Intel’s announcement of metal-gate transistors with high-k gate dielectrics swept the mass media by storm in late January. Mainstream news outlets that don’t know a transistor from a resistor scrambled to find industry analysts who could explain the technology in newbie terms. Intel’s accomplishment was touted as a major breakthrough—the biggest advance in transistor design in more than 30 years.
Within hours of Intel’s announcement, IBM pushed its top engineers in front of microphones to assure a nervous world that they too have developed metal-gate transistors with high-k gate dielectrics.
What’s amazing about this furor is that Intel’s “news” is more than three years old. On November 5, 2003, Intel issued a press release announcing exactly the same thing. (I still have my copy.) The day after that announcement, Intel presented a technical paper on the subject at a conference in Tokyo. Intel’s press release from November 2003 reads almost word-for-word like the news stories that broke in January 2007. The press release even said that the new transistors could appear in Intel processors “as early as 2007, as part of the company’s 45nm manufacturing process”—an uncannily accurate prediction.
R flacks should study this incident. It’s a good lesson in exploiting the news media’s short attention span.
Intel’s 2007 announcement did include two details missing from the 2003 release, although many news stories omitted those details. One of the new dielectric materials is hafnium, which replaces silicon oxide. (The metal-gate material is still a secret.) And the new transistors will indeed appear in an Intel processor manufactured in the company’s 45nm fabrication process this year. (It’s Penryn, the next Intel Core 2 Duo.)
I don’t minimize Intel’s accomplishments. The new transistors significantly reduce current leakage and switching power, allowing Intel to build faster, cooler, lower-power processors. But I view this step as more evolutionary than revolutionary. Similar advances in recent years include silicon-on-insulator transistors, strained silicon, and copper interconnects. Future advances in the same vein are liquid-immersion lithography and tri-gate transistors, both coming soon. Engineers everywhere are busting their butts to prolong the life of Moore’s Law. Intel’s engineers deserve a big share of the credit—which, in this case, was 39 months overdue.
Tom Halfhill was formerly a senior editor for Byte magazine and is now an analyst for Microprocessor Report.