Sometimes transferring knowledge means transferring an exact copy, especially when you don't really know why things work the way they do.
This is a big problem for Intel - chip manufacture is a complex process, the full details of interactions within the system are not known, and errors need to be avoided. It is essential that technology transfer to mass production takes place as quickly as possible, without disruptive quality issues, and with the highest possible speed. No time is available to debug new problems that occur during the transfer.
Intel came up with a solution that at first sight is counter intuitive - a solution designed to avoid making improvements - Copying Exactly..
You know the old phrase "if it works, don't mess with it"? What Intel found was that if it's really complex and works, then seriously don't mess with it; not even to the tiniest amount.
According to the article "the evolution of Intel's copy exactly technology transfer method" by Chris McDonald, Intel historically tried many different strategies to transfer production knowledge.
For the 1.5 micron chip, work process flows were far simpler than they are today, Intel used a small band of technical experts to orchestrate successful technology transfer, often introducing improvements to the equipment process on the way. Because of the small number of variables, it was simple to troubleshoot any results that did not come out as expected.
For the 1.0 micron chip, started get more complicated and a structured method was introduced, where changes could still be introduced. Figure one shows how the first production factory, Fab 1, delivered yields that were similar to the development plant, but the second and third factories, which made changes to the process and equipment, gave much poorer yields, and it took years to catch up. The same learning curve was repeated independently by every factory.
For the sub micron chips, Intel introduced their "copy exactly" principle for technology transfer. In its simplest form this says that "everything that might affect the process, or how it is to be run, is to be copied down to the finest detail unless it is either physically impossible to do so". This is a very different philosophy, and quite difficult for engineers to follow, because engineers are typically trained and rewarded for making improvements. However Intel found that almost every "improvement" led to a decrease in yield. As a result , they ensure rigid copying; the same pumps, the same length of hoses, the same raw materials, the same manufacturers and so on. You can see the results in figures three and figure four, where the first and second fabricator plants were able to match the development plant exactly.
Here's what the author of the article concludes
"The Copy EXACTLY! method has proven itself as a technique for semiconductor technology transfer. A new process flow and products can be introduced to production in minimum time with equivalent yields and without the introduction of product-quality issues. Both manufacturer and customers can reduce their time to market. This approach could equally be employed in other industries where the technology is complex and has many interacting variables affecting the end result"
"Repatriating a high-tech manufacturing plant to the United States is not simply a matter of hiring the local talent. It requires good-old foreign know-how. “We call it ‘copy exact,’ ” Forcier said. “We bought a company in Korea that had the technology around this type of battery and had developed the manufacturing process there. We basically brought that here, copied it exactly and scaled it up.”
They also brought a team of six Korean engineers to help transfer the technology to the U.S. and sent a team of Americans to Korea to learn.
In the context of high technical complexity, "Copy Exactly" may be the best way to transfer success.