300 mm Progress Report

Two dozen fabs around the world are now experiencing the benefits of production on 300 mm wafers. The ability to fabricate more than twice as many die on a 300 vs. 200 mm wafer for a relatively small premium in equipment and facility costs has led to significant cost savings of about 30% per wafer. As the semiconductor industry begins to add capacity after an extended downturn, those companies that have made the investment in 300 mm are expected to have a key competitive edge.
 
Looking back, the move to 300 mm has been a long and bumpy road. It was 1995 when the industry agreed that 300 mm should be the next wafer size and that production on these large-diameter wafers should begin in 1998.¹ Equipment suppliers heeded the call and introduced the first generation of equipment in 1996 and 1997. Then came a downturn, which delayed the move to 300 mm production. The industry rebounded in 2000, but so quickly (and so uncertainly) that most companies felt safer adding capacity to 200 mm lines rather than making the riskier move to unproven 300 mm processes. The 2000 rebound turned out to be short-lived, so it is only now with a new upturn in the making that a significant amount of new 300 mm capacity is expected to come on-line.

According to Strategic Marketing Associates (Santa Cruz, Calif.), the 24 300 mm lines now in operation are mostly volume production lines, but some are R&D or pilot lines. Figure 1 shows the capacity of the top seven companies that have 300 mm capabilities, considering the equipment presently in place. However, most of these fabs are not operating anywhere close to full capacity. The research firm notes that, if fully ramped, these fabs could produce as many as 350,000 wafers a month. Converted to 200 mm wafers, that's almost 800,000 wafers or 40 200 mm fabs with a capacity of 20,000 wafers a month. In terms of theoretical full capacity (what the fab is designed to produce when fully ramped), Intel and UMC have more capacity on-line than any other company (Fig. 2). Strategic Marketing Associates also reports that as many as 15 new 300 mm fabs will begin production between now and the second quarter of 2004. Figure 3 shows current, theoretical and planned 300 mm activity.

1. Present capacity of the seven companies that have 300 mm production capabilities. (Source: Strategic Marketing Associates)

2. The capacity of the top seven companies if the fabs were ramped to the level for which they were designed. (Source: Strategic Marketing Associates)

3. 300 mm production capacity as it is today, theoretical capacity of existing fabs, plus the amount of capacity that’s in the early stages of planning or construction. (Source: Strategic Marketing Associates)

Longer term, iSuppli predicts that, by 2010, 20% of semiconductor manufacturing will be done with 300 mm wafers. About 40% of 300 mm fabs will be located in Taiwan by 2005, contributing to the total of 70% to be found in Asia. According to Len Jelinek of iSuppli, initial success in 300 mm manufacturing has been achieved by companies that have a dominant market share, allowing them to sell the increased throughput. Longer term, companies that are unable to transition to leading-edge technology and large wafer sizes cannot remain cost-competitive (see Jelinek's article, "The Promised Land Is Still 300 mm Away").

The industry has already had several experiences in moving to larger wafer sizes, with each new wafer generation taking significantly longer than previous generations. The moves to 100, 125 and 150 mm each took about three years, while the move from 150 to 200 mm took five years, and the move to 300 mm eight years (to reach 100 million square inches of silicon in production). What's different about 300 mm is that the cost of conversion has mostly been borne by the equipment industry, as opposed to a single company as was the case in the past (IBM is said to have borne the cost of the transition to 200 mm and Intel the one to 150 mm).² "Although many leading chipmakers have already invested in fully integrated pilot lines at the 300 mm wafer size, high-volume manufacturing at 300 mm has been pushed out by all but a few," said Ashok Sinha, senior vice president of silicon processing systems at Applied Materials (Santa Clara, Calif.). "Unfortunately, the 300 mm transition had to overlap with other difficult transitions, from aluminum to copper, from FSG/SiN-based dielectrics to low-k, and finally from 0.13 µm technology node to 90 nm node."

The delay in moving to 300 mm has been both good and bad. On the positive side, it has allowed some of the problems of moving to the larger wafer size to be worked out. Those interviewed for this article agreed that the major technical issues associated with the move to 300 mm have been solved. By now, many equipment suppliers have already introduced a second or even third generation of 300 mm equipment. Also, manufacturing efficiencies realized by new automation and data control strategies pioneered for 300 mm processing have been optimized.

On the other hand, 300 mm equipment often has many new components, and reliability may take some time to be worked out. "There are issues with tool reliability," said Wilbert van den Hoek, chief technology officer and executive vice president of integration and advanced development at Novellus. "We've had 10 years of experience with 200 mm; 300 mm tools definitely don't start off where 200 mm tools left off. Tool uptime is the same if not better than the 200 mm tools, but the MTBF (mean time between failures) and MTBI (mean time between incidents) aren't at the same level yet."

Equipment suppliers also complain that they have had to absorb the cost of the transition with little payback. "The 300 mm technology was developed on the backs of the equipment industry," said van den Hoek. "We had to bring tools to market early on and then there was another five years before there was any revenue. If you look at it from the equipment industry perspective, it cost us billions of dollars and we are still fairly far away from recouping that money. Financially, the 300 mm transition is not something we've been very excited about over the last seven years."

Of course, the story would be quite different if it weren't for the long downturn and continued pressure on equipment pricing. "Our experience has been that equipment cost has been one of the most positive factors with regard to the movement toward 300 mm," said TI's Tolson. "Equipment suppliers have been right on target with regards to their tool pricing increases for our plans on the next-generation 300 mm tools."

It should be noted that there's been some significant progress in the 200 mm world, which has in some ways helped delay the need to move to 300 mm. Ironically, some of the manufacturing efficiency developed for 300 mm production has been implemented in 200 mm lines. "We are getting some manufacturing method efficiencies in 200 mm and extendibility that would not have happened were it not for the driving force of 300 mm," noted Steve Fulton of International SEMATECH (Austin, Texas). "Those advantages that are in the industry would not have happened if it were not for the 300 mm infrastructure and factory implementation."

What a 300 mm factory has that in part enables higher efficiencies is a higher degree of automation, including pervasive use of front-opening unified pods (FOUPs) and mini-environments; very high levels of individual tool automation; a highly integrated factory-level computer integrated manufacturing (CIM) system; automated materials handling systems; factory layout optimized for automation, not operators; and a significantly reduced number of operators.³

Some worry that the as yet untapped reservoir of 300 mm capacity is a disaster waiting to happen. The concern is that, when these fabs do come on-line, there will be tremendous excess capacity, which will lead to price and earnings erosion and make the inevitable next downturn that much more severe. While nobody is predicting that semiconductor companies will blindly add capacity when there's no demand for it, the sheer amount of potential new capacity — the equivalent of 40 20,000 wafers/month 200 mm fabs, plus the 15 new 300 mm fabs in the planning stages — is daunting.

A second concern is that the industry is being divided into two groups: those that have 300 mm and those that don't. When the inevitable next upturn comes, the thinking goes, those that do have 300 mm capabilities will have such a tremendous competitive edge because of lower production costs and their ability to turn capacity on so quickly that less prepared competitors will rapidly lose market share and be eliminated. This is a valid concern, but mainly for companies making commodity products such as DRAMs.

Adding to the complexity of the situation is the role of new technologies. "The industry is running into fundamental limits, which is leading us to make bigger changes every technology node," van den Hoek noted. "We're introducing more new materials and making much larger changes."

Early adopters of 300 mm such as SC300 had the small luxury of being able to work at 0.25 µm geometries. Today, the stakes have been raised and companies have to not only struggle with the transition to a larger wafer, but introduce a slew of new technologies on those larger wafers, including smaller dimensions (90 nm), copper dual-damascene processing, low- and high-k dielectrics, and perhaps wafer bumping and backgrinding.

Presently, the model at most semiconductor companies is to develop new technologies on 200 mm equipment, then transfer that technology to the 300 mm production line. This is because 300 mm wafers cost much more than 200 mm wafers, and 300 mm capital equipment and operational costs can also be greater. "Currently, we believe that it's more cost-effective and doesn't deter our ability to convert those technologies to still develop on 200 mm," said TI's Tolson. "Wafer price alone — our research and development eats up literally thousands of those a month — is five to six, even 10 times the cost of 200 mm wafers. There are a lot of things in the maturity of 300 mm manufacturing that have to occur to enable development to be done cost-competitively on 300 mm. Downside: you do have some transfer issues."

The potential problem with this model is that equipment suppliers may no longer wish or be able to support the development of new technologies on 200 mm platforms. "If the only reason to have a 200 mm version is to enable the customers to do 200 mm development, the financial return on this 200 mm version is not going to be there for the suppliers," van den Hoek said. "On the other hand, the customer's risk of inserting a new technology at 300 mm production, without having been able to introduce this technology in their 200 mm development, is very large. This is even more of a problem now, because so many changes are introduced every new technology node. I think this question is going to become a bigger issue and needs to be answered in the industry."

Speaking of the economic model at International SEMATECH, Denis Fandel added, "We anticipate that there will be some sort of eventual phasing out of 200 mm technology capability, only because of the limitations of suppliers and the ability to fund development on both wafer sizes at the same time."

Although it may not work for all types of equipment, one solution might be the approach taken by Lam Research (Fremont, Calif.). "We made a strategic decision starting with our etch products in 1998 that our new product development was all going to be wafer size independent," noted Dave Hemker of Lam's new product development group. "We kept the footprint the same as the 200 mm tool it was replacing and made it 300 mm capable. We know 300 mm is inevitable for everyone at some point — or for the vast majority — but we're not concerned as to whether it is happening now or tomorrow."

The good news is that new technology development on 300 mm may be soon here, at least for DRAMs. "It appears that a high-volume ramp will take place at the 90 nm node for logic and 110 nm for DRAMs," Applied's Sinha said. "Nearly all advanced development for 65 nm and beyond is now being first done on 300 mm rather than 200 mm. So, any day now, we should see the cost benefits of 300 mm, when demand for leading-edge designs outruns capacity."