Head: Low-Power Design Trends
Deck: Connectivity, rich graphics and multicore drive new designs
By Cheryl Coupé
Low-power trends are impacting embedded designs from the smallest handheld devices that need to squeeze out the longest possible battery life, to high-performance, always-on systems that need to run greener. “There are a number of different usage models for why people want low power,” explains Jonathan Luse, director of marketing for the Low Power Embedded Products Division at Intel. “Some people want low power because they have untethered, battery-oriented devices, and their motivation for wanting low power a lot of times is the extended battery life. There’s another aspect of low power as well, which is in many cases for tethered (or even untethered) devices, where the low power is all about performance-per-watt density.”
For many embedded applications, performance-per-watt levels have evolved to enable a wide range of new applications for computing in remote and unusual locations in which power delivery is a challenge. Examples include solar-powered mobile tower applications, or rail car vibration monitoring systems that convert the moving car’s mechanical vibrations to electricity that powers the system.
Chip manufacturers are responding to demands for low-power designs with a range of options, from the Intel Atom processor family to the AMD G-series platform. Both Intel and AMD are watching low-power design trends closely. We talked to Intel’s Luse and Buddy Broeker, director of AMD Embedded Solutions, to get a sense for the low-power trends they’re watching.
Connectivity megatrend behind low-power evolution
The move to mobile and handheld formats in everything from consumer electronics to industrial and medical devices is a driving factor for low-power embedded designs. But according to Luse, there are a couple of megatrends behind the scenes. “Things that used to be discrete or isolated compute devices are becoming connected,” he says. “If you look at some of the devices that were providing automation or intelligence, a lot of them were doing it without being aware of their surroundings. As they become connected, they’re becoming aware of things up- and down-stream from their devices. That megatrend fundamentally changes the role of a lot of these embedded devices.”
But that’s not the only trend in play. Developments in wireless connectivity extend that model even further. In the past, many of these industrial or retail devices were isolated and standalone. Today, they’re likely to be connected to a broader network. So a point-of-sale device is no longer just a cash-transaction terminal; it’s an inventory-control system, an advertising device, and a data-mining system. That phenomenon is occurring in a wide range of traditional embedded segments and is leading to an explosion of innovation, which Luse expects to increase as the range for wireless broadband connectivity expands.
AMD’s Broeker agrees, but sees some fragmentation that needs to be addressed. “I see a huge opportunity over the next few years in the world of connected devices. There’s some innovation that needs to happen for that, and it’s not that the wireless technology doesn’t exist. You have all of these devices that are different – they don’t know how to talk to each other. Right now it’s very fragmented. What needs to happen for that explosion of connected devices is there need to be standards; there need to be middleware layers written so devices know how to talk to each other. But no doubt the opportunity is huge.” Broeker quoted a recent statistic that connected media devices will hit something north of 350 million units by 2014.
Rich visual experience now available with lower power
Broeker believes that embedded developers have more choices today than ever before, stating, “It used to be if you wanted to do a low-power design you had no choice but to do something with ARM.” Today, adoption of the x86 architecture in embedded applications continues to accelerate, and the integration of graphics and low-power technology is a key part of that. “You’ll see more and more devices with a rich visual experience because of graphics and video-decoding performance that is orders of magnitude lower power than it was just a couple of years ago.” Broeker describes past requirements for DirectX 11 graphics that required a 125W graphics card. He compares that to current thin-client customers who are building fanless systems driving two or four high-definition displays with streaming video using a new AMD G-series platform that includes an AMD accelerated processing unit (APU), two 64-bit x86 cores, a DirectX 11 graphics engine, and universal video decode – all in a part that requires less than 10W.
Design challenges in move to multicore
Low-power designs provide additional challenges to developers, especially as they move to multicore platforms. “A lot of work needs to be done to help embedded customers move to multicore in an efficient way so they maintain the power envelopes that are tolerable for their application,” Luse states. There’s still a lot of single-threaded code in the embedded world, which has slowed adoption of multicore in embedded, but Luse says that’s about to accelerate as partners in the ecosystem develop tools to help parallelize code.
Broeker sees similar evolution in the move to multicore in embedded designs, saying, ““The low-power system designer finally has the opportunity to use multicore processing in a reasonable power envelope.” Broeker is excited about Open CL and sees that as a key trend moving forward, especially in heterogeneous multicore systems that require a single programming environment. “You’re going to see customers using Open CL to get even higher performance on applications that are prone to parallelization,” he states.
Scalability key to migrating designs
The evolution of multicore designs from desktop and server applications to mobile and low-power devices brings opportunities as well as added complexity. Luse explains that the most important thing developers can do is to make their software scalable so they can take advantage of the technology as it migrates. Software design done well becomes somewhat future-proof if it can be migrated from one generation of processor to the next. Luse says, “One of the best advantages Intel has is the ability to have that software investment protection from one generation to the next, as well as up and down the Intel architecture stack. That allows customers to develop – and preserve the investment in development – from one generation to the next.” And that allows developers to spend more time on application-specific enhancements.
AMD lays claim to similar scalability advantage, especially with respect to rich imaging. Broeker states, “This is where providing platforms that are very scalable is important. You may see medical imaging customers building platforms like an MRI on a cart as well as a handheld device. To the extent that we can provide upwards and downwards scalability in terms of performance and power, it’ll just make developing those platforms that much easier for our customers.” Broeker notes that customers’ smaller engineering staffs also drive the importance of scalability. “Hardware no longer pegs product development,” he says. “The long pole in the tent now is software.” AMD offers a range of reference designs to customers to help them get to market sooner.
Additional innovation within the ecosystem will also help drive opportunities for new applications. From a platform perspective, complementary silicon and board-level devices – including voltage regulators, power-management ICs and complementary chips supporting new segment-specific I/O – have to progress. The other area that is ripe for innovation in the embedded market specifically is the software stack. Luse predicts that as embedded computing devices become more commonplace, there will be pockets of segment-specific or application-specific software stacks that will match medical, retail or industrial protocols, as well as ongoing development in real-time and specialty operating systems.
Roadmap for low-power platforms
While the performance that new processors can deliver at 5W or 10W is impressive, embedded developers are already thinking in terms of what they could accomplish with similar performance at 1W or even 1/2W. A common refrain from developers is that if there is one thing they could still use, it’s lower power – even though the term ‘low power’ is relative to the needs of specific applications across the embedded marketplace.
Broeker describes a roadmap for AMD platforms that drives power, area and cost. “That’s the formula we’ve used,” he says, “and you’ll continue to see that going forward.” The approach has been successful, as AMD states that its 64-bit business has grown 350 percent.
Luse compares the Atom processor to a Pentium 2 Xeon, which consumed about 40 watts of power nearly ten years ago. The current generation of Atom processor consumes 3 to 5 watts of power with about twice the compute performance. Luse expects to see a similar performance-per-watt improvement over the planning horizon of about the next ten years, as long as Moore’s Law applies. “There’s a very big strategic thrust for Intel to continue to move down that journey of lowering power and improving performance simultaneously,” says Luse.
Looking forward, Broeker says, “I think you’ll see convergence and a blurring of the line between personal computing and tablets and handheld devices. As we push x86 down and ARM tries to push up, you’re seeing the battleground defined.”
“I look at it as more of a journey than a destination,” says Luse. “I don’t look at it as ever being a situation where you’ve arrived. No matter what power level I’m at, I can always say, ‘I wish I could get 30% less power’ and find applications for that.”
Pull quote: You’ll see more and more devices with a rich visual experience because of graphics and video decoding performance that is orders of magnitude lower power than it was just a couple of years ago.
Pull quote: A lot of work needs to be done to help embedded customers move to multicore in an efficient way so they maintain the power envelopes that are tolerable for their application.