The 300-mm Age Has Finally Arrived

But What about Its Long-Term Impact?

Bijan Moslehi, PhD, is chief technology officer and senior vice president, semiconductor technology research, for The Noblemen Group (www.noblemengroup.com), a boutique investment banking, strategic advisory, and business development firm. Moslehi has some 20 years' experience working in the semiconductor and semiconductor equipment industries. He can be reached at bmoslehi@noblemengroup.com.

After nearly a decade of intense industrywide effort and billions of dollars of investment, the age of 300-mm manufacturing has finally arrived. In 2003, according to SEMI's Silicon Manufacturers Group, 300-mm wafers accounted for a bit more than 8% of total wafer-area shipments, which is expected to grow to 13% this year and to exceed 20% by 2006. Close to 30 production fabs and R&D pilot lines are in operation, accounting for around 5% of the reported weekly wafer starts. Almost as many new lines are in various phases of implementation, from planning to construction to early start-up, and most could be expanded to add substantial extra capacity. As many as 70% of these plants are or will be located in Asia, with the remainder built in the United States and Europe.

The move to 300-mm wafers has proven to be the toughest, most challenging, and most expensive wafer-size change and manufacturing transition that the semiconductor industry has ever experienced. In the early to mid-1990s, the initial plans called for an early implementation at the 250-nm node by 1998, which did not materialize. This false start was quite premature, and the industry was not ready for it. The 1998 downturn, followed by the short-lived boom of 2000 and the protracted bust of subsequent years, significantly pushed back these plans.

Several difficult technology transitions have coincided with the move to 300-mm technology, including the adoption of copper/low-k interconnect and the ramp of 130-nm and smaller design rules using subwavelength lithography. The implementation of 130 nm and low-k integration have reportedly been costly and painful, with many chipmakers suffering from multiple failures and particularly an inability to hit their target yields. All these factors have contributed to delays in the transition to the larger wafer size and, as a result, many tool suppliers still have not achieved a reasonable return on their 300-mm investments made over the past decade.

The economics of 300-mm technology have come under renewed scrutiny and close reexamination. One key metric is the notion of the cost-crossover point, where for a given product and process technology, the cost per die (or area-weighted wafer cost) of 300-mm wafers would be equal to or less than that of 200-mm substrates. For a given fab, product mix, and process technology, many factors influence the fab cost structure and the resultant cost-crossover point and its timing. In certain cases, some elements of the cost structure may be unique to a particular fab; for specific products, that may be particularly true for some 200-mm plants.

The depreciation costs of equipment, fab buildings, and facilities account for the largest portion of the die or wafer cost and can reach as high as half of the total. The cost of wafers, consumables, and other process materials as well as equipment and facilities maintenance costs constitute most of the remaining part. Direct labor only accounts for <5% of the cost in today's modern, automated fabs. Furthermore, the installed fab capacity, fab capacity utilization, cycle time, and yield directly affect the cost of finished product.

While larger fabs are more economical to run, they are also harder to fill, and are more susceptible to idle capacity and low utilization during a down cycle. Fab location, together with local tax breaks and incentives, also plays a role. Therefore, a more thorough economic analysis must be performed for at least a full industry cycle, if not for the useful life of the fab.

At least in the short term, the early demise of 200 mm may have been somewhat exaggerated. There has been a notable level of capacity orders recently for 200-mm process tools. The used-equipment market has also been active, particularly in China. Furthermore, the 200-mm lines could directly benefit from progress made in the productivity enhancement and manufacturing efficiencies of 300-mm lines.

Despite a significant drop in wafer prices, there is also some concern that, in terms of per-unit area, the 300-mm wafers are still more expensive than the smaller substrates. However, as the number of 300-mm fabs has grown, this price gap has progressively narrowed.

The extended useful life of process tools and the huge supply of inexpensive used equipment, combined with partially or substantially depreciated fabs, can create economically favorable conditions for selected 200-mm fabs. Some suppliers have reported that certain chipmakers are rethinking their 300-mm plans and are evaluating the economics of extending the life of their existing lines on a case-by-case basis. This appears to be a temporary situation, an apparent advantage that will be gradually eroded as 65-nm and smaller process technologies are introduced and 300-mm fabs continue to evolve. In a related development to watch, two fabs are being upgraded from 200 to 300 mm. The current situation creates yet another challenge for tool suppliers, who need to continue developing both 200- and 300-mm equipment for advanced processes. Many suppliers have addressed this issue by providing upgradeable bridge tools or wafer-size-independent systems.

Several DRAM and microprocessor manufacturers, as well as a few fabless customers of foundries say that they have met both the yield and the cost-crossover targets, claiming their 300-mm production has become more cost- effective than the 200-mm runs. A better capital investment efficiency has also been cited as a positive result of the move to the larger wafers. However, the cost of technology development on 300-mm wafers has been a major concern. Some chipmakers have used 200-mm wafers for process development, with a subsequent transfer to 300 mm, although this has started to change at 65 nm. The factors cited above underscore why there is so much confusion about the economics of 300-mm fabs; in general, each manufacturer and each fab experiences varying degrees of success as it goes through its own learning curve.

While filling a high-volume 300-mm fab is certainly a challenge, maintaining high utilization throughout various stages of the industry cycles is even more daunting. This could be ameliorated by dedicating these fabs to leading-edge process technologies, which traditionally have enjoyed a higher utilization rate, even in a down cycle. But this is a partial solution, since advanced technologies typically account for <10% of the total chips produced. In addition, the demand for these advanced products has recently been weaker than expected, a disturbing development that indicates that performance may be getting ahead of what certain market segments need and are willing to pay for.

DRAM manufacturers have been the most aggressive in adopting 300-mm technology, followed by major foundries and microprocessor suppliers. Most DRAMs and MPUs are eventually expected to be made on 300-mm lines. Typically, those devices account for about one-third of annual chip production, and foundries process about one-sixth of the total, a share that will continue to grow.

Most 300-mm fabs generally run a low mix of high-volume standard parts using sub-100-nm process technologies. A high product mix, even in a high- volume fab, would present many unique economic and operational challenges to chip suppliers and foundries. This means that many smaller companies and low-volume fabs with a high product mix will find it very difficult to transition to the larger wafers, especially since at least $1 billion in annual revenues is needed to justify a high-volume 300-mm fab.

As a result, the role of foundries and other outsourcing strategies will be more important than ever. There will also be new opportunities for tool and automation suppliers to provide solutions for the economical processing of high-mix/low-volume products in 300-mm fabs. Such solutions include fabwide single-wafer processing combined with efficient, economical processing of single-wafer lots, which will require further advances in e-manufacturing, advanced process control, and fab automation. Single-wafer processing in cost-effective minifabs that could economically produce wafers at low volumes would provide the ultimate desired solution. Recently, a minifab concept, which uses multifunction tools and reduces the total number of tools, was successfully demonstrated by Japan's Highly Agile Line Concept Advancement (HALCA) project.

Although the majority of new fabs under consideration are 300-mm lines and most new-equipment orders are also for the larger wafer size, it is clear that 300 mm may not be the right solution for all fabs or all products. Particularly in the short term, many advanced 200-mm fabs should have an extended and significant useful life. It boils down to one key question the entire industry wants answered: What will the longer-term impact of the transition to 300-mm manufacturing be on the equipment market and semiconductor economic cycles?