Status and trends in 300mm manufacturing
01/01/2009
S.W. Jones, IC Knowledge LLC, Georgetown, MA USA
In the second quarter of 2008, 300mm wafers passed 200mm wafers as the highest production volume wafer size on a sq. in. basis [1]. Announcements of 200mm fab closings by memory producers in favor of lower cost 300mm production seem to come almost daily. Clearly, 300mm has been a success for the semiconductor industry, and yet there are lingering issues particularly for equipment companies. In this paper, some history and current and future trends in 300mm wafers are reviewed.
Many observers were surprised by the initial “slow ramp” of 300mm manufacturing. The first 300mm fab was built by Siemens (before spinning out Infineon) and Motorola (before spinning out Freescale), and then two years passed before the second 300mm fab was operational. The surprising element of this was that anyone was surprised! After all, Intel built the first 150mm fab, and then there was a delay before the industry adopted the new wafer size. Likewise, IBM built the first 200mm fab, and then there was a delay before the rest of the industry adopted the new wafer size. In fact, the ramp up of 300mm has been remarkably similar to the ramp up of 200mm wafers. Figure 1 plots the percentage of worldwide silicon produced versus time for 200mm and 300mm wafers, illustrating that the 300mm ramp is 6???12 months ahead of the 200mm ramp.
Double-edged sword
While 300mm has been successful for semiconductor producers, it has been problematic for the equipment companies that have spent nearly $12B developing 300mm tools, only to see demand for the tools delayed for years beyond what they expected.
 Figure 1. Percentage of total silicon area produced for 200mm and 300mm wafers vs. time. |
The second issue is more systematic and continues to this day. A 300mm equipment set is substantially less expensive than a 200mm equipment set for the same level of die out. For example, Intel announced that they spent $1B less on 300mm capacity in 2004 than the equivalent capacity would have cost by investing in 200mm fabs. Furthermore, Intel reported a savings of $4B on capacity spending since its 300mm program began through the end of 2003 [2]. We estimate that spending on fab equipment for a 300mm plant is ∼60% of the cost of a 200mm plant with equivalent die out. This represents revenue the equipment industry would presumably have seen if the transition to 300mm wafers had never happened, and came at the same time that the semiconductor industry was transitioning from 14%+ annual growth to ∼6% annual growth [3].
Recent events
Recent issues with the global economy are putting a squeeze on businesses across the globe and the semiconductor industry is no exception. During the first six months of 2008, the industry showed strengthening growth month-over-month, vs. 2007, peaking at 12.9% for June month-over-month, and 6.9% year-to-date versus 2007. For July and August, month-over-month and year-to-date growth declined, and what once looked like a year that could end with 8% to 10% overall growth is now looking more like 4% to 5% growth, with most of the growth in the first half, and a flat-to-down second half.
Semiconductor companies have understandably become very cautious about adding capacity. For 300mm fabs, a steady stream of delays has been announced. Micron is delaying IM-Flash S in Singapore and is now planning to install equipment in 2009. An announced Micron 300mm fab in Boise, Idaho has seen no activity for five months. Numonyx NV’s NOR Flash fab in Sicily is on-hold with no plans to restart work. Qimonda has put a DRAM fab in Singapore on hold. TI’s DMOS 7 Fab is on hold and with TI’s fab-lite strategy, may never be built. Toshiba is slowing the capacity ramp for its Fab 4 NAND Flash fab, and the company has pushed out its Fab 5 NAND Flash fab to 2010 from 2009.
A significant number of 300mm fabs in various planning stages or construction have not announced delays to-date. Samsung has seven fabs in construction or planning, ProMOS has three, Powerchip has three, Intel has two, Inotera has two, TSMC has four modules and UMC has one module; and with the transition to Foundry Co., AMD’s New York fab may finally get underway.
 Figure 2. Year-end operating 300mm wafer fabs and wafer capacity by year [4]. |
There is no question that capacity management in 2009 will be challenging. Foundries, in particular, may be very tight for capacity in 2009. We are currently forecasting adequate total capacity in 2009 and barring further push-outs, expect overall utilization in 2009 to be ∼90%. We are also forecasting 84 operating 300mm fabs at the end of 2008, 98 operating 300mm fabs at the end of 2009, and 111 operating 300mm fabs at the end of 2010. Total 300mm wafer capacity is forecast to be 30.20M wafers in 2008, 36.62M wafers in 2009, and 44.36M wafers in 2010 (Fig. 2) [4].
300mm manufacturing vs. 450mm
The debate about 450mm wafers is going full steam. The largest, most successful semiconductor companies in each segment???Intel for microprocessors, Samsung for memory, and TSMC for foundries???are all pushing for 450mm. The smaller, less well-funded semiconductor companies are less enthusiastic, and the equipment companies are generally hostile to the idea.
Our investigations of 450mm suggest that it will provide a cost reduction similar to the 200mm to 300mm transition, and therefore should ultimately prove to be an economical platform for semiconductor production. The key issue revolves around making it a successful wafer size for the equipment companies. SEMI has estimated that $25B will be required to fully develop 450mm wafers and equipment. It is also reasonable to expect one, or possibly a few, initial pilot facilities followed by a long delay before 450mm wafers go mainstream.
In our opinion, the biggest issue is that if 450mm wafers provide the same kind of capital efficiency improvement as 300mm wafers, equipment companies won’t have a viable business model. This area will require intensive cooperation and some painful compromises between the semiconductor producers and the equipment companies.
Conclusion
300mm wafers have become the most widely used and economical platform for semiconductor production, particularly for leading-edge devices. Baring the introduction of 450mm wafers, 300mm will continue to be the wafer-size-of-choice for leading-edge production, and 300mm capacity will continue to increase over the next three years with ∼21%/yr. capacity increases currently forecast.
References
[1] “STATS: SICAS Capacity and Utilization Rates Q2 2008,” Semiconductor Industry Association. http://www.sia-online.org/cs/papers_publications/statistics.
[2] Intel 4th quarter 2003 conference call.
[3] Scotten W. Jones, “2008 IC Economics Report,” IC Knowledge (2008), http://www.icknowledge.com/our_products/economics.html.
[4] “300mm Watch Database,” IC Knowledge (2008), http://www.icknowledge.com/our_products/300mm_watch.html.
Scotten W. Jones received his Bachelors of Science from the U. of Rhode Island and is a senior member of the IEEE, a member of the Electrochemical Society, and is the president of IC Knowledge LLC, P.O. Box 20, Georgetown, MA 01833; ph.: 978-352-7610; email [email protected]; www.icknowledge.com.