There is no longer any doubt. Moore’s law, the rule of thumb used to estimate the growth of computing power for the last 50 years, is slowing down. The number of transistors on an integrated circuit has, as Intel’s cofounder Gordon Moore predicted, roughly doubled every 24 months, but last year, Intel’s CEO admitted that the current pace is now closer to two and a half years and decreasing
As Steven Bowers, an assistant professor in the Charles L. Brown Department of Electrical and Computer Engineering, points out, the issue is that decades of innovation have pushed the complementary metal-oxide semiconductor technology, the mainstay of computing, close to its physical limits. One answer, Bowers believes, is an approach called heterogeneous integration. “Heterogeneous computing means picking technologies that are optimal for a specific purpose and tightly linking them together,” he says. The key to successful integration lies at the interface between the technologies, Bowers’ area of expertise.
Multi-functional integration is the strategy that Bowers and colleagues in the department and at the University of Illinois at Urbana-Champaign are using to address a fundamental obstacle to the realization of systems like the Internet of Things — power usage. The thousands of tiny sensor nodes distributed throughout the environment or embedded in physical objects must use power extremely sparingly if the system is to function for any length of time. Bowers’ team is focusing on minimizing the energy drawn by the always-on wakeup radio receiver in each node. The receiver activates the node upon receiving a specific signal.
An Ambitious Goal
The Defense Advanced Research Projects Agency (DARPA) recently awarded Bowers’ team, which includes professors Scott Barker and Benton Calhoun at UVA and Assistant Professor Songbin Gong at Illinois, a $3.4 million grant to design a near-zero, ultra-low-power radio.
“DARPA seeks to extend unattended sensor lifetime from weeks to years,” Bowers says. “Their goal is to reduce the energy drawdown during the sensor’s asleep-yet-aware phase to less than 10 nanowatts.” This is an extremely ambitious goal, and combined with a required sensitivity of detecting a 1 nanowatt signal, requires improvement of a factor of a thousand compared with state-of-the art sensors. DARPA chose this target because it is roughly equivalent to the self-discharge during storage of a typical watch battery. In other words, the radio that Bowers’ team hopes to create would use the same amount of energy that the battery is losing anyway.
The Best of All Worlds
To achieve this objective, Bowers and his colleagues are combining elements from a variety of different technologies to optimize each part of the radio. One of the consequences of reducing the radio’s power is that it loses sensitivity. It could miss the signal to activate the rest of the sensor or, conversely, respond to background electromagnetic noise.
To boost sensitivity without consuming power, Barker and Gong are building a passive filter and voltage amplifier that uses microelectromechanical systems, a mechanical device on a chip made from lithium niobate. Their section of the radio turns the electrical signal into vibrations, which allow for high quality noise filters.
For Bowers’ and Calhoun’s purposes, silicon-based complementary metal-oxide semiconductor devices do a better job. Bowers converts the analog signal from Barker’s filter into a DC signal that a low-power digital circuit can detect. Finally, Calhoun is creating a device that further clarifies the incoming signal and determines if it is the proper one to activate the device.
“It really helps when making decisions about sensitivity and power consumption not to feel constrained by the limitations of any one technology,” Bowers says. “It opens new doors for us technically, and in the process sets the stage for new uses for networking — for instance for wireless health — that otherwise would be impossible to achieve.”