Will 45 Nanometer Chips Make Two Warring Camps?
February 5, 2007 Timothy Prickett Morgan
Chip makers IBM and Intel have announced their research efforts to push the boundaries of chip making processes as they try to keep the Moore’s Law price/performance curve alive. IBM and Intel both divulged some of the techniques that they will employ in their 45 nanometer chip making processes, trying to steal headlines from the other. But that may not be the real story. It probably won’t be long before there are only two organizations in the world with the resources to extend chip technology and push Moore’s Law. One of these companies will be Intel, of course, and the other will probably end up being a co-operative of everyone else, including IBM, Advanced Micro Devices, Freescale Semiconductor, Sony, Nintendo, Toshiba, and Microsoft. Right now, Texas Instruments does the research and development that allows it to make digital signal processors and other chips, such as Sun Microsystems‘s UltraSparc-IV and Sparc T1 processors; Fujitsu similarly does the research to create its own processes to create the Sparc64 line of chips. It may not be too long before Sun and Fujitsu join the Intel camp, much as Hewlett-Packard did when it decided in 1996 to use Itanium processors in its high-end servers instead of homegrown PA-RISC processors made in HP’s own factories. Intel and Sun have just forged an agreement to have Sun use Intel’s Xeon processors in its entry and midrange servers, and considering the animosity between Sun and Intel for the past decade, it might be hard to imagine the two companies cooperating on chip making processes. According to Carl Johnson, the main analyst at semiconductor watcher InfraStructure, TI has just laid off 500 of the PhDs who work on advanced chip making processes and is going to stop pursuing Moore’s Law by itself. Now, maybe TI will team with Intel-HP or the IBM-AMD collective. But whatever it does, Sun is counting on advanced chip processes for its future “Niagara” and “Rock” multithreaded chips–processes that are intimately tied to the design of chips themselves. You cannot divorce the chip from the process–and this will be particularly true as we move from current 65 nanometer technologies through 45 nanometers, down to 32 nanometers, and on to 21 nanometers. The more you shrink, the more the properties of the process affect your yields, and then your ability to actually ship a chip and then a product based on it. If Sun goes into the Intel camp for chips, it might as well go into it all the way and use Intel as a foundry. The other option, which would be to use Fujitsu as a foundry, has been tried before. Fujitsu used to be a second source for UltraSparc-II chips in the 1990s, but Sun foolishly went sole source with TI with the UltraSparc-III chips, which were massively delayed and caused Sun a lot of embarrassment and arguably billions of dollars in lost sales. Fujitsu is now late with the “Jupiter” Sparc64-VI dual-core chips that are going into the “Olympus” servers that were jointly developed with Fujitsu. Fujitsu is already an Intel partner, selling Xeon and Itanium servers, and it would not be difficult to conceive of all four vendors–Intel, HP, Sun, and Fujitsu–all working together on future chip processes and using Intel as a foundry, much as IBM, AMD, and the game console and electronics makers have already thrown in together and IBM has become the foundry for all but AMD. And that could change, too. In the meantime, IBM and Intel want to keep everyone focused on the advances they have made with 45 nanometer technologies, because they want to ensure their customers that they can deliver on Moore’s Law for a little bit longer with some assurances. Over at IBM’s East Fishkill, New York, chip-making plant, Big Blue says that it has worked with AMD, Sony, and Toshiba to create a technology called high-k metal gate that will allow IBM to put 45 nanometer processes into place using existing fabrication lines in 2008. Low-k dielectric technologies are already in use in 90 nanometer and 65 nanometer processors, as is silicon on insulator (SOI) technologies and copper wiring (all of which were perfected by IBM after many years of research). IBM did not detail what specifically it was doing, but the expectation is that scientists have figured out yet another material to mix into the silicon dioxide insulator to allow chip circuits to shrink further. While IBM did not say how it implemented the high-k metal gate technology, Intel said that it had created a similar technology and was using a rare metal called hafnium as part of the process to get its circuits down to the 45 nanometer size. There are other parts to the recipe, but Intel is keeping its lips sealed. TI said that it had also created high-k metal gate processes for 45 nanometer chips, and will sample the technology at the end of 2007 and put it into production by the middle of 2008–when Sun’s first Rock Sparc processors are due, by the way. Intel hopes to have its 45 nanometer processes humming by the end of the year, and IBM is expected to gets its own online sometime in the first half of 2008.
|