Shifting process nodes and new applications are driving demand, but utilizing this technology requires a different design mindset.
Energy harvesting is gaining traction with a surge in ultra-low-power IoT applications, ranging from inventory tracking, wearables and drones, to vibration sensors for motors in industrial settings.
The idea that machines could run without batteries—or that energy could be harvested either from motion or ambient sound waves or chemical reactions to augment battery power—has been in the works for some time. In fact, self-winding wristwatches date back to just after World War I. But all of this is coming to life as technology improves, and as the need to power trillions of sensors and devices becomes a requirement for growth of the IoT.
At least part of the success can be attributed to process node advances, which in this case means from one older node to the next. At older nodes, moving from one to the next still provides enough power/performance improvements to make it possible to utilize energy that is scavenged from the environment.
“If you look at 90nm, the minimum operating voltage for most of the IP was about 1.6 volts,” said Ron Lowman, strategic marketing manager for IoT at Synopsys. “As engineering teams migrate down to 40 and 55nm with microcontrollers, you’ll see more energy harvesting. For example, most remote controls have two AA batteries because they need 1.8 volts to operate the IC. As you move to 40 and 55, it’s 0.9 volts, so you can do a single AAA battery. And that’s just the end of life for an alkaline battery.”
Another key metric here is 0.5 volts, which is typical for a single solar cell. As process technologies are able to reach that number, and IP is able to operate at that voltage, adoption should rise. But the formula isn’t completely straightforward. When dealing with legacy technologies, there also must be internal DC-to-DC down converters.
“There is efficiency loss when you do that, and costs associated with that,” Lowman said. “There are always tradeoffs between costs and additional IP on-chip to support that. But as technology progresses, we’ll continue to see more innovation and more use of those things where they are using small amounts of energy from other sources, be it solar, thermal, vibrations, or something else.”
Vic Kulkarni, senior vice president and general manager at Ansys, points to 0.7 volts as a key number. He said the most recent ITRS 2.0 Roadmap (v2015) indicates that energy harvesting will start to become prevalent for Internet of Everything (IoE) applications starting 2019, when Vdd supply begins to drop below that. This approach is ideal to supplement a primary power source, if available, as well as enhance the reliability of the overall system, he noted.
“Many real-life applications using energy harvesting system power are now becoming practical, especially in the IoT edge node networks,” Kulkarni said. “For example, wireless sensor nodes (WSN) network systems such as ZigBee ideally need energy harvesting power sources. This is going to be especially true, in what the MEMS Industry Group defines as the trillion-sensor world, when everything in the world gets connected and wireless nodes in remote sites must depend on energy harvesting as their energy source. The harvested energy can power small sensors and MEMS.”
Kulkarni believes IoT devices targeted for fitness and healthcare will be the first generation of applications that will start using thermo-electric energy harvesting based on the differential between human body and ambient temperature. That approach typically can generate about 10 to 20 microwatts per square centimeter of body surface.
But putting energy harvesting to work requires a different way of looking at a design, as well.
“You have to build systems that are essentially normally off,” said Drew Wingard, CTO of Sonics. “It’s the absolute opposite side of the coin from people who think about putting the functions in first, then going back later and thinking about what they can afford to shut down. It’s completely the other way around. It’s more like, ‘What’s the minimum I have to have on?’ And then you do the analysis saying, ‘Is that already too much, given how much energy I can productively harvest?’”
It also requires an understanding of how much energy can be stored when it’s not being harvested, because most systems do not continuously harvest energy. These systems require motion, light, or something to harvest to have that backup energy storage, which is generally measured in one or more duty cycles. Take a solar-powered light, for example. The user wants it to output the maximum light possible in the several hours right after dusk, so the battery runs out in the middle of the night and nobody cares because the use models support that.
“If I was doing solar energy harvesting for something that really only had to interact with its environment once a day, then maybe I wouldn’t need to do that,” Wingard said. “But the vast majority of energy harvesting applications have a duty cycle that is much shorter than that, and it’s not acceptable for them to only take one measurement a day, or to communicate upstream once a day. In those situations, you end up having the amount of energy you use becoming a first-class issue. You start with a blank sheet of paper and determine the fewest number of transistors that can be put into the part of the design that always has to be on. ‘What are they for? Are you sure they have to be there?’ You build up from there toward the functionality. This is exactly the environment in which you say, ‘I don’t care about power at all. I care about energy.’ If turning the whole thing on for a very short period of time so you can get the job done as quickly as possible, and turning it all back off is the lowest-energy approach, then that’s what you do. That’s a very different mindset.”
Wingard noted that the wristwatch industry did an enormous amount of work back in the 1980s using sub-threshold concepts for micro-power circuits. “All of that is real and will be very valuable in the energy harvesting circuitry itself, but also in how you think about integrating into existing communication systems that have already pre-defined duty cycles. And that’s hard.”
Micro-power circuit techniques will likely be utilized in many other devices, but the circuits will have to be much more energy-optimized. “You’re going to have to do the power state control autonomously, Wingard said. “Essentially, at the end of every substantial operation, you’re going to want to shut back down, which probably means you can’t use a microprocessor to do it. If you can’t use a microcontroller, you’re going to have to do it in hardware, so you start to think about the basic building blocks you build your system from as being things that inherently know when it safe to shut themselves down—and which have the hardware signaling to do that.”
One way to put those pieces together into the overall management system is to use an energy processing unit that brings together both active and static power savings techniques, coupled with a methodology that can be scaled and repeated using successive refinement.
While the benefits of energy harvesting are intriguing, Ansys’ Kulkarni points to some engineering challenges. Among them:
• Power generation source. This requires an ultra-low voltage DC/DC step-up converter and operates on input voltage as low as 20mV with a 2.2V V-LDO to power an MCU.
• Energy storage, size and cost.
• Increased design complexity.
There are several techniques, such as DVFS, which can yield considerable energy savings, but designers of WSNs must understand integration of IP components, workloads, intelligent partitioning, and so on. In addition, the substrate noise injection from an aggressor MCU into the RF section must be analyzed carefully because it can cause significant interference by degrading the signal-to-noise ratio. Proper modeling of near-threshold device operation is critical in this case.
Many of these challenges fall outside of the digital domain, too. “In solar panel yard lights, for example, fundamentally these systems are about harvesting energy over a long period of time, then deploying it when and how it is needed at the end,” said Jeff Miller, product marketing manager at Mentor Graphics. “They all have the same fundamental components. There is a harvester, some electronic component that takes that harvested energy, which is going to be varying voltages and generally small amounts of power available. And then it puts it into some kind of storage device like a battery or a supercapacitor or something like that, and uses it up in a much shorter burst.”
Sensing applications are seen as the next big driver of energy harvesting. That includes wearables, which can take advantage of movement, or things that go into mechanical equipment that can count on a certain level of vibration, such as being placed in truck tires or construction equipment, or even in HVAC systems near a fan. Many times these systems essentially have to put themselves in a deep sleep until enough energy has been harvested so they can do one thing, whether that’s take one measurement point or send a packet of data back up to the gateway to ultimately send it up to the cloud, Miller said.
“There is this element of this blast of activity, where you are burning off all that power that you collected, then this long wait while you build it back up again,” he said. “That ratio depends on the system, and in the case of a yard light, you’ve got eight hours of build-up, and three hours of discharge. In a vibration harvester it might be a minute of buildup and a microsecond of discharge. It’s all over the map, depending on the system.”
To enable energy harvesting to the fullest extent, there are a number of technologies that will speed things along, including near-threshold voltage approaches, which begin with the memory and logic libraries and the ability to operate at extremely low voltages, Lowman noted. “Voltage is one of the critical pieces of reducing the overall energy because it’s a squared function in the energy calculation, so it’s one of the most important pieces.”
Close work with foundries allows the technology to be ported and enables the foundries, but it also ensures that interface and analog IP can support the proper voltage levels that those processes support, he said.
When design engineers are determining they have the least number of cycles, leveraging the available tools is a first step. After that it is imperative to understand what the application needs to do when utilizing the proper low power modes, the proper instruction sets, maybe even developing custom instruction sets for hardware acceleration.
“If you look at a lot of universities today, one of the main focuses is on a lot of hardware acceleration to accomplish these types of functions, and it’s strictly for energy efficiency, strictly for better performance,” Lowman said. “We are seeing a lot of heterogeneous processing within different designs, so there will be multiple cores doing separate tasks that can optimize for power and get a specific task completed, rather than having a lot of software overhead to do things inefficiently. Where you’ve had existing architectures out there, people need to add new functions, and they end up just increasing the frequency of their processor and adding those functions. But if you designed it more intelligently, you can actually reduce the frequency and add hardware acceleration to accomplish common tasks. And, of course, doing that at very low voltages with logic libraries and memory compilers support — combining those are very effective at supporting energy harvesting applications.”
Energy Harvesting Update
One-on-one with Georgia Tech’s Manos Tentzeris on what’s possible today and what will unfold over the next five years.
Energy-Harvesting Floor (Power/Performance Bits: Oct. 25)
Engineers at the University of Wisconsin-Madison developed a flooring material which can be used as a triboelectric nanogenerator to convert footsteps into electricity.
New Challenges For Wearables
Second of two parts: Energy harvesting and improved battery technology are no substitute for best practices in electrical and software engineering.