Practical Chip Design

Jul 18 2008 2:04PM | Permalink | Email this | Comments (0) |
Blog This! using:  Blogger.com | LiveJournal |
Digg This | Slashdot This | add to Del.icio.us

A panel discussion organized by SEMI Thursday morning offered a debate on the question of transition from 300 to 450 mm wafers. Rather than "debate," the session might better be described as a review of SEMI's recently-published white paper, "SEMI/Equipment Suppliers' Productivity Working Group 450 mm Economic Findings and Conclusions." This necessarily presented a rather one-sided view of the question, as panelists went so far as to use graphs from the highly skeptical white paper to illustrate their presentations. But some of the points made were quite interesting.

The basic view of the panel seems to be that current enthusiasm for 450 mm wafers is based on "folklore" and a misreading of the current global market. As SEMI vice president for global standards and technology John Ellis said in his introduction to the discussion, "historically, the industry has reached decisions such as this on the basis of intuitive consensus." This is not a great idea, he argued, if intuition was wrong on some key points.

That concern led both SEMI and the International SEMATECH Manufacturing Initiative (ISMI) to build an economic model of the transition and evaluate it. Not surprisingly, given that the former organization is dominated by equipment suppliers and the latter by large semiconductor manufacturers, the two analyses came to very different conclusions. Ellis explained this difference by saying that while SEMI had researched each of the parameters in the model and estimated them based on best-knowledge current conditions, ISMI had inserted numbers into the model that were consistent with their intuitively-derived answer.

Ellis and the other panelists then went on to illuminate some of these differences in more detail. The first item they attacked was the seemingly intuitive idea that increasing the size of a wafer by itself increases fab productivity.

Not so, according to the historical data, the panel said. Ellis cited one Intel manager's report saying by Intel's own analysis, the 150 mm to 200 mm transition had produced no net benefit in die cost. Digging into the 200 to 300 mm transition in more detail, the panelists attempted to separate out the productivity contributions specific to the larger wafer from those due to other changes that were introduced at the same time. The result of their analysis was striking: "Productivity rose in this transition mainly due to the introduction of better materials-handling, higher station throughput, introduction of advanced process controls, and FOUPs [front-opening unified pods, if you are not a fab fan]," Ellis said. "There was little contribution simply from the larger wafers."

Sticking some salt in the now-impressive wound, the panelists went on to speculate, based on an extrapolation in the white paper, that the industry as a whole still has not recovered the cost of the transition to 300 mm, and might never recover it.

How could this be the case? In more detailed discussion, panelists sketched in the results from their economic model. Importantly, rather than assuming an average increase in die throughput for the entire line based on the change in area, the model looks at throughput changes in individual stages. That immediately highlights the difference between what the panel called beam-oriented devices and chamber-oriented devices. In the former category—including lithography, implant, and inspection tools, according to Ellis—productivity actually increases little if at all with wafer size. The tool still has to step or scan the surface, and still moves at about the same rate. In fact, with the bigger, heavier wafers that process, like materials-handling, may be slowed.

Even with chamber-oriented devices, twice the area doesn't mean twice the capital productivity, pointed out Dave Hemker, vice president of new product development at Lam Research. "You can make a plasma chamber larger, but in doing so you still have to maintain the same very uniform plasma density," Hemker warned. "That isn't easy. It means a bigger turbo pump, and it means a much more powerful RF generator. We have learned that RF generator cost does not go up linearly with power—it begins to increase much more rapidly shortly beyond current power levels. So you cannot achieve linear scaling in cost with wafer area."

This distinction is critical in looking at 450 mm, said Masayuki Tomoyasu, senior vice president at Tokyo Electron, because today beam tools make up a larger portion of the finished wafer cost than at 200 mm. "If that portion of the cost doesn't scale, what do we gain?" he asked. Further, he pointed out "most of the benefit we received from the 200 to 300 mm transition was from FOUPs, better automation, and similar improvements. We can't do those changes over again—we can't introduce another FOUP, or automate the fab all over again."

The panel also attacked the case for needing larger wafers in the first place. Ellis said it was a myth that monotonically increasing die size is forcing a move to 450 mm to maintain the die-per-wafer ratio. "In fact, the average size of dice has been stable to slightly decreasing," he said.

Nor, panelists said, is the industry as a whole seeing the kind of demand that was used to justify 300 mm. "At 300, we were seeing dramatic demand growth," observed Iddo Hadar, CTO at Applied Materials. "Today, the growth projections for the industry are much slower. That requires a different logic than we used back then. We see 450 mm wafers being unnecessary to meet demand until the 2020s.

"If we had exhausted the opportunities for productivity growth on 300 mm wafers, it would be different," Hadar continued. "But we have not. In fact there is a risk that focusing on 450 mm would distract us from further work on 300 mm equipment, just when the biggest bang for the buck is at 300. What we are saying here is not that we are pessimistic about 450 mm, it is that we are strongly optimistic about the returns still to be had from further 300 mm R/D."

To this point both industry consultant Walter Class and Skip Miller, director of strategic marketing at ASML, assented. Both emphasized that traditionally, and according to the economic model, the best way to drive productivity has been to aggressively shrink features, resulting in smaller dice or more value per die. And most panelists underlined a point made by Ellis in his opening remarks, that the equipment industry already was spending less than it needed to on R/D just to stay on the ITRS Roadmap for feature size.

Taken together, the panel comments painted a picture of a proposed transition that would entail very significant technical and business risks for both equipment-makers and fabs, that would not promise a significant change in productivity, that was unnecessary to meet forecast demand, and that would likely distract the equipment industry from developments that actually would help. This could only change, the panelists said, when some of the underlying inputs to the economic model changed: demand growth, the size of the technology barriers, or perhaps the appearance of some processing breakthrough that could only be implemented on 450 mm equipment. The time may come for 450 mm, the panel suggested, but that time was not now.