AMD Gains on Intel in Marketplace
Advanced Micro Devices (AMD)—which has introduced new product lines, instituted technical innovations, and improved its production processes—has gained on archrival and microprocessor-market leader Intel in several areas recently.
During the third quarter of 2005, AMD's global marketplace share of processor sales rose to 17.8 percent from 16.2 percent in the previous quarter and 15.9 percent in the third quarter of 2004, according to Mercury Research, a market analysis firm. AMD's long-term goal has been to capture 30 percent of the market.
Intel's market share dropped to 80.8 percent in 2005's third quarter from 82.1 percent a year earlier and 82.2 percent the preceding quarter, Mercury noted.
According to Current Analysis, another market research firm, AMD had 49.8 percent of US retail sales of microprocessors for desktops and notebooks in October, compared to 48.5 percent for Intel. As
Figure 1 shows, this marked the first month that AMD surpassed Intel in this market, which doesn't include business or direct sales.
Figure 1. Current Analysis, a market research firm, predicts revenue from sales of the increasingly popular control systems will grow between 3.5 and 4 percent per year. "Continuing to hold this lead in the holiday season would be a colossal win for the company," said Matt Sargeant, Current Analysis' director of research.
In October, Current Analysis said, AMD's US retail-market share was 67.7 percent for desktops, up from 52 percent the previous month, and 31.5 percent for notebooks, up from 26.2 percent.
However, Intel, whose chips are more expensive than AMD's, still leads in revenue generated by US microprocessor retail sales, 57.6 percent to 40.1 percent.
Marty Seyer, AMD's senior vice president for commercial business, said unit sales increased for each of the company's product lines—chips for servers, desktops, and laptops.
AMD's gains are a big deal in the huge, slow-moving microprocessor marketplace, said Mercury analyst Dean McCarron. Having competitive products in all the different market segments has been at the heart of the gains, he said.
And, stated Seyer, offering innovative technologies has helped the company's market standing. These innovations include AMD's approach to multicore technology as well as an architecture that directly connects the processors, memory controller, and I/O to the central processing unit, thereby eliminating front-side bus architecture bottlenecks, he explained.
The company's Automated Precision Manufacturing process—which lets it deliver more than one type of chip on a single wafer—has also helped, according to Seyer. This provides production flexibility and lets AMD adjust and cater to demand better, he noted.
Some analysts contend Intel's legal woes, including global antitrust challenges, have helped AMD. Japan's Fair Trade Commission charged that Intel used rebates to force PC makers to limit the use of AMD chips. In addition, AMD has sued Intel for alleged anticompetitive practices.
McCarron said it may be too soon to mark AMD's gains as a long-term trend.
Intel spokesperson George Alfs said looking at the data over long periods of time is more meaningful and shows that during the last five years, Intel has had a dominating market share of between 80 and 85 percent.
The future key, according to Seyer, will be AMD's ability to meet energy-efficiency demands by providing chips that offer good performance per watt of electricity consumed.
New Technique Boosts Optical Networks' Speed
Stanford University researchers are working on a technique for using standard chip-making technology to build a device that switches laser beams on and off 100 billion times per second. If successful, this would make commercial optical networks 10 to 100 times faster.
The Stanford engineers developed a modulator-on-a-chip made from silicon and germanium that alternately blocks and transmits light from a continuous laser beam, splitting it into a stream of ones and zeros. The ability to switch the beams on and off at high speeds increases an optical network's data rate.
If successful, the device could initially enable a network to achieve data rates of 40 Gbits per second. With more work, the modulator could enable 100 Gbps rates, explained Stanford professor and researcher James S. Harris.
"There is a clear lack of optical data technology beyond 40 Gbps, and even at this rate, there is an unmet need for large-volume technology," said Karen Liu, research director of Ovum, a market research firm.
Researchers have previously found ways to turn laser beams on and off at high rates but only by using gallium arsenide and indium phosphate in modulators. However, these materials don't work well in typical semiconductor-manufacturing processes because they can contaminate silicon.
In response, the Stanford researchers experimented with various materials and discovered they could make an effective modulator-on-a-chip with silicon and germanium.
Manufacturers' ability to use the two materials, and thus to work with standard chip manufacturing techniques, would make the production process less expensive than building each modulator from scratch.
In addition, integrating the modulator and the accompanying electronics on one chip, instead of having them operate as separate components, would reduce costs and yield a more precise system.
If successful, Harris said, the Stanford optical modulator would be used first in networks, next as output drivers between chips in a system, and then for on-chip communications.
"The significance of [Stanford's] research is that there was better-than-expected performance from the system," noted Liu. "Over time, now that the possibilities in this direction have been identified, quite a few further advances can be made."
For the modulator to be commercially successful, said Harris, his team must show that it will work well in real-world situations and that it can be manufactured in large quantities.
Research Program Learns from Its Own Findings
Two University of Texas at Arlington professors have developed a program that finds patterns in data sets and then further analyzes the patterns to discover even deeper relationships.
The ability to deeply analyze and find patterns in data could make the Subdue (substructure discovery using examples) program particularly valuable in areas such as bioinformatics analysis, predictive tox-icology, and the examination of aviation incident reports and seismic events, said professor Lawrence Holder. The application could also find patterns in observed activities and behavior that could help identify potential fraud or even terrorist attacks.
Subdue, which professors Holder and Diane Cook developed, finds patterns in relational data represented as graphs. Holder said interpreting data and determining relationships between information points is easier when they are expressed as a graph.
Data mining finds initial patterns in information. However, Subdue goes a step further by examining and then learning from the initial relationships it discovers, he explained.
The program, which is available for use on the project's Web site (http://ailab.uta.edu/subdue), looks for patterns among data points on a graph. If Subdue finds patterns, it collapses those points into a single data vertex on the graph. Once these patterns are established, the program analyzes the new graph for additional relationships.
It can find patterns within freestanding data sets. It can also identify whether a data set has relation- ships similar to those found in other sets. With this latter capability, Subdue could compare a proposed chemical compound with others known to be toxic and predict whether it would also be harmful.
In addition to data mining, Subdue works with technologies such as artificial intelligence, using techniques like machine learning, knowledge representation, natural-language understanding and generation, and computer vision.
DOWNLOAD PDF 
