The Bends in the Pipes
Intel copies a working fab down to pipe diameters and the number of elbow bends. Not superstition — policy, adopted after a cooling hose decided the yield.
When Intel builds a new semiconductor plant, it copies the old one down to pipe diameters and the number of elbow bends.
01One of them would not reach yield
Before the policy existed, a new fab took nearly a year to reach the yields of the plant it had been copied from. Same procedure, same tooling, people trained the same way.
The gap was real, it was expensive, and nobody could point at what was causing it. Everything anyone could name had already been matched.
A year of a new plant running below the one it was modelled on is not a rounding error in semiconductors. It is the difference between a fab that pays for itself on schedule and one that does not.
02The difference was a hose
Intel adopted Copy EXACTLY! after discovering that minor physical differences — the length of an electrode cooling hose among them — could cause catastrophic drops in yield when a process was transferred to a new site.
So the rule became absolute. Everything that might affect the process gets copied down to the finest detail, unless it is either physically impossible.
A cooling hose. Its length. Recorded nowhere.
One of the most heavily instrumented manufacturers on earth studied its own technology transfers and found that the variables governing yield were living in equipment layout and operator habit, not in the documentation. The written procedure was never wrong. It was incomplete, and nobody could see where.
That is the uncomfortable part. Nobody had been careless. The document said everything its authors knew to write down, and the thing that mattered was not among it — not because anyone missed it, but because nobody had ever had cause to think of a hose length as a process variable.
03Copy the reality, not the description of it
After Copy EXACTLY!, new fabs routinely matched their source plant's yields from the very first production run. Intel had stopped trusting the document and started cloning the thing itself.
The same story plays out at a smaller scale in every startup after a major overhaul. Same standard operating procedures, same pre-commissioning checklists, and the unit still refuses to settle until an experienced operator tweaks a valve or adjusts an offset that appears nowhere in the manual. They have carried that adjustment as a personal habit for a decade.
The adjustment is real, it works, and it exists in exactly one place: the hands of whoever learned it. When that person moves shift, retires, or is simply off on the day it matters, the unit takes longer to settle and nobody records why.
Intel spent billions cloning entire facilities to get around this. Most operations cannot. What is left is the cheaper half of the same idea: treat those habits as process data, and capture them before they leave the shift.
Next startup, stand next to your best operator. Write down the one adjustment they make that is not in the procedure. That is your cooling hose, and right now it exists in one person's hands.
Sources: Chris J. McDonald, "The Evolution of Intel's Copy EXACTLY! Technology Transfer Method", Intel Technology Journal, Q4 1998.
Two plants built from one document — same procedure, same checklists, same equipment list. One takes nearly a year to reach the other's yield, and everything anyone can name has been matched. The difference is a cooling hose. Its length. It was never written down anywhere. Intel stopped trusting the document and cloned the building. The habit still sits outside the copy.
Get the next piece.
Get the free field guide — The Handover Gap: three disasters, one lesson, and a checklist for your next shift. Plus the occasional piece worth your shift. No noise, no pitch.
Run a shift somewhere? I compare notes with operators and supervisors on how real floors handle handover and capture — what works, what quietly doesn't. No pitch; I'm not selling anything. Tell me how yours does it: [email protected]