First of two parts: Devices will need to be designed from the ground up, and so far there isn’t enough information to determine what’s needed.
It was Dick Tracy’s wristwatch communicator that triggered the public’s appetite for wearable electronics. Introduced in a 1946 syndicated comic strip, the idea was so compelling that it inspired the release of hundreds of wrist-based devices ranging from walkie-talkies to calculators to GPS trackers, heartbeat and movement monitors. Yet despite the public’s fascination with this kind of technology over the 69 years since then, nothing has stuck around very long.
That doesn’t mean anyone is giving up on this concept. In fact, many more devices are on the way. But the conversation behind them is shifting. Debate continues behind closed doors about whether a single killer application really does exist. The leading candidates are health-care related applications—being able to track enzyme changes or heartbeat changes that signal a heart attack ahead of time, for example—along with mobile payments and security. Still, the current thinking is that success in this market will be built on a combination of applications, probably with a back-end business infrastructure that subsidizes the cost of end devices.
Getting there will require some significant technological changes, though. A large, clunky device that has to be unstrapped every night, or even every few nights, and plugged into a cable won’t foster widespread appeal. What’s significant here is that the smart watch is becoming the first mass-market mobile device for which power is the primary concern, rather than just one of several factors along with performance and area.
But lengthening time between charges and improving the form factor isn’t so simple—or so straightforward. It requires a bottom-up rethinking of chip architectures, as well as what kind of processor and memory technology will be used, how various components are connected, whether communication is on-chip, off-chip—and how far off-chip— as well as how software will be written and utilized, and the security scheme used to protect data. Each of those factors can have a big impact on power, and together they can make or kill a market.
It also requires a top-down rethinking of what functions are critical, how often they need to be accessed and how quickly, and what functions can be turned off or left out entirely. Being able to identify a pending health problem is critical. Being able to direct you to the dairy aisle in a supermarket when you’re looking for milk is not, and it may take more energy to provide directions than to steadily monitor your vital functions.
“The wearables device market requires system-wide thinking,” said Scott McGregor, president and CEO of Broadcom. “It includes everything from power on the infrastructure side to making the device work with batteries in the proper form factor. It’s total system design, from process geometry to only having certain parts ‘on’ at any time, to understanding latency and I/O. It’s also about expectations, which may be limited as we get going. But when we get enough functionality, this market will take off. It’s like a wave rolling in that won’t stop. There will be a knee in the curve in the next couple of years.”
The whole industry is preparing for that spike, from big-name device companies to chipmakers to infrastructure providers, venture capital firms, and a long list of smaller players vying for a piece of the market. The good news is that with so many different companies are converging on this market means there is enough at stake to build the infrastructure, the technology and to develop the necessary standards to ensure at least some success. The bad news is that there is no clear direction for how that should be done.
“The IoT market is a serious market opportunity,” said Mark Bohr, senior fellow and director of process architecture and integration at Intel, noting that wearables will be a very viable subset. “It’s all about ultra low power, ultra low leakage, and probably less about cost than people have been saying. Some of the older technology nodes will still scale, but even if you’re not shrinking features you need to tune and adjust those processes.”
The starting point for sufficient reduction in power is the architecture. One of the reasons current devices have fared rather poorly—at least in comparison to the smartphone or tablet, for example—is that early entrants rushed products to market based upon existing components and approaches.
“We tend to think about the system architecture and then the gate and implementation level and all the techniques that go along with that,” said Mark Milligan, vice president of marketing at Calypto. “The wearable market, in particular, has to include all of the above. You need to have a plan for everything from system architecture to micro architecture to gate at the same time. And we’ve seen some really novel circuit architecture being created, along with system partitioning for what you need to keep the power down and selection of IP like a lower power processor than what you might normally use. Power is first, and all techniques are being deployed.”
Milligan calls this “the ultimate get-it-right and hit-it-right” challenge using a minimum number of components. “Power is number one. But a close second is having the right set of functionality, and at this point no one knows what that is. It’s not like there is ‘the chip set’ for wearables. We still need to decide what is needed and what we can live without.”
The challenge isn’t just about the individual technology components, though. It also involves how those pieces are put together, and perhaps even more important, why they’re being used in the first place—with power as a starting point. That can be significantly more complex than making tradeoffs for performance or cost because it involves everything from where and how data should be stored to how to get it there in the first place.
“You really never want to go off chip for anything,” said Steven Woo, vice president and distinguished inventor at Rambus. “Bluetooth Low Energy will solve some of the issues for moving data off the chip, but the bigger problem is that it is very difficult to maintain a battery capacity that can run one month or one year.”
Also included in that discussion is process technology and packaging. While the current thinking is that many IoT devices will utilize older process technology, wearables—particularly the ones people strap on their wrist—are likely to be an exception. In some cases, that could mean moving to the most advanced process nodes, where the latest power-saving techniques are used. But numerous engineers interviewed for this story say it also could be a driving force for stacking die, where wider signal channels and shorter distances can sharply improve energy efficiency, as well as providing flexibility for future changes in an uncertain market.
New technologies, new approaches
This is a very different focus for most design teams. Ever since the introduction of the microprocessor—and even well before that—the real emphasis has been on speed. Power has been about the ability, and the cost, of increasing speed, and for many devices it is still the primary concern. A smart phone that can search the Internet and download data faster for the same amount of energy than a different model is considered the better device. But in the wearable market, that’s not the case. Sensors may be able to pick up data over time, but the more attractive solution will likely be one that imports and analyzes data in the same or greater amount of time using less energy.
“The key is total power in the signal chain,” said Larry Przywara, product line group director for audio/voice IP at Cadence. “We’ve put a lot of emphasis on the processor and decreasing power and the core. We can have less than 17 microwatts for a processor using 20nm process technology. But that’s just one facet. For components like a microphone, the processor accounts for 10% or less of the total energy.”
There is no clear answer about how to go about saving power across an entire device, though. While everyone agrees this is a system-level issue, the market for the current round of wearables is still in the experimental stage. There is no clear direction for connecting things or for what the best power management strategies are. Moreover, the supply chain is so fragmented that for many companies their only point of influence is at the block level.
“You have to look at every part,” said Aveek Sarkar, vice president of product engineering and support at ANSYS. “Logic needs to be scrubbed as much as possible, of course. You have use-case scenarios for those. But with wearables it’s easier to determine all the known operating modes and see where you’re wasting power. So even though the power requirement is much more stringent, it’s easier to solve. The number of operating scenarios is not that wide-ranging. Wearables are almost like a smart phone on the hand, but you don’t have to run as many applications. And you will most likely leverage the smart phone more as the I/O interface.”
But the market isn’t static when it comes to use cases, either. One big knob to turn in this sector involves memory, and what is garnering particular interest in this space lately is Resistive RAM.
“Embedded RRAM is a non-volatile technology and it’s relatively low power to store bits,” said Rambus’ Woo. “With RRAM there are two terminals on bit cells and they basically form a filament between terminals. When voltage is applied, the resistance between the terminals drops.”
This makes it almost ideal for wearables because it isn’t subject to the refresh and burst modes of DRAM. “We have big hopes for RRAM because it’s 1/8th the size of SRAM, it uses less power and it’s ideal for big caches,” said Jamil Kawa, group director of the Solutions Group at Synopsys. “The problem until now has been variability in the manufacturing process, which is what makes it messy to use, but there are reports that companies are narrowing that variability. It’s also idea for medical and biomedical uses, where low-frequency response time is acceptable. It’s not as fast as SRAM, but it’s decent, and it’s CMOS-compatible.”
Another big knob to turn for efficiency is software. It’s no secret that badly written software can waste energy even in the most efficient designs. Far less attention has been paid to designs that sidestep power-management features built into the hardware because of the increased cost of doing so.
That’s beginning to change, particularly in the wearables market where power is the top concern. As a result, compute strategies are being built around that with software as an essential part of the design—something that has been talked about for years but rarely implemented. In some cases that involves heterogeneous multi-core designs with more granular use models based upon energy.
“We’re seeing more multi-core, multi-OS, inter-process communication,” said Andrew Caples, senior product manager for the Nucleus product line at Mentor Graphics. “At a minimum you want to shut down blocks. But you also want to shut down processors when you can. Silicon for years had low power features that were hard to take advantage of. You could write code for LP states. It may seem trivial to turn it on and off, but you also can shift the clock so you have high, middle and low frequency. You can take devices off line and bring them back. But you can’t just superimpose code to make all of this happen. If you don’t start with that mindset you get a very different result and you get code bloat. In wearables, the power management framework is a compelling feature. In fact, it’s one of the more critical features.”
Coming in part two: Designing for a rapidly changing market, sidestepping the battery, and business models that can help pay for more expensive designs.