Milliwatts with Mega Impact
Anantha Chandrakasan, director of MIT’s Microsystems Technology Laboratories, sees big potential for electronics with tiny power needs.
In the MTL At left, Vivienne Sze, SM ‘06, PhD ‘10, Yogesh Ramadass, SM ‘06, PhD ‘09, and Joyce Kwong, SM ‘06, PhD ‘10, discuss an energy-efficient microcontroller chip that operates at very low voltages. At right, Patrick Mercier, SM ‘08, and Denis Daly, SM ‘05, PhD ‘09, test the supply voltages for a chip used to control a moth’s flight.
When Anantha Chandrakasan got up to give his talk at the 1994 International Solid-State Circuits Conference, a crowd quickly filled the room and spilled out into the hallways. Dozens of people couldn’t get close enough to hear. So the organizers decided to do something that they’d never done before for a talk on new research and have never done since: they asked Chandrakasan, who was then a doctoral student at UC Berkeley, to give his presentation again, so that the throng of people who missed it the first time could hear what he had to say.
Chandrakasan’s talk was about his work designing a power-efficient chip–a radically new approach to circuit design that would spark a revolution in the field. Dennis Buss ‘63, SM ‘65, EE ‘66, PhD ‘68, now chief scientist at Texas Instruments, was so impressed that he sat through the talk twice. At the time, typical circuits used a hundred times more power than the designs Chandrakasan was presenting. “It shocked the industry,” Buss recalls.
For decades, electronic devices had been getting faster and smaller at a rapid rate. The faster they got, the more power they required. But the proliferation of portable electronic devices like laptop computers and cell phones, and the promise of a new generation of tiny medical and environmental monitoring devices, suggested that decreasing power consumption could offer enormous benefits. The tantalizing possibilities included greatly extended battery life and even, in some cases, getting rid of the battery altogether. No wonder people were lining up to hear about it.
The revolution that Chandrakasan helped launch in the early 1990s got chip makers thinking differently. Instead of focusing exclusively on processor clock speed, they began to see power efficiency as a goal. This shift in point of view made devices like netbooks and smart phones possible. Now Chandrakasan, a professor of electrical engineering and director of MIT’s Microsystems Technology Laboratories (MTL), is overseeing what he hopes will be a second revolution–one that will slash power consumption again. Electronics that don’t need traditional batteries at all may at last be on the horizon.
Slow but Mighty
Chandrakasan’s historic ISSCC talk was the culmination of work that began in the summer of 1991, at the instigation of his thesis advisor, Robert Brodersen, SM ‘68, EE ‘68, PhD ‘72. That spring, while listening to a discussion at another big conference about power consumption in portable devices like cell phones, Brodersen had an epiphany. “It clicked with me, and I said gosh, power is the most crucial thing to think about,” he says. So he, Chandrakasan, and another graduate student, Samuel Sheng (now chief technical officer at Telegent Systems), began meeting several times a week to kick around ideas for reducing the power needs of electronic circuits. “Those guys just went after this thing,” Brodersen recalls.
In addition to thinking about how to get circuits to use less power, they also considered the implications of achieving that goal. They focused on what would be needed to produce full multimedia computing and communications capability in a small, thin, portable pen-input device that could keep working for hours on batteries alone. Presciently, they named their project the “Infopad”–almost two decades before a very similar device with a very similar name was to set all-time records for sales of a new kind of electronic product.
When they started, it was far from clear that what they were attempting was possible, and there were plenty of skeptics. Few others were even trying, Brodersen explains: “The basic feeling in the industry at that time was that there was no power problem anymore.” After all, a new generation of chips based on complementary metal-oxide semiconductor (CMOS) technology had already produced a big improvement in power consumption. CMOS circuits didn’t consume power constantly the way earlier circuits had, because they used power only while performing calculations. “Most people thought CMOS solved the problem,” he says.
Not Chandrakasan. “I knew that with these wireless devices, energy was going to be a key issue,” he says.
As they brainstormed together that summer, Brodersen, Chandrakasan, and Sheng realized that the kind of efficiency they were looking for was going to take some kind of major change. One possibility they considered was a drastic reduction of chips’ operating voltage. But that had its own problems: trying to reduce the voltage degraded performance so much that the chips quickly became useless, says Brodersen. They needed something else.
Finally, they hit on the idea of parallelism, and Chandrakasan did the calculations and simulations that proved it would work. Conventional circuits could be made to operate at low voltage if their speed was also very low. Doing more things at once, they realized, could compensate for the lack of speed so that the same amount of work got done.
By that summer’s end, they had licked the problem–at least in principle. They published a paper in 1992 in the IEEE Journal of Solid-State Circuits, the leading journal in the field, outlining their vision for a power-efficient chip that would compensate for speed loss through parallel operations. The paper described methods for making computer chips and other integrated circuits that could operate on a one-volt power source, instead of the five volts that was then standard–something that Chandrakasan says people didn’t think was possible at the time. As the lead author, Chandrakasan summarized what was to become his doctoral thesis work. More than a decade later, that report about a student’s research project remained the second-most-cited paper in the journal’s history.
By the time Chandrakasan gave his talk at the ISSC conference in 1994, the vision had become a reality. Having laid the theoretical groundwork in the earlier paper, he demonstrated the production of a working six-chip set that could perform all the computing, audio, and video functions needed for his prototype Infopad. It consumed just five milliwatts–about a hundredth as much power as comparable circuits at the time.
Sitting in his office at MIT, Chandrakasan smiles as he recalls the moment when all the parts came together. Alone in his Berkeley lab in the wee hours of the morning, he finally got full-motion video to begin streaming to the monitor from his 1.1-volt circuit. “That was a great moment, to see the full system working,” he says. “It was very exciting.” But he didn’t feel he could wake his professor in the middle of the night; he waited until the morning to call Brodersen with the news.
Such low-voltage, low-power electronics have now become widespread, especially as new generations of smaller, more powerful electronic devices such as smart phones have proliferated. “The concepts that the industry considered to be radical, innovative, and visionary in 1994 are in common use today,” says Buss.
An Idea Takes Flight
After receiving his doctorate in 1994, Chandrakasan came straight to MIT, where he became director of the MTL in 2006. He immediately set to work to try to make electronics that could thrive on an even stingier power supply. These days, he and his students are working toward chips that run on 0.3 volts. MTL alumni Vivienne Sze, SM ‘06, PhD ‘10, and Daniel Finchelstein ‘05, PhD ‘09, have already developed an ultra-low-power high-definition video decoder chip that operates at 0.7 volts. The MTL researchers are trying to push power requirements so low that electronics could run, battery-free, on “waste” energy scavenged from tiny movements or body heat. And they are starting to work on ways of putting such chips to use–a challenge that means maximizing the efficiency of all the elements of a complex system and, simultaneously, of the ways they link together. “You have to look at the whole system, and ensure that every block is low-power,” says former MTL student Denis Daly, SM ‘05, PhD ‘09, who now works for Cambridge Analog Technologies, a local startup made up mostly of MIT alumni and faculty. “You’re only as low-power as your weakest link.”
When the MTL scientists began researching how to integrate such low-power components into complete systems, a moth helped point the way. In 2006, MIT researchers received a federal grant to develop a system capable of controlling the flight of a living moth or other insect as a small, potentially self-sustaining platform for gathering environmental information. “Moths have very sophisticated flight abilities,” explains Patrick Mercier, SM ‘08, a doctoral candidate in electrical engineering and computer science, one of the students in Chandrakasan’s lab who took part in the project. “Mechanical devices don’t even come close to how efficient they are.”
It quickly became apparent that the electronics needed for the task would have to meet daunting limits on size, weight, and energy consumption. So the project was divided into components: communications, power supply, and control systems. Over the years, more than a dozen students from several research groups have collaborated on the project, focusing on various aspects of the system and conferring with each other to make sure their parts will all fit together physically and electronically. MTL contributors Mercier and Daly focused on the low-power transmission and reception systems needed to send commands to the moth.
Together, the team managed to develop a package that weighed about one gram–less than half as much as a penny. It included the control circuits, battery, and radio receiver, all mounted on a miniature harness that could fit on the abdomen of a five-centimeter-long Manduca sexta (a hawkmoth) without interfering with its flight. Tiny wires were used to connect the circuit to the insect’s nervous system, creating what Mercier refers to as a “cyborg moth.” (Neuroscientists from the University of Arizona and the University of Washington worked with the team to develop the interfaces to the moth itself.)
The key to the ultra-low-power radio device was using ultra-wideband transmissions in very brief bursts–very different from the long-duration narrowband transmissions used for conventional radio systems such as Bluetooth connections, which require up to a hundred times more power. On average, the whole system used less than one milliwatt.
By 2009, the team had achieved the goal, producing a complete system in which the tiny, complex components did their jobs, working together to steer the moth’s flight and proving the potential for very small, ultra-low-power systems.
Daly envisions a day when even smaller, lower-powered systems based on this research could be deployed in vast swarms. “They could be like dust motes that you could distribute over a large forest to detect fire, for example,” he says.
Mercier was initially drawn to the moth project because he saw great potential for health-care applications: it laid the groundwork for a whole new generation of small, lightweight, and perhaps even implantable devices for medical monitoring, diagnosis, and treatment. Chandrakasan and his MTL colleagues envision self-contained “electronic band-aids,” as they call them, that will incorporate sensors, a battery, computer chips to analyze the sensor data, and a radio transmitter and receiver to communicate the data–all packaged into a device small enough to wear as a skin patch.
“We’re developing continuous ambulatory monitors,” says Mercier. Such devices could be used, for example, to observe heart activity 24/7 in patients with heart disease, allowing them to go about their daily activities with sophisticated machinery unobtrusively attached to their arm or chest. “We want someone to be able to wear it and not even know they’re wearing it,” he says. If the device incorporated GPS and cell-phone technology as well as sensors, it could identify the person’s exact location and automatically call for help in a medical emergency. Electronic band-aids might also be used to monitor brain waves in patients prone to seizures, perhaps detecting an imminent seizure in time to prevent it: the device could automatically trigger a pulse to an implanted electrode that would disrupt the pattern of brain activity.
The technology that’s needed to extract useful information from the data such sensors collect is what Joyce Kwong, SM ‘06, PhD ‘10, who now works at Texas Instruments, focused on during her years at the MTL. She built a chip that examines electroencephalograph (EEG) data for abnormalities in the signal. Its energy needs are minimal, in part because separate accelerator modules around the main processor offload some processing tasks to smaller dedicated circuits. “That translates into longer battery life,” she explains. “Instead of a couple of hours, it could run for a couple of days.”
In the short term, doctors might use this technology to monitor patients after they leave the hospital. But the low-power devices could eventually be used in rural areas and poor countries, where hospitals are few and the nearest doctor may be too far away to reach someone in a crisis. “These systems don’t require much maintenance by a doctor, and they’re intended for a patient to wear at home,” Kwong says. “They can include a radio on the band-aid, which sends information through a cell phone, and then it is relayed to the Internet.” Doctors who have the expertise to interpret the data could analyze it and make a diagnosis, no matter how far away the patient might be.
Kwong says the chip is really a “flexible processor” that could be programmed to analyze different types of physiological data. And because it’s small and based on standard manufacturing technology, it could be made for pennies, Mercier says: “If the volumes are large enough, the prices are dirt cheap.” Such chips could eventually become disposable diagnostic systems, nearly as inexpensive as regular band-aids.
And that could be just the beginning. Mercier, who has been working on such systems for the last five years, talks about the potential for several devices attached to different parts of the body to monitor a variety of health indicators all at once. The data would then be relayed in real time to one central processor for analysis. The result would be what he calls a “body area network.”
Getting Rid of Batteries
Some of the prototype chips being developed in the MTL are no bigger than a sesame seed–so tiny that someday they could be fully implanted in the body. Ultimately, researchers hope, they could run entirely on scavenged energy, never needing to be recharged.
Yogesh Ramadass, SM ‘06, PhD ‘10, is one of several MTL students who have worked on harnessing trickles of energy that normally go to waste. At the 2009 Energy Night at the MIT Museum, organized by the student-led MIT Energy Club, Ramadass showed off one of the products of his research. Among the dozens of booths and posters showcasing energy-related projects, he wasn’t hard to find: for most of the night, he was surrounded by a crowd of people staring at the bizarre apparatus he wore on his arm.
Designed to draw power from the small difference between the temperature of his skin and the temperature of the air around it, the prototype was hardly unobtrusive: near the crook of his arm was a shiny, spiky aluminum heat sink, a kind of radiator that dissipates heat to the surrounding air. The heat sink directed the heat of his body through a thermoelectric harvester connected to a tiny chip that he’d designed to make unusually efficient use of the thermoelectric effect exhibited by certain semiconductor materials: when one side is hotter than the other, the temperature differential generates an electrical voltage. The fascinated throng peppered him with questions, especially about the possible future uses of such a system. The most intriguing uses do remain in the future: the prototype produces too little electricity to operate a cell phone, though it generates enough to power a watch or a calculator.
Ramadass, who now works for Texas Instruments (which is one of the major sponsors of the group’s research, along with Intel and others), pursued two different approaches to energy scavenging. In addition to exploiting tiny differences in temperature, he has also developed experimental devices to harness energy from small motions and vibrations. These devices use piezoelectric materials, which produce an electric current in response to pressure. Ultimately, he says, small electronic devices might be hooked up to systems that can scavenge both kinds of energy, in order to maximize the available power.
There are still hurdles to be overcome. For one thing, the piezoelectric devices that harvest energy from movements and vibrations produce AC current, while solid-state circuits need DC. Because the amount of power available is so small, Chandrakasan explains, the circuits that convert the AC current to DC will need to be very efficient. Harnessing the thermoelectric power requires extra steps, too. Body heat can produce “about 50 millivolts” of output voltage, he says. That simply isn’t enough to run logic circuits. A voltage booster (transformer) can be used to increase the voltage, but again, the circuits that perform the conversion need to be efficient enough to wring useful power out of a tiny supply.
Thinking about the extent to which the Infopad that he imagined in 1994 prefigured the iPad that consumers can buy today, Chandrakasan wonders what kinds of devices we’ll all be looking at, using, and playing with a decade or so from now. “I’m thinking about what are going to be the challenges we’ll have in 2020,” he says. “There will be dramatically new functions, like computational photography–the ability to manipulate images in real time–and things like gesture inputs, and many different kinds of user interfaces.” Minimizing power consumption will be what makes such computationally intensive functions practical for many applications.
Chandrakasan is also thinking about smaller, cheaper, simpler devices that could make a difference in the many parts of the world that lack access to even basic modern conveniences. In a new collaboration, he and Subra Suresh, the former dean of MIT’s School of Engineering and the new director of the National Science Foundation, are aiming to develop a tiny, low-power device that could instantly diagnose malaria. Today, few rural clinics are able to conduct the laboratory test on blood samples that’s needed to deliver accurate results.
The idea is “to take a drop of blood, and do a microfluidic lab-on-a-chip so you don’t need a fancy microscope,” Chandrakasan says. “It would all be done electronically–you’d trap the cells and measure their electrical activity.” The whole apparatus, he says, would be “the size of a postage stamp.”
The realization of that possibility, of course, is still many years in the future. “How can you do diagnosis in a place far away from clinics, out in a village?” Chandrakasan wonders aloud. But he looks forward to watching this vision, too, become a reality. Sixteen years after his groundbreaking appearance at the ISSC conference, he is now serving as that conference’s chair. So whether important new advances in low-power circuitry come from his own lab or from somewhere else, he is ideally positioned to see them first.
Keep up with the latest in intelligent machines at EmTech Digital.
The Countdown has begun.
March 25-26, 2019
San Francisco, CA