The little wavelength that could
Can semiconductor manufacturers avoid a costly conversion?
By Geoffrey James, Contributing Writer -- EDN, June 1, 2006
|
Sections:
Disaster ahead?
|
The future of photolithography is supposed to lie in extreme ultraviolet (EUV), which uses a beam of light far smaller than today's 193-nanometer deep-ultraviolet (DUV) technology.
"EUV should come into play at the 32-nanometer node, somewhere around 2010," says David Attwood, a professor at the University of California, Berkeley, who specializes in electromagnetics at short wavelengths.
IBM, however, recently announced a way to extend DUV down to 32 nm and perhaps even to 22 nm and beyond. In doing so, IBM may have scuttled EUV as a financially viable technology.
IBM's new technique, "high-index immersion," uses "an organic fluid that allows the light to be focused more precisely," according to IBM researcher Bob Allen.
"Although the technique still needs further development, we believe that this technique is likely to be more economical than other alternatives at sub-30 nanometers, including EUV," Allen adds.
This is good news for semiconductor firms, because converting to EUV will be a "major re-tooling event, with the cost for the industry running into the billions," says Risto Puhakka, vice president of VLSI Research. And that expense might turn out to be a very bad investment if EUV fails to prove as long-lived as DUV.
As Moore's Law has progressed, the basic economics of chip making have been changing. At each successive process node, achieving an acceptable yield becomes geometrically more complex and expensive. A state-of-the-art fab, for example, will cost an astounding $3.5 billion in 2007, according to TSMC, greatly limiting the number of companies that will be able to afford to build them.
Design costs are also ballooning at each process node, making the newest processes only practical when chips are manufactured in increasingly high unit volumes. As a result, the industry is seeing fewer state-of-the-art design starts than in previous years, targeted at fewer state-of-the-art fabs.
A by-product of this trend is that it decreases the unit demand for each successive generation of photolithography equipment.
"Sales of immersion scanners are running at about 30 per year," explains Puhakka. "These are very small numbers, considering the extensive development that's gone into the new technology."
By the time EUV is slated for prime time, Puhakka expects, yearly demand for lithography tools will be as tiny as 10 per year. With volumes that small, photolithography tool makers will be forced to charge a very high per-unit price for the tools, simply to justify making them.
 David Attwood, University of California, Berkeley |
That will be true whether or not the industry converts to EUV or just to a new and improved DUV. However, converting to EUV will be far more expensive than a DUV upgrade, because EUV involves an entire retooling of the photolithography industry. "It involves a complete changeover in every part of the supply chain in order to manufacture a set of completely new tools, not to mention the tools needed to make those tools," says Craig West, director of applications at Toppan Photomasks.
That expense would be justified if EUV proves as long-lived as DUV, which has survived through four nodes—longer than any previous photolithographic methodology (see graph). For that to happen, though, EUV will need to be introduced at 32 nm, survive through 22 nm and 16 nm, and then bring the industry safely to 11 nm. After that, there's probably nowhere to go with silicon, regardless of photolithographic technology.
"The limit of component size for anything recognizable as a semiconductor appears to be somewhere around nine nanometers," explains Klaus Rinnen, a semiconductor analyst at Gartner.
Disaster ahead?
However, it's not at all clear that it will actually be possible to create circuits out of such tiny components. In the past, predictions of the demise of silicon circuitry have been premature. Fifty-year industry veteran Fred Zieber, of Pathfinder Research, once heard the then-head of IBM Microelectronics claim that "we'll be lucky if we make it to half a micron." However, even Zieber agrees that, this time, the end is nigh. "At some point—it could even be before 22 nanometers—it breaks down and the economics begin to change," he explains.
If workable circuitry proves possible at 22 nm and smaller, the cost of designing and manufacturing such chips may be economically feasible only if unit volumes reach several orders of magnitude beyond the magnitude of any of today's semiconductor production runs.
If either the physics of semiconductor circuitry or the economics of semiconductor manufacturing break down just as major semiconductor firms convert to EUV, the result would be an industry wide disaster of Biblical proportions. Those firms participating in the abortive leap forward would be stuck paying for a massively expensive conversion whose costs could never be recovered.
To understand how such a debacle might look, Dr. Robert Castellano of The Information Network cites the time when the U.S. semiconductor industry converted en-masse to new mask-making technology for 64K DRAM.
"The Japanese stayed with the previous generation, and, when the new technology didn't work as expected, that failure wiped out the U.S.-based DRAM business," he says.
Imagine that same dynamic, playing itself out worldwide, with the big losers being the handful of top semiconductor firms that still had the financial clout to buy into the EUV conversion. It would be the end of the worldwide industry as we know it.
In other words, it's vastly to the advantage of semiconductor manufacturers to stay on the DUV wagon as long as possible, which is why IBM's announcement is so significant.
Wavelength Longevity
| 1 | 436-nm wavelength | Mercury vapor lamp (g-line) | Blue | 500-nm features (half micron) |
| 2 | 365 nm (i-line) | Mercury vapor lamp | Violet | 350 nm features |
| 3 | 248 nm | Krypton fluoride excimer laser | 250 nm | |
| Deep UV | 180 nm | |||
| 130 nm | ||||
| 4 | 193 nm | Argon fluoride excimer laser (AFEL) | Deep UV | 130 nm |
| 90 nm | ||||
| 65 nm | ||||
| 5 | 193 nm | AFEL with water immersion | Deep UV | 65 nm |
| 45 nm | ||||
| 32 nm | ||||
| 6 | 193 nm | AFEL with high-index immersion | Deep UV | 32 nm |
| 22 nm (TBD) | ||||
| 7 | 30 nm? | unknown | Extreme UV | unknown |
