GREENVILLE, S.C. — New semiconductor manufacturing facilities could be built at 20 percent below current costs–turning out devices six times faster and producing 25-36 percent more revenue than existing fabs–according to the assertions of a just-completed study by Fluor Daniel, an architecture and engineering firm with expertise in building wafer fabs.

The six-month “Strategic Future Fab Study” (SFFS) project consulted 22 leading companies in the semiconductor industry, many of them semiconductor equipment and materials vendors. It was conducted for Fluor Daniel by Paul Castrucci & Associates of South Burlington, Vt. Dr. Castrucci, a veteran of semiconductor manufacturing at IBM, was once COO at Sematech.
Malcolm Williams, director of engineering for Fluor Daniel’s Electronics Division here, said his company projects 75 to 100 new fabs will be built worldwide by the turn of the century. “We wanted to understand the best way to meet this demand, and whether or not that was reflected in today’s conventional wisdom. We asked for a view of the future in semiconductor manufacturing from all the people who play a key role: fab equipment manufacturers, suppliers of chemicals and gases, experts in computer-integrated manufacturing, simulation experts, architects and potential clients for new fabricators.”

The assignment was to design a “standard” logic fab, designed to process 0.35-micron device geometries on 200-millimeter wafers with a four-layer metal CMOS process, a minimum of 25 part numbers and 500 wafer starts per day. Sixty percent of the wafer starts would be devoted to ASICs, with the remainder being MCUs/MPUs and static RAMs.
The study keyed in on 16-megabit SRAMs, “P7″ MPUs with 25 million transistors and standard cell/gate array logic with 800,000 gates per device. Four alternative future fab designs emerged, including a design for a conventional class 1 “ballroom” with stand-alone equipment. These were then evaluated with a variety of parameters.

“Mini-environments emerged in a new role as the key to the flexibility required to maximize revenue,” said Dr. Castrucci. “Our simulations of manufacturing systems show that integrated process tools, when combined with mini-environments and ’smart tag’ logistics systems, buy maximum flexibility and minimum production cycles of six days. That translates into greater profitability–from 25 to 36 percent more revenue per time period. Over the course of four years, it means an additional $987 million when compared to the conventional class 1 ballroom with stand-alone tooling.”

Specifics of the “minifab” include class 10,000 work spaces surrounding class 1 integrated cluster tools, with each cluster capable of up to eight process steps. Without “space suit” garments for fab workers, garments cost savings were $1.5 million a year, and power savings compared with ballroom construction were $1 million a year. Use of mini-environments, which keep wafers clean by enclosing them in ultraclean containers as they are transferred among processing equipment, reduces “several months” in the start of production, the study says.

A 34,500-sq.-ft. cleanroom not only offers sixday cycle times, but also shortens yield learning curves to one year, according to the Fluor Daniel study. By shortening the new-product time-to-market, the breakeven point is accelerated. The study also found that the most costefficient approach was to replicate the building as demand increased in multiples of 500 wafer starts per day.

The mini-fab’s production cycle of six or seven days contrasts with the 60 to 90 days seen today. In today’s conventional class 1 ballroom fabs with stand-alone equipment, wafers spend about 70 percent of their time on line waiting to get into the next tool, the study says.

With shorter cycle times and an advanced yield management system, yield targets were achieved in one year, rather than two. Yield bonuses could substantially increase profitability in the early years of a new product line.

In the most advanced concept presented by the study, Fluor Daniel foresees a production control center capable of presenting a 3D simulation of a “virtual factory,” enabling managing current production control and logistics for multiple part numbers and volumes, simulating future production and training production personnel.

The flexibility of the facility calls for allowing the first tool clusters installed to turn out product while space in the same room is being tooled up for capacity. With most fabs manufacturing four generations of product at any given time, mini-environments allow technology changes to take place without disrupting the entire facility.

“These fabs of the future can be built right now for 1996-1997 product requirements,” said Robert Young, VP of Fluor Daniel’s Electronics Division. “They can be on-line in 18 to 19 months, compared with the 22- to 36-months cycle that is now common.”

Jim Griffin, VP of micro-electronics at Integrated Circuit Engineering Corp. (ICE), one of the participants in the study, said the proposed fab of the future takes “a fresh approach” but “Like any model, it hasn’t been proven.”