SEPTEMBER 2007 (Vol. 40, No. 9) pp. 20-22
0018-9162/07/$31.00 © 2007 IEEE
Published by the IEEE Computer Society
Published by the IEEE Computer Society
|Company Says Diagonal Wiring Makes Chips Faster|
|New Technology Transmits Data via Visible Light|
|Researchers Herd Computers to Fight Spyware|
PDFs Require Adobe Acrobat
Company Says Diagonal Wiring Makes Chips Faster
A leading semiconductor-design software-tool vendor has developed an approach that would increase chip performance by running wiring diagonally—as well as horizontally and vertically, as is the case now—between transistors, on-chip memory, and other processor elements.
Cadence Design Systems says this technique would also reduce manufacturing complexity and improve energy efficiency.
Typically, the copper or aluminum wires in a chip run only along horizontal and vertical axes, an approach known as Manhattan routing. The interconnections deliver power and carry data, explained Chi-Ping Hsu, Cadence's corporate vice president for integrated-circuit digital and power forward.
In today's complex chips, which can contain "tens of miles" of wiring, there can be as many as 15 layers of circuitry, Hsu noted. Each layer generally has wiring running in just one direction. This avoids having wires in a single layer cross one another, which could cause short circuits. Additional interconnects run between the layers.
Cadence's X Architecture approach uses wiring that runs diagonally, along with the connections that run horizontally or vertically, in some layers.
Diagonal wiring enables shorter, more direct connections between processor elements in many cases and thus reduces the manufacturing complexity that Manhattan routing can require over longer distances.
The X Architecture saves energy by having signals travel between points over direct diagonal paths, which are shorter than the more indirect paths that use both the horizontal and vertical wiring in multiple layers, as Manhattan routing can necessitate.
Adding diagonal wiring also gives designers more routing options and thus greater flexibility, Hsu said. The architecture lets chip designers make more design trade-offs to get the chip characteristics needed for specific applications, such as lower dynamic operating power in mobile devices.
Also, introducing diagonal paths lowers the amount of wiring and concentrates more connections within each layer, which could decrease the number of layers. This would cut production costs. In addition, less wiring would reduce problems such as crosstalk and current leakage into the substrate, which typically decrease chips' performance and energy efficiency.
Agere Systems, which LSI Corp. recently acquired; ATI, which AMD purchased last year; and Teranetics use the X Architecture in some of their chips.
Cadence has had trouble convincing chip makers to use the X Architecture partly because it is largely untried, would force them to buy new equipment, and is supported by only one vendor, said Gabe Moretti, a veteran observer of the electronic-design-automation industry who operates the Gabe on EDA Web site ( www.gabeoneda.com).
Also, he added, "I believe that what has been demonstrated so far has yielded only marginal improvements over traditional routing techniques."
According to Hsu, the companies trying the X Architecture now are aggressive, risk-taking, and pushing the state of the art.
New Technology Transmits Data via Visible Light
Researchers are working on a technology that would utilize the LEDs in common light fixtures to transmit data over short distances.
"When LED light fixtures become commonplace, their use as a communications device will also become commonplace," predicted Shinichiro Haruyama, a Keio University professor, vice president of the Visible Light Communications Consortium (VLCC), and one of the original VLC researchers.
"LEDs are becoming popular," he explained, "because their light-emitting power is becoming large enough and their lifetime is longer than those of conventional incandescent lamps and fluorescent lights." Also, he noted, LEDs' power efficiency is greater than that of incandescent lamps and about same as that of fluorescent lights.
VLC—recently implemented in Japan—would be most useful in locations where cellular and other radio-based systems don't work optimally, such as dense urban areas.
Researchers created VLC modulators that turn the power to an LED on and off—too rapidly for the human eye to process—thereby creating binary data's ones and zeros.
A VLC system would include LED lights; photodetectors, which sense light and convert it to electricity; and a small transceiver chip that relays data between systems.
The technology could work with light sources other than LEDs. The researchers say they chose to experiment with LEDs in the belief that they will probably be used more than other types of lights in the future.
VLC's top transmission speed with common LEDs will probably be 20 or 30 megabits per second, although it could top out as high as 80 Mbps, according to Haruyama. Although the technology may be slower than some other types of short-range communications, the main benefit will be its pervasiveness, he explained.
A type of LED designed strictly for communications could offer data rates up to 500 Mbps in the future, although it would not provide a very bright light.
The technology's transmission range in tests conducted with an LED traffic light was 30 meters. However, Haruyama said, improving the receivers eventually could increase the range to a kilometer.
VLC could link to other types of wired or wireless communications networks.
In addition to communications, VLC could be used for location detection. The light in a building could send a unique location identifier, the address and room number, and the structure's longitude and latitude to a visible light receiver.
Haruyama said it remains to be seen whether adding communications features to LEDs would make them significantly more expensive and, if so, whether manufacturers would do so.
Light-based communications are not affected by electromagnetic interference. However, the technology is limited to locations at which transmitters could send light to receivers either directly or by bouncing it off another object.
Already, said Haruyama, a Hong Kong University professor is using an LED traffic light to send sound, and some Daimler Chrysler researchers are using LED head lamps and tail lamps for sending data between cars.
The Tokyo-based VLCC ( www.vlcc.net)—which includes companies such as Mitsubishi Heavy Industries, NEC, NTT DoCoMo, Sony, and Toshiba—coordinates VLC research and standardization. The consortium is looking at potential business models for commercializing the technology.
Researchers Herd Computers to Fight Spyware
A joint research project of Harvard Law School's Berkman Center for Internet & Society and Oxford University's Oxford Internet Institute is using a technique called herd computing to fight spyware and various types of malware.
Herd computing uses multiple computers from voluntary participants. Software runs on each client, gathers data about the computer's vital signs—such as CPU performance, memory consumption, pop-up windows, system crashes, and the activities of processes running on the computer—and sends it to a server.
The server then collects and analyzes the data, presents it in a form viewers can see, and anonymizes the material to eliminate IP addresses and other information that identifies individual machines.
Participants and network administrators monitoring the herd can view the processed information via either a browser or a light client with browser functionality.
Data transfer between elements can occur via any of a number of wired or wireless Internet-based communications technologies.
In essence, herd computing lets participants who are considering running new code see how many other machines have chosen to run the software, as well as the effects it has had on the computers, explained Oxford professor Jonathan Zittrain, who cofounded the Berkman Center.
Using participant experience or the observed behavior of applications, members could decide whether to download the software, based on their organization's computing environment and usage policies, Zittrain said.
Herd computing lets participants configure their computers to decide the extent to which they want to participate and thus how much processor and bandwidth they will contribute, he noted.
The researchers plan to include open APIs in future pilot herd networks so that users can write widgets enabling them to work with collected data as they choose.
News Briefs written by Linda Dailey Paulson, a freelance technology writer based in Ventura, California. Contact her at firstname.lastname@example.org.