A Fab's Christmas Eve Call: Ship the Chip or Blow the Deadline
Archi4 min read
Contents
- The Lot That Had to Choose
- Two Orders, Two Meanings
- What the Contract Said, and What It Did Not
- The Options on the Table
- Option One: Split the Lot
- Option Two: Ship the Automotive Part
- Option Three: Ship the Power IC
- The Decision
- The Cost, in Plain Numbers
- Why This Choice Keeps Coming Back
- What the German Supplier Did Next
The Lot That Had to Choose#
On December 23, a foundry in Taiwan's Hsinchu Science Park faced a scheduling decision that would not make the trade press. One 300mm wafer lot, 25 wafers, sat at the end of a diffusion furnace. The lot carried two customers' designs: an automotive microcontroller on a 40nm process, and a power management IC on the same node. The fab could finish one lot before the holiday shutdown, not both. Someone had to wait until January 4.
The decision went to the operations director, a 22-year veteran of the plant. She had four hours before the furnace slot expired and the tool cooled. The answer she gave is the kind of choice the chip industry makes every day and almost never discusses: which customer absorbs the delay.
Two Orders, Two Meanings#
The automotive customer was a Tier 2 supplier in Germany. Its microcontroller goes into a seat control module. A 12-day slip would push a production run at a European assembly plant, and that plant had already burned through its safety stock after a separate shortage last spring.
The other customer was a power management IC maker in California. Its part goes into a consumer charger, a product with a 9-month life cycle and a launch date in early March. The chip had one more mask layer to go and a revenue commitment tied to a retail window.
Neither order was late. The fab had promised both by the same week. The conflict existed because one tool had gone down for 36 hours in November and the recovery plan had quietly stacked two lots onto the same slot.
What the Contract Said, and What It Did Not#
The fab's master agreement gives allocation priority by historical volume, not by end market. The German supplier had been a customer for nine years and accounted for about 6 percent of the plant's 40nm output. The California company had been there 14 months and accounted for 11 percent, but on a different tape-out schedule that made its lot easier to move.
The Options on the Table#
By mid-morning the operations team had three paths.
Option One: Split the Lot#
The fab could run 12 wafers for the automotive part and 13 for the power IC. This was technically possible and would have been the obvious answer to someone who has never watched a fab's queue. It also meant two partial lots, two qualification rechecks, and a yield risk on both. The process engineer put the yield penalty at 2 to 4 percent on each. On a mature node with thin margins, that is real money.
Option Two: Ship the Automotive Part#
The German customer's order was larger and older. Running it first would honor the relationship and keep a quiet, steady account. The California company would get a written apology and a new date. The account manager warned that the company had already been told once that its launch schedule could not slip further.
Option Three: Ship the Power IC#
The California part had the nearer deadline and the higher per-wafer revenue. Finishing it would protect the fab's newest customer and a contract that had been hard to win. The German supplier, by this logic, had deeper inventory and could absorb 12 days more easily than a launch could.
The Decision#
At 4:10 p.m., the operations director chose option three. The power management IC finished the furnace, went to test, and shipped on December 28. The automotive lot sat. The German supplier's procurement team was notified the same afternoon by phone, not email, and given three things: a new ship date of January 6, a credit on the next lot, and a direct line to the fab's planning lead.
The reasoning, as she described it afterward, was not about revenue. It was about which delay could be undone. The automotive module had an inventory buffer of about five weeks at the customer's plant. The consumer charger had a buffer of nine days. One slip was expensive. The other was a missed launch, and a missed launch does not come back.
The Cost, in Plain Numbers#
The power IC lot shipped 3 days late against the original promise. The automotive lot shipped 14 days late. The fab paid a penalty credit on the second order and lost about 2 percent of its quarterly 40nm revenue to the scheduling shift and the re-qualification work. No one was fired. The operations director's year-end review included one line: "Prioritized launch-critical lot over volume account; customer retained."
Why This Choice Keeps Coming Back#
The foundry business runs on promises made months before the silicon exists, and the promises collide when a tool fails. Most of those collisions are invisible. A customer notices only when a date moves.
The decision here followed a rule that planners at several fabs describe in similar terms: protect the deadline that cannot be renegotiated, then repair the relationship that can. That rule is easy to state and hard to apply when the older customer is the one that pays the bills. The 40nm node is not glamorous. It is where a lot of cars, chargers and industrial controls actually live, and it is where scheduling fights have the sharpest edges.
What the German Supplier Did Next#
The Tier 2 supplier did not walk. Its procurement lead asked for one change to the contract: a standing allocation review every quarter, with the fab's tool health data included. The fab agreed. That clause now sits in three other customer agreements, added quietly in January. No press release, no conference talk. Just a line in a document that will matter the next time a furnace has to choose.
More articles