450 mm: A Promise Postponed

With the possible exception of immersion and extreme ultraviolet (EUV) lithography, there have been few instances where our industry has found itself as sharply polarized about a technology transition as it is over the perceived need to migrate from a 300 mm wafer to a 450 mm wafer. We thought it would be useful to examine the pros and cons of this transition.

In general, the equipment manufacturer's point of view regarding this shift seems to range from cautious to hostile. The suppliers that we queried agreed to speak anonymously. What follows is a consolidation of their views and opinions.

Overall, some of those interviewed view 300 Prime as the way to go, rather than migrating to a larger wafer. The 300 Prime goal is to improve 300 mm productivity, while also serving as a bridge to 450 mm production by providing productivity improvements that will be scalable to that size.

Although none of those who we interviewed view a possible transition to 450 mm as a technical impossibility for processing equipment, it was often pointed out that, industry-wise, there are limited resources available to invest in development. As one put it, “Shifting focus to one thing usually comes at the cost of another. We must set priorities consistent with the interests of the whole industry.” Thus, the hurdle seems to be mainly economical — the high cost of developing next-generation equipment, and the concern that it may never be paid back by the potential penetration of that manufacturing toolset. Others indicated doubts that adequate substrates of that size could be provided in sufficient volume. There appears to be a general consensus that so long as customers are satisfied with the yields they are getting, they will not view transitioning to a larger wafer as a matter of interest, certainly not for the near term.

Some manufacturers expressed a strong disagreement with the International Technology Roadmap for Semiconductors (ITRS) perspective, which has 450 mm wafers in production by 2012. One of those interviewed stated that, “The ITRS has acted irresponsibly in this prediction.” This person added that as far as he could see, it is obvious that 450 cannot be in production by then because no one is working on developing the toolsets to make it happen. “They should reexamine this,” he stated, “to avoid confusion and enable people to focus on what matters, such as 300 Prime.”

An important part of the industry’s ability to stay on its cost/performance curve has been the transition to larger wafer sizes. The most recent move was from 200 to 300 mm, and now 450 mm is being considered in less than an enthusiastic manner. (Source: Sematech)

When it was suggested that postponing the move to 450 mm might delay innovation, an industry observer pointed out that in the semiconductor world, major technology innovations run on a cycle roughly three years in duration. While historically these have been associated with transitioning to a larger wafer, ideas and initiatives to improve equipment and processing technologies are first discussed internally or in conferences and committees. “What happened,” he said, “was that the window of opportunity to implement these developments came during a wafer size transition.” He added that this would then be the time when fabs would have to be built from the bottom up, providing this wider opportunity to implement changes.

Now, however, the belief seems to be spreading — at least in some quarters of the equipment manufacturing sector — that behind the benefit of a wafer size transition are productivity innovations and the cost of toolset scaling. “It would be better to unbundle the productivity improvements from the wafer scale-up cost, which is what 300 Prime aims for, and go after significant and discontinuous productivity innovations that have been associated with wafer size transitions without paying wafer scale-up costs.” The biggest opportunities are viewed as being in cycle-time reduction and increased fab agility.

Would some like to see 300 mm be the final wafer size? This is difficult to say. Some manufacturers think that the way processes for 32 and 22 nm seem to be shaping up, it does not appear that a wafer scale-up would be of great benefit. As it was expressed, “Perhaps we aren't too clear about what happens at 22 nm, and certainly beyond it. At that point, maybe the right wafer size might be 1000 mm, or we'll be use a completely different manufacturing process. However, from what we know about manufacturing technology right now, there's no strong motivation for 450.”

This opinion is no wholly unjustified. Many equipment suppliers are still smarting over the return associated with the move from 200 to 300 mm, which was complicated by a massive downturn, so their lack of enthusiasm to jump on the bandwagon is understandable.

Other perspectives

Several leading chipmakers do not agree with this view. Some seriously warn that unless the jump takes place soon, the industry will face higher costs per square centimeter of silicon. We interviewed three industry experts who agreed to go on record on this subject.

The first was Peter Silverman, director of Intel's Equipment Technology Strategy, Technology and Manufacturing Group (Santa Clara, Calif.).

SI: Why go to 450 mm?

Peter Silverman, Director of Intel's Equipment Technology Strategy, Technology and Manu-facturing Group, Santa Clara, Calif.

SI: Equipment suppliers seem concerned over this move.

Silverman: They're taking a somewhat simplistic view. They argue that if we remain at 300 mm and capacity demand rises, more 300 mm fabs will be built and production will be the same as if we went to 450 mm. This ignores the fact that as a maturing industry, we have a very elastic market. If die costs rise — as they will — growth will decrease. Thus, 450 mm could be better rather than worse for the equipment suppliers.

SI: What about the development burden?

Silverman: Everybody agrees that the major hurdle to 450 mm is developing the toolset and spreading out development costs. Although not an overwhelmingly large amount of money, it is a large investment. The biggest risk, however, is for the same thing to happen at 450 as it did at 300 mm if the economic cycle gods conspire against us and we go into a slump. When the industry was ready to convert from 200 to 300 mm, we went into a big downturn, which made matters difficult for suppliers. Equipment manufacturers worry that this might repeat, and we must discuss ways to share this risk. We must work together toward this transition, and it must be done in a reasonably short time scale. If you look at the cost per square centimeter of silicon, by this decade's end we'll find costs are higher than ever before. We must quickly address this.

SI: There is still learning to be done about 450.

Silverman: Insofar as working with and processing a 450 mm wafer is concerned, that's true. In terms of making the wafers, wafer suppliers are quite confident; 400 mm wafers have been demonstrated.

Things should be significantly easier on the equipment side. The transition from 200 to 300 mm required going from unautomated to fully automated factories. This was a true paradigm shift — wafers going into FOUPs — a big change. Going from 300 to 450 mm is merely scaling. Many current automation schemes will be reusable, particularly software, which was a major hurdle going to 300 mm. From a technical perspective, there's nothing to be concerned about.

SI: Won't 450 mm fabs be prohibitively expensive?

Silverman: A single 450 mm fab will cost less than two 300 mm ones for the same capacity. True, it's an enormous expenditure, but this industry is used to dealing with huge amounts of money. Depending on capacity, $5B or $6B for a 450 mm fab is conceivable, but the key factor is, once the factory is full, what's the cost of the material coming out? If it's lower and the factory is full, then cost is not a huge issue.

SI: Won't the move to 450 mm reduce competitiveness?

Silverman: Some argue that larger, more expensive 450 mm fabs would speed up consolidation, dampening competition and innovation, reducing the synergy that the industry has experienced over the last forty years. This isn't a legitimate argument. Even today, at the leading edge, there aren't that many companies, and that hasn't slowed competition one bit because it is built into the industry's core.

Tom Abell, manager of the 450 mm Transition Program at the International Sematech Manufacturing Initiative (ISMI), Austin, Texas, agrees with this.

SI: What is 450 mm's prime mover?

Tom Abell, manager of the 450 mm Transition Program at the Inter-national Sematech Manufacturing Initiative (ISMI), Austin, Texas

SI: Why not just fine-tune the present infrastructure to maximize yield?

Abell: There is certainly some opportunity with the existing infrastructure for this; however, in some cases, design constraints were fixed by decisions made over a decade ago about how fabs will operate, how the equipment will be designed, etc. What we face today resulted not necessarily from bad decisions, because we didn't really understand how the economic landscape or the business drivers would change. Some of the degrees of freedom in fixing problems aren't easily changed. With the existing infrastructure, it might be possible to get a percentage of improvement. This is what 300 Prime is considering: how much opportunity is present in these fixed constraints. Small changes that might be retrofitted into an existing factory might not get you much; more radical changes may be required to attain significant benefits. Eventually, the question becomes when is it better to simply make a wafer size transition and change the entire infrastructure, instead of attempting to retrofit individual aspects of the current infrastructure?

SI: The engineering to transition to 450 will be rough.

Abell: Nobody has any illusions about the engineering difficulties associated with the transition. There are problems such as those that can originate from a longer wafer edge or from silicon technical challenges; these are things we learned going from 200 to 300 mm. There are engineering challenges; however, they're just that — engineering challenges — and they can be solved.

SI: For a price.

Abell: Certainly, the overriding question is how much it will cost to solve those problems. However, it doesn't appear that much more than good — admittedly expensive — engineering development will be required. Many of the problems, such as that of the wafer's bevel edge, are being grappled with in 300 mm. These must be solved regardless of wafer size.

SI: Many standards are still undefined.

Abell: When we consider what standards might be needed for 450 mm, at least in the early stages, it comes to diameter and thickness. Other technical issues, such as contamination, are much more process- and technology-related — unrelated to diameter. Undeniably, the larger size will exacerbate some problems, but there's no reason why we couldn't apply the same solutions or same approaches to 450 that are being worked on for 300 mm. We've gone through a tremendous learning curve with 300 mm, and considerable modeling has been done to determine the optimum thickness to minimize the amount of bulk silicon required for the wafer, as well as to predict processing stresses. We're using as much computer modeling as possible and see it as feasible; we have workable avenues to pursue.

SI: Any concerns over the silicon supply?

Abell: Some have mentioned that with the upsurge in the photovoltaic market, the basic material might not be as available for 450 as it was for 300 when we transitioned from 200 mm. The polysilicon required for PVs isn't as much of a direct competition for the single-crystal silicon. Production of a wafer that size seems more of a concern.

SI: Will the equipment be available?

Abell: Undeniably, some equipment manufacturers have shown little enthusiasm over 450 mm. They're justified — the last transition was extremely painful. We started 300 mm planning in the mid-1990s, when everything looked rosy. Suppliers invested considerably into R&D with the expectation that the cycle would continue. Meanwhile, chipmakers were more interested in a cost-effective way of increasing capacity than in die cost reduction, because fabs were being built so fast. When the dotcom crash came followed by 9/11, manufacturers were left holding a bundle of R&D dollars and dealing with unhappy shareholders.

SI: Are there any concerns about scaling to 450 nm?

Abell: During the transition to 300 mm, there were concerns about process scaling difficulties. Suppliers used considerable computing power to model, and the alpha tools were surprisingly close to the models. Much was learned. The sophistication level between the first 200 mm equipment I saw back in 1991 and that of some of the early prototype tools was astounding, showing how much suppliers had learned about design methodology. This makes me confident that another such transition won't be insurmountably difficult. Admittedly, there are always some physical challenges — plasma uniformity across that large an area, for example.

SI: Any differences in this transition?

Abell: A key difference in the transition from 200 to 300 mm was that we really didn't have a truly automated material handling system in factory automation. Considerable new development and new technology became necessary in factory automation infrastructure. Now we're building on that, not starting from practically ground zero as we did then. Some changes and updates will be necessary to improve efficiency, but from a factory layout integration perspective, going to 450 should theoretically be simpler than from 200 to 300 mm; just improvements on existing infrastructure.

Jim Kupec, COO of eSilicon (Sunnyvale, Calif.), sees a slow transition to 450 mm.

SI: Will 450 mm be ready for 2012?

Jim Kupec, COO of eSilicon, Sunnyvale, Calif.

SI: You prefer 300 mm technology?

Kupec: It seems more cost-efficient to fine-tune 300 mm processes to increase and maintain yield than going through an expensive transition to a larger wafer size. There will be a considerable 300 mm capacity increase in Asia, and competition and utilization will drive 300 mm wafers prices to more competitive levels. This should take place over the next five or 10 years, and further delay the possible introduction of 450 mm.

SI: Are you concerned over wafer availability?

Kupec: People say that there isn't a fundamental limitation on wafer supplies, regardless of the rise in demand for silicon by other markets such as photovoltaics, for example. However, I see a fundamental limitation in 450 mm fabs. I refer to the toolset cost, and to the equipment developers' willingness to take a bet on 450 mm equipment. They know that 300 mm equipment will continue to sell for a long time, and this isn't an inducement to move to 450 mm. On the chipmaker side, for example, TI has announced that they are going to move their development processes to foundries to minimize their costs. This isn't directly tied to a shift to 450 mm, but to the cost of developing advanced technology nodes on the existing 300 mm wafer size.

SI: Any concerns about scaling?

Kupec: With enough time and energy, these problems are solvable. The question is whether there is enough time, energy and money to make it happen vs. continuing to use the current wafer size. It isn't just a matter of will, but of money. I don't see any insurmountable technical barriers if the industry wants to solve these problems, but I can see people far more willing to spend money on going down the technology path from 65 to 45 to 32 nm than on going to a larger wafer.

We're too concerned about the move to 450 nm. Lest we forget, the industry's workhorse — 300 mm — is here now and doing very well. We should go on making money now and not worry about going to 450 mm…or 675 mm.