MOUNTAIN VIEW, Calif.--Much more than most companies, Intel's success depends on the technology that will arrive in its field years hence. As a result, the company has more than 1,000 researchers beavering away to gauge and develop that technology.
And the company wants everyone to know it.
At its Intel Research Day at the Computer History Museum here Thursday, the company touted a wide range of projects that extend beyond the company's core business of making computer processors. On display were projects to improve the WiMax regional wireless network technology, improve mobile devices' processing power while reducing their energy consumption, refine software to make larger-scale data storage faster, and transmit electrical power wirelessly within a modestly size room.
Intel also gave the work a higher-profile name Thursday, with Chief Technology Officer announcing that the Corporate Technology Group now is called Intel Labs. The group's role is to evaluate not just what works, but to find out what doesn't before Intel invests a lot of money in that area, Rattner said.
Power-efficient Atom systems
Intel rules the roost for PC processors, but it's an also-ran when it comes to cell phones and other mobile devices, in part because its x86 processors consume more power than rivals, including those of the ARM lineage. Intel's Atom chips are the company's current attempt to crack the market, and the next-generation "Moorestown" processor boasts lower energy consumption requirements.
"I've been doing this about 15 years now. We've had advancements, but never the magic doubling of battery life," said Paul Diefenbaugh, principal engineer
At the research day, Intel showed off technology that lets a Moorestown system use less power by using a more aggressive version of existing power-saving idea, sending a computer into somnolent states as deeply and frequently as possible. "We realized the problem was really about the platform," Diefenbaugh said, because saving small amounts of power in the processor was futile when something like a USB controller chip was consuming more power and keeping the system from entering a low-power idle mode.
Platform-level engineering is easier with Moorestown, which combines many computer system elements onto a single processor, integrating graphics, a memory controller, and more in a technology generally called system-on-a-chip. That means it's relatively easy for one part of a chip to signal when it's idle and doesn't need power and when it's about to get busy and need more power, Diefenbaugh said.
Intel showed a running Moorestown system that cut power consumption by 50 percent to 90 percent compared with the current "Menlow" model by using research versions of this power-saving technology. Rattner said that production versions would see power savings of "up to" a factor of 50 with Moorestown compared to Menlow.
Although Intel showed a wide range of technologies, some are closer to the company's core business than others. Rattner and Mike Mayberry, vice president of Intel's technology and manufacturing group, described one: silicon photonics, in which light rather than electricity transmits data from one chip to another.
Today photons carry data across long distances with fiber optics, but Intel is among those who believe it will eventually travel directly from one chip to another, with transceivers built into the silicon chips to send and receive light pulses.
"We're hard at work to demonstrate a complete silicon photonics transceiver this year," Rattner said. "We won't tell you exactly our bandwidth goals, but they're very impressive."
In the nearer term, light will be used to transmit data among servers in a data center and then within a computer chassis, Mayberry said, but photonics embedded completely in silicon should arrive afterward. "We're talking about potentially the middle of the next decade," Mayberry said.
Mayberry also said Intel is working on bringing new technology for creating silicon chip circuit patterns from research to manufacturing stage. That next technology uses extreme ultraviolet light, which has a shorter wavelength and therefore can be used to help etch smaller features to help keep up with Moore's Law predictions for ever-more processing electronics in a given amount of chip area.
And Intel wants a place in next-generation memory technology, too. On the agenda today are "floating body" cells, phase-change memory, and seek-and-scan probes, each of which hold promise but have drawbacks, he said.
Intel manufactures and promotes solid-state disks (SSDs), which replace spinning platters of conventional hard drives with packages of unmoving, fast-responding flash memory. The biggest hurdle with SSDs today is their higher cost.
Intel is working on benefiting more from SSDs without going whole hog, though. The company's approach goes beyond the idea of using an SSD as a high-speed cache for a storage system that relies more on conventional hard drives.
Instead, Intel has created a variation of the ext3 file system Linux uses to store data. The Intel version checks the hard drive command requests and prioritizes the ones it judges to be high-priority data so the single SSD in a 12-drive storage system handles that data, said Matthew Eszenyi, a technology strategist.
Adding the SSD cache doubles the overall system speed, he said, and using the prioritized data system doubles it again, Eszenyi said.
Wireless power transmission
Electric toothbrushes and other devices can be charged without wired connections, but Intel has been working on technology that works over much longer distances. At the research event, the company showed off a new variation of the idea that transmits power through the air to run a speaker without any other power source.
Two flat copper coils are used in the technology, each tuned to resonate at a particular frequency. That means when electromagnetic energy is released from one, the other picks it up in much the same way an opera singer can shatter a wine glass by singing at just the right pitch, said researcher Emily Cooper.
Ultimately, Intel sees the idea as useful for delivering power to a laptop computer inside a room, but it could be used over shorter ranges, too--for example to replace the fallible wires that connect laptop screens through a hinge, Cooper said.
The wireless transmission shows efficiency of 90 percent at distances of up to a meter, she said, and Intel has shown it powering a 60-watt light bulb, too.
Multicore data dealings
Intel's tera-scale processing project--which Rattner said is expanding by a factor of 1,000 to become the exa-scale project--is designed to tackle the challenges of serious multicore processing. Today's chips typically have eight or fewer processing engines called cores, and communications among them are relatively straightforward along a bus--a linear data pathway that links the cores together.
But with more cores, things get more complicated. Aniruddha Vaidya showed a mesh of 36 cores--a 6-by-6 grid made of programmable chips rather than an actual single slice of silicon as eventually will be the case.
The cores on the periphery can connect to resources such as memory or graphics, but the cores in the interior connect only to other cores. To transfer data, each core must often transmit data from one to another in multiple hops.
In the 36-core mesh, data takes an average of 4 hops to get where it needs to go, Vaidya said.
Part of the reason for the research is to develop necessary higher-level features. The mesh can be partitioned into multiple independent patches to support virtualization or security needs, he said, and the data-routing technology can adjust when individual nodes fail.
Boosting WiMax capacity
Intel has long touted WiMax technology for bathing an area in broadband wireless, though it's had less success fostering adoption. Intel showed two WiMax technologies at the event.
First was a method squeezing 40 percent more capacity out of a WiMax networking station when handling voice over Internet Protocol (VoIP) calls. The system groups calls with similar characteristics so call-control data can be shared across each group rather than sent individually, said Vijay Kesavan.
Second was a peer-to-peer networking idea that ends up giving each device on a wireless network more network capacity. The technique helps smooth out areas with weak wireless network coverage and could let a person use a WiMax-enabled PC shoulder the battery burden instead of a nearby WiMax-enabled phone, said Intel researcher Ozgur Oyman, but it doesn't work as well when many of the devices on the network are moving instead of stationary.