15 June 2010—This week, at the Symposium on VLSI Circuits, in Honolulu, Intel will reveal the engineering behind its newest Atom system-on-a-chip (SOC) platform, which the company unveiled six weeks ago in a bid for the smartphone and tablet markets.
Pankaj Kedia, who directs Intel’s Ultra Mobility Group, says that the chip is the company’s response to an ongoing shift in the phone market. ”Smartphones are becoming computing devices that happen to make a phone call,” he says. At this point, smartphones are a minority of the cellphone market: In 2009, according to Forrester Research analyst Michele Pelino, just 17 percent of U.S. adults had a smartphone (including Android, BlackBerry, iPhone, Palm, and Windows Mobile devices). Within three to five years, Kedia says, Intel expects to see that number increase substantially.
One weakness of today’s smartphones, according to Kedia, is that their ARM-based processors require designers to choose one device, develop an application for it, and then rewrite that application for other smartphones. An application developed for the BlackBerry simply won’t run on an iPhone, even though both have ARM processors. With the x86-based Atom Z600 chip (code-named ”Moorestown”), that will no longer be the case, he says.
”When we talk to developers, their eyes light up,” Kedia says. ”Today in the mobile space, they have to pick a device and write to it.” He says Intel’s move will make designing for smartphones more like designing for PCs. ”You don’t pick an HP-core PC,” he says. ”You just pick a PC, write an application, test it on one PC, and it runs on any PC anywhere in the world.”
Most important, an application designed for the Atom Z600 series would run not only on any Atom-based smartphone but on any Atom-based device. A single application could be compatible with smartphones, tablets, netbooks, embedded systems, and more (provided, of course, that the operating system is the same).
But to make this dream a reality, Intel had to drastically bring down chip-power requirements while maintaining performance. A smartphone chip must be on a strict diet when it comes to power consumption.
To that end, the smartphone version of Atom can go to 1.5 gigahertz, but the SOC core itself sips power: In a low-power state, it consumes as little as 100 microwatts. In usage terms, that means a typical smartphone could last 10 days in standby mode on a single battery charge, play up to 48 hours of continuous music, or allow 4 to 5 hours of HD video or Web browsing, according to Intel.
This is a significant improvement over the previous Atom netbook generation, which combined the 45-nanometer high-k metal gate x86 core with a 130-nm-generation chip set. ”It sucked power like nobody’s business,” says David Kanter, an analyst with Real World Technologies in Colton, Calif. But the power and performance needs of a netbook differ greatly from those of a handheld device.
To get that first-generation chip into a mobile device, Intel had to reduce its power consumption in both active and idle modes.
Some of the savings come from integrating what were functions of the power-hungry 130-nm-generation chip set from the previous Atom generation into the 45-nm Atom CPU core. These functions include memory controllers and display controllers with 3-D graphics, as well as video processing, which will allow users to download and display 1080p HD content—unprecedented on a smartphone—and also upload HD content (say, a mobile video to YouTube in 720p). The platform controller hub (where the input/output blocks are) is on a second chip, built at Taiwan Semiconductor Manufacturing Co. using normal, silicon-dioxide-based 65-nm transistors rather than high-k/metal gates.
Another advantage comes from relying on a wider variety of transistor types than were used in the previous generation. Engineers weighed trade-offs between transistors that could provide high performance and those that could have low current leakage, and therefore waste less power. ”What we did is pick the best transistor type that’s suited for a given functionality to support higher performance and low power” to that particular part of the chip, says Rajesh Patel, a lead architect on Moorestown who will present at VLSI later this week. Using that strategy, Intel engineers reduced the power lost to current leakage by more than 65 percent while maintaining the same performance as that of the previous Atom generation.
Another important innovation was a new, fine-grained power-management scheme. The non-CPU portion of the chip set (for example, the codecs) employs distributed power gating, which allows particular systems to shut down when they’re not needed. And as for the CPU core itself, Patel says, ”We looked at every little logic in there that burns power and said, ’Do we need it? Why is it still on? Can we turn it off?’ ”
Yet another crucial change was the development of an operating-system-level power-management scheme. In the previous generation of Atom, Patel says, power-management techniques were applied at the chip level. However, on Moorestown, those same techniques are applied at the system level by the operating system: MeeGo, Android, or Moblin. ”We shut down the pieces of the device that aren’t being used. When you’re playing music, you don’t need the display, you don’t need wireless, you don’t need much I/O or the power-management chip. So we shut it down. When you’re in standby mode, you don’t need much of your system; you just want your radio to be pulsing for incoming calls.”
Despite all of these promising features, some experts believe that Moorestown is not Intel’s true endgame for the smartphone market. The company’s real focus is on Medfield, Moorestown’s 32-nm successor, which will be introduced in 2011.
Medfield will integrate the I/O hub in addition to everything it has thrown into the Moorestown. ”I suspect only a few smartphones will use Moorestown,” says Kanter. Intel’s 32-nm SOC process, he says, is much more appropriate for a mobile phone chip than Moorestown’s 45-nm SOC process. ”With Moorestown, Intel is just getting its feet wet.”