EN
Back to the archive

The encyclopedia · Engineering & Operations · Technical decision · 1993–1998

Intel made each new fab copy the R&D line exactly, so yields matched from day one

Instead of letting factories improve the process during transfer, Intel froze recipes, pipe bends and software, so new fabs matched R&D yields in record time.

Intel Corporation

the move

Semiconductor processes have hundreds of interacting steps, so transferring a process to a new fab while 'improving' it along the way created delays: Intel's own paper records a 0.8-micron startup delayed three months and about a year before yields matched the R&D line.

For the 0.5-micron generation Intel developed Copy EXACTLY!: everything that might affect the process or how it is run is copied down to the finest detail—pipe diameters, board revisions, software, even preventive maintenance practice—unless copying is physically impossible or there is an overwhelming competitive benefit to change.

A system of four-level matching (physical inputs, process outputs, module structures, product parameters), change-control boards, equipment difference forms, audits and joint specifications made the philosophy enforceable. Two new 0.5-micron fabs came on line with the same yields as the R&D line, in record time, with no quality delays.

why it works

  • Traditional transfers allowed 'improvements' that turned into thousands of interacting unknowns.
  • Copying exactly made troubleshooting trivial: a subtle recipe typo was found within a week.
  • R&D and production improved together, so learning curves were shared instead of repeated.
  • The capital cost of a fab made failed ramps too expensive to risk.
the payoffCopy everything now, improve later, togetherclever

what transfers

When a process is too complex to troubleshoot variable by variable, freeze the variables: transfer by copying, and let improvement happen after the baseline is identical everywhere.

what came after

Copy EXACTLY! became part of Intel's manufacturing strategy through the 0.25-micron generation and beyond, extended into 'Systems Synergy' where fabs are near-identical across sites, and is still cited in semiconductor manufacturing literature.

references

spotted an error? The archive wants to know.

same kind of clever