computer science from Cambridge University. While at Olivetti Research and later Xerox PARC, he was best known for his work on the Active Badge, a system for automatically locating people in a building, and the first context-aware computer system (PARCTab). Contact him at 2200 Mission College Blvd., Santa Clara, CA 95052; firstname.lastname@example.org.
Keith I. Farkas is a senior researcher at Hewlett-Packard Labs. His research interests span the areas of personal and enterprise computing and include such topics as software and hardware techniques for managing energy consumption, microprocessor architecture, and software-based techniques that enable the automated control and management of utility computing environments. He is currently working in the latter area, focusing on model infrastructure and increasing the utility of monitoring data. He received his PhD in electrical and computer engineering from the University of Toronto. He is a member of the IEEE and ACM. Contact him at HP Labs, 1501 Page Mill Rd. (MS 1177/3U), Palo Alto, CA 94304; email@example.com.
Chandra Narayanaswami manages the Wearable Computing group at the IBM T.J. Watson Research Center. (His complete biography appears in "From the Editor in Chief," on page 3.) Contact him at firstname.lastname@example.org.