On the scene of the Intel Developer Forum (IDF), the pattern of Intel caused a sensation. Paul Otellini was holding his hands a silicon wafer which he assured that it was a prototype for the manufacture of chips that will be able to share data at speeds of the order of the terabyte per second. As a comparison, the library of Congress fully digitized, regarded as the most voluminous of the world, contains 20 terabytes of data. They will also offer calculation powers measured in teraflops. The journey is easy to measure: the first supercomputer who posted a one teraflop computing capacity has been installed as ten years ago, at Sandia Lab military laboratory, near San Francisco.
Of course, for such performance are not possible with current technologies for the production of microprocessors. Use in effect of copper wires to transmit data between the various components of the chip. All experts estimate that one reaches soon possible maximum speeds, including the fact of the quantities of heat by this mode of transmission of information. This will limit then the race to power computers.
The recent advances made by Intel in the field of optical transmission opens new opportunities. Following the inaugural presentation of Paul Otellini, chief technology officer of the California group, Justin Rattner, came, in turn, comment on this new technology direction. According to him, Intel could produce in series in five years these new chips integrating technologies laser hitherto reserved for components used in communications networks. A few days prior to the IDF, Justin Rattner had indeed presented a new concept of chips designed in collaboration with the University of California at Santa Barbara (UCSB), near Los Angeles.
Intel named this Hybrid Silicon Laser technology as it combines the advantages of the transmission of data at the speed of light in the techniques of cheap production of Silicon. It is obtained using all the Silicon, one hand, and Indium Phosphide, on the other hand, a material commonly used for producing lasers for the telecommunications industry.
Communications equipment use long lasers to send impulses through networks optical fibre at the speed of light. But these lasers claim particularly expensive optical components which, until now, would reserve this type of applications. Using these lasers with conventional silicon components, Intel and the UCSB researchers would therefore succeeded in bringing pulses of light transported by these lasers up to these electronic components can transform again into electrical signals, and therefore digital data. If these lasers can carry light without generating too much strong heats on the chip itself, the industrial solution seems in sight. "Our recent results are very significant, it was the last real technological barrier to eliminate" provides Mario Paniccia, Director of Intel's Photonics Technology Laboratory.
But even in this case, remains the economic equation. Because if Intel and the UCSB researchers claim to be able to integrate hundreds of these localise into a single chip hybrid, the current cost of such lasers about 50 $ unity makes impossible an economically viable marketing to replace microprocessors for computers. Intel is confident in its ability to bring through extensive industrialization, this unit cost only 1 dollar by the end of the Decade.
A market still open
What will therefore serve soon dispose of bullets which the calculation of the more mundane ability will be measured in teraflops and their capacities of terabytes communications "In the coming years, new online services hosted by computer centres connecting millions of servers, allow users to access to personal data, multimedia and applications from any portable terminal" provides Justin Rattner. In other words, any user can therefore tomorrow, from its iPod portable music player, watch TV channels live high definition or orally examine search engines using techniques of data mining, to respond in the same way to billions of simultaneous requests...
But if technological future seems well plotted, there is no indication that it will be confiscated by Intel. Indeed, other firms in particular start-ups are working on the same type of technology and, apparently with some advance.
That is why last August, Luxtera, a California-based start-up, presented the first copies of a chip combining also optics and Silicon. The commercial version of the 2010 Lux should be presented by the end of the year and business leaders ensure that they are as well designed markets for consumer electronics (transmission of video signals in high definition) than those of the centres of very high-capacity servers. Unlike copper wires, the optical to indeed carry information over large distances. Other firms, such as Zarlink Semiconductor, also work in the same directions. Some know go faster (or better) and Intel