SAN FRANCISCO Intel Corp. chief technology officer Patrick Gelsinger will reveal several research breakthroughs that extend the reach of silicon devices in his keynote address Thursday (Feb. 28) at the Intel Developer Forum. They include wireless CMOS-radio-based sensor networks, microelectromechanical systems (MEMS), and photonics chips that compute with photons instead of electrons.
These technologies will not replace silicon, Gelsinger will say in a speech entitled "Expanding Moore's Law," but rather will extend its reach. Moore's Law, which posits that technology advances will allow the transistor density of a silicon chip to double roughly every 18 months, will extend "as far as the eye can see," Gelsinger said in a briefing.
"We are looking out onto a broad landscape of new technologies. The first one is wireless, which I entitle 'radio-free Intel.' The second is wireless sensor networks where you take the wireless technology and build self-configuring, dynamic networks using on-chip MEMS sensors. And the third is silicon photonics, where we imagine the opto-processors of the future," said Gelsinger.
IDF has become a "geek fest," where even Intel competitors make product announcements. According to Gelsinger, 150 companies participated this year and made over 100 new product announcements, not counting the 30 product announcements Intel made. Gelsinger said that "geeks" collectively asked Intel to outline its long-term goals after last year's IDF, resulting in this year's presentation of its five-year goals.
"These are technologies we are actively researching, and we are showing some for the first time ever we want to paint a picture of where we might be going in the next five years, but these are not product statements," said Gelsinger.
By implementing radio circuits directly in CMOS silicon, according to Gelsinger, future Intel systems-on-a-chip will have built-in reconfigurable wireless network hookups. And with intelligent software-defined radio algorithms, these wireless connections will be dynamically reconfigurable to take advantage of changing spectrum opportunities as users move about, transparently switching from LANs to WANs, for instance.
"Our vision is that this CMOS radio circuitry becomes the corner of every die. And as they get cheaper and smaller, riding Moore's Law, guess what? We'll have these little slivers of silicon that you can paint into your walls and become a sensor environment for every place that you live, work and communicate," said Gelsinger.
The second "leg" of Intel's CMOS radio aspirations, according to Gelsinger, is based on its "labettes," established near such universities as Carnegie Mellon, Berkeley, and the University of Wash. In cooperation with its labettes, Intel is investigating, among other things, how MEMS technologies can create microscopic mechanical devices that solve the tougher electronic problems of putting RF "black magic" on a standard CMOS chip.
"We are using MEMS to build the RF componentry directly into silicon the capacitors, inductors, switches and all those. And we are showing our first MEMS circuits produced by a Intel fab. With MEMS we'll have enough processing power that we will be able to run multiple software-defined radios out of the same piece of silicon," said Gelsinger.
By arranging the RF circuitry to be configurable, Intel envisions its future MEMS-based CMOS chips to dynamically reconfigure for "always on" wireless connections, regardless of where a user roams, from wired LAN to wireless LAN to WAN.
"The idea is that I can start out on my wired LAN, but when I unplug my computer it automatically switches to the wireless LAN, then I leave the building and I seamlessly switch to my wireless WAN. And my software was intelligent enough to be listening on those other frequencies in order to determine when is the best time to switch between networks, while keeping all your IP Internet Protocol and voice connections live," said Gelsinger.
He also described for the first time Intel's effort to harvest ultrawideband radio technologies which are due to be "christened" by the Federal Communications Commission soon. Ultrawideband radios use spread-spectrum algorithms to divide data into parallel streams that can be broadcast at very low power levels across the entire radio spectrum, and are then reassembled by ultrawideband receivers. Because the energy of a spread-spectrum transmission is lower in any one band than its traditional noise floor, such ultrawideband transmissions can theoretically operate in the background while all the current activity in those bands continues unaffected.
"While Bluetooth is a wonderful technology, it is low-bandwidth and short range," Gelsinger said. "If you really want wire replacement, which is the goal of Bluetooth, you need big bandwidth. For instance, USB-2 is 480 Mbits/second almost ten times faster than a 5-Mbit/second Bluetooth network. Whereas, we believe that ultrawideband will be a technology that at short range will provide very high bandwidth. We are doing research to hopefully bring ultrawideband technology up into the 500-Mbit/second range, because at that point you really do have a wire-replacement technology."
As the Defense Advanced Research Projects Agency funds national security applications such as "smart dust" cheap wireless MEMS sensors that can be dropped from drone aircraft onto a battlefield Intel envisions civilian applications of such wireless sensor nets.
"Our overall effort in this area we call 'proactive computing' because we are embedding computing capabilities in places it hasn't been before," said Gelsinger.
Follow the bouncing ball
At Intel's lablette near the University of California at Berkeley, the company has been quietly developing smart-sensor networks that combine self-configuring algorithms with MEMS sensor miniaturization, resulting in sensor chips with their own dynamic on-chip wireless communications capabilities. At his presentation, Gelsinger will demonstrate their progress with a live, self-configuring wireless network that dynamically reconfigures on-the-fly to run two different applications. A wireless sensor net will be contained in several hundred beachballs that Intel will release onto the crowd listening to Gelsinger's keynote.
"The beachballs bouncing around the audience show the dynamic reconfiguration that is going on while the network is running. Imagine you are at Disneyland and there are sensors all around and you stick one on your kid so you always know where they are," he said.
The demonstration shows how the network dynamically reconfigures itself, depending on each beachball's location as they bounce around the room. A display shows how the network self-organizes into a topology that matches the physical location of each beachball. Then when the beachballs settle down, audience members will remove the sensors from the beachballs while the network is loaded with a "voting application" where the beachballs' sensors becomes the input device for voting.
"Nobody knows what the voting application is, but I'll tell you: It's deciding where Intel's next IDF should be. People can choose between Hawaii and Silicon Valley for the next IDF, but it's just to show that we can load different applications into the network," said Gelsinger.
Optical silicon chips
Today's optical filters cost upwards of $10,000 each. The devices, which select among the different frequencies of laser light in a communications channel, may select a single communications channel by implementing a narrow bandpass for a certain range of frequencies. Intel's optical chip a research device shown at IDF for the first time puts the functionality of today's $10,000 chip onto a $1 chip.
"We are showing an optical filter that is built directly in silicon. There is a different optical defraction characteristic for light based on either the bias or temperature of silicon. And you actually change this defraction index to select different bandpass frequencies," said Gelsinger.
While remaining entirely in the domain of light, the selectivity of Intel's all-optical bandpass filter can be tuned by setting simple software options. Thus instead of buying dedicated optical filters, Intel's reconfigurable optical-silicon chip can be programmed to become whichever type of filter is required at the moment, from low-pass, to bandpass, to high-pass, and at any desired wavelength.
"You can imagine your network operator in, say, Orlando, sends a network message down to an optical filter chip in Chicago, telling it to reconfigure itself to pluck out, say, a 50-GHz-wide channel for a customer there," said Gelsinger.
Intel has already been putting more and more of the electronics of fiber-optic communications onto its chips, but this all-optical filter demonstrates how Intel Research is also investigating how to put the actual optical operations themselves onto silicon chips. In the end, Gelsinger predicted, Intel will be building all-optical silicon chips that use photons instead of electrons to compute.
"We want to take optics from the WAN to the enterprise to the LAN to the data centers to the rack and finally from chip-to-chip," he said. "Today, making an optical connection costs many tens of thousands of dollars. Our job is to make that pennies, resulting eventually in the opto-processor a microprocessor with direct optical interfaces."
This hypothetical "opto-processor," as Gelsinger calls it, would eliminate the need to have two separate systems at a data center. Optical communications is presently handled by one set of equipment at a data center and data processing is handled by another set of equipment, but Gelsinger said Intel's future opto-processor will eliminate this setup by collapsing both functions into silicon chips that compute with photons instead of electrons.
More Intel Developer Forum coverage.