Dramatic Gains in Performance on the Horizon
Peter Singer, Editor-in-Chief -- Semiconductor International, 7/1/2006
It's always a delight when one of the large IC manufacturers releases the details of next-generation technology. Lately, the announcements have been on 45 nm technology, including one last month from Texas Instruments . Although not expected to go into production until 2008, companies such as TI want to get locked in with chip designers as early in the process as possible. Of course, this must be frustrating for those companies that are just now announcing acceptance of 65 nm technology, which actually is in production, but such is the world we live in.
The good news is that the 45 nm processes announced to date offer some huge improvements in performance and productivity. In TI's announcement, for example, the company states that the performance of multimillion-transistor, system-on-a-chip (SoC) processors will be increased by 30% while power consumption is reduced 40%. In addition, the move to smaller dimensions makes it possible to produce twice as many chips per wafer (compared with existing processes).
The even better news is that these dramatic improvements have been achieved — or more precisely, there's a very high level of confidence that they will be achieved when everything actually goes into production — with fairly evolutionary advances in materials and processes. Even though some process steps have yet to be finalized, TI's Ben McKee, vice president of silicon technology development for 45 nm, said that low-power devices will use the same materials as previous generations, and both low-power and high-performance devices will still rely on planar transistors fabricated in bulk silicon.
As reported in "TI Announces 45 nm Process ," what is new is the use of immersion lithography, millisecond "flash" anneals to form the source/drain junctions and, for high-performance chips, metal gates and silicon germanium. Interconnects will still be copper with thinner seed and barrier and a slightly lower k value for the dielectric (effective k of 2.5, down from 2.9). Strain technology, although not new, will be more aggressively implemented.
What TI (and presumably other companies that haven't yet made announcements) has avoided is a more advanced transistor structure, such as the lauded finFET. Although these structures have exciting promise and are sure to be eventually required, they also create manufacturing challenges, particularly in terms of lithography, photoresist removal and wafer cleaning. "As long as we can increment material and move forward with the same kind of a structure, that's how we will most quickly put cost-effective technology into production," McKee said.
While the manufacturing challenges with finFET-type devices do not seem insurmountable — plenty of companies have demonstrated working finFETs — they will almost certainly be more costly. But perhaps not as much as you'd expect; analyst firm Semico recently reported that silicon on insulator (SOI), which is required for finFETs, adds only 4-6% to the total manufacturing costs (see "SOI Equals Big ROI, Says Semico ,"). Conventional wisdom had it adding 10-15%. Easy for me to say, but this seems like money in the bank, whether these types of devices come into play at the 32 or 22 nm generation or beyond.
It may be difficult to imagine what applications chip designers will be targeting in 2008, but it's a safe bet that there will be a big focus on consumer products. Speaking in a recent webcast, George Scalise, president of the Semiconductor Industry Association, said, "Over half of the demand for semiconductors now is being driven by the consumer. Products such as 3G cell phones are a good example. Where they represented about 5% of the cell phone market a couple years ago, they now represent over 50%. In addition, the content in those 3G phones is much higher." With 45 nm technology, perhaps wireless users "will be able to run more simultaneous applications, such as a game with 3-D graphics in parallel to a video conference between the players, with e-mail synchronizing in the background," as TI suggested in its announcement.
Hmmm. Personally, I'll be happy when my laptop battery lasts more than 40 minutes, but it looks like the semiconductor industry will be able to keep some remarkable performance gains coming for years to come. What consumers choose to do with it is another story.
