INDUSTRY NEWS

Big changes needed to satisfy cleanroom, boardroom, exec says

The chief operating officer of Philips Semiconductors has put his colleagues on notice that the industry must make better use of its capital assets or risk seeing an industry dominated by a few giants able to afford the coming $2-billion-plus fab.

HEX: Philips COO proposes using modular fab, left, to combat rising costs of chip manufacturing.

In an October 6 speech at the International Symposium of Semiconductor Manufacturing in San Francisco, Stuart McIntosh asserted that semiconductor manufacturers can increase capital effectiveness by converting to the use of modular fabs and forging new alliances with tool suppliers. These new partnerships would ensure that complex process gear is "running at optimum efficiency from day one and through succeeding generations of process technology." At the same time, toolmakers would benefit from new business and strengthened bonds with each customer.

McIntosh proposed the use of an integrated system of tool design that would enable fabs to maximize their productivity. A basic seven-module design consists of a central area for distribution and general functions surrounded by process modules. "Instead of building and commissioning the whole $2-billion fab in one go, modules could be built and started as required," the COO told the 700-member audience. Once the first module is built, processing could begin, "even though the whole fab complex may eventually consist of many sets of modules, installed as demand requires."

Even efficient fabs operating their complex and expensive equipment around the clock achieve an overall capital effectiveness of 50%, the 26-year industry veteran pointed out. The reasons for this are varied. They include both planned and unplanned downtime, marketing throughput rates and "real" throughput rates, uneven workflows, and test-wafer runs. And, he noted, today's equipment is "very, very complex." Unfortunately, as costs head for the stratosphere, the industry can no longer accept this level of efficiency.

A $1-billion fab with an annual output of 360,000 8-in. wafers has yearly depreciation costs of $200 million and operating costs of $400 million, adding up to an annual tab of $600 million, McIntosh said. Each wafer needs to bring in $3000 "simply to achieve a 50% margin--gross, net, or manufacturing, take your pick." However, a $1.5-billion fab will have annual depreciation and operating costs of $900 million, requiring a price per wafer of $5000 to achieve that 50% margin. The fab therefore must bring in $1.8 billion annually.

This economic scenario "may remain possible for complex, high-margin processors such as Pentiums, but for other products such as memories . . .it's doubtful," McIntosh noted.

"Until now manufacturing and technology—the 'cleanroom' guys—have been the industry's driving force, and with this sort of success, the guys in the boardroom have been very happy to come along for the ride, even if that ride is sometimes a bumpy one. But the stakes are increasing," he continued.

"In the mid-1980s a fab cost about $100 million and had an expected life span of nearly 10 years, because it started life as a memory fab and then, as memory technology moved on, was converted to logic technology. Now we have billion-dollar fabs where the depreciation alone is $4 million a week." The difference between memory and logic technologies is now "so significant that the previous migration model is hardly possible."

Continuing to run the business this way is no longer feasible, particularly for the equipment market, McIntosh insisted. Toolmakers will need to recover their high development costs "over a relatively small number of tools," because unit sales cannot possibly grow at the same pace as the cost of developing increasingly complex machinery. "I think you will be surprised to realize how few DRAM fabs will be required to meet world demand. I know I was. It's somewhere between 10 and 20."

Envisioning a team effort to improve the efficiency of the "current fab-based system," McIntosh emphasized that he sees the biggest potential for improvement coming from the equipment community. Upgradeable, modular tools and an integrated fab design would steer the industry away from the current setup where equipment vendors try to maximize the effectiveness of individual tools. None of these tools works in isolation, though, the executive pointed out. "We are not actually looking to maximize the productivity of any individual piece of equipment; we should be looking to maximize the whole fab's productivity."

Philips is working with suppliers to develop a business model in which the Netherlands-based chipmaker would buy a special process step from the supplier, rather than purchase the actual tool. As an example, the executive cited a case where "a stepper-and-track combination would sell the required number of good alignments per day."

McIntosh later noted that this concept is an extension of already ongoing cooperation between supplier and manufacturer, but on a much more integrated level. Under this new business model, equipment in the fab is operated and maintained by a partnership between the supplier and the chipmaker.

"This leaves the semiconductor manufacturer free to concentrate on the areas he needs to be good at, bringing systems on silicon to market fast and cost-effectively," McIntosh concluded.

Although some of these ideas are being implemented in the form of foundries and fabless companies, Philips is putting its money where its mouth is, McIntosh indicated. The chipmaker will take "several months . . .to fully implement this new way of interacting with suppliers," starting with part of a fab before expanding the concept.

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit