Fab Allocation Checklist: What One Chip Team Weighed Before Signing a Three-Year Capacity Deal

Archi4 min read

semiconductorssupply-chainprocurement

Contents
  1. The decision on the table
  2. 1. Map real demand, not the sales forecast
  3. 2. Separate the wafer from the die
  4. 3. Price the flexibility you give up
  5. 4. Check the foundry's own constraints
  6. 5. Build the exit math before the entrance
  7. 6. Ask what the customer will actually pay
  8. What the team decided
  9. A short version for teams facing the same call

The decision on the table#

A fabless chipmaker with about 60 employees had to answer a single question before the quarter closed: sign a three-year capacity reservation at a foundry, or stay on the spot market and hope. The company shipped power management parts for industrial customers. Its volumes were steady but modest, roughly 4,000 wafers a year at 180nm and a smaller run at 90nm.

The foundry offered guaranteed allocation in exchange for a take-or-pay floor. Miss the floor and the company still pays. Hit it and the per-wafer price drops about 8 percent. The catch was a deposit equal to two quarters of forecast spend, money that would otherwise fund a second product line.

This is the checklist the operations lead built to make the call. It is worth reading because the same tradeoffs show up at companies far larger, and the failure modes are remarkably consistent.

1. Map real demand, not the sales forecast#

The sales team projected 30 percent growth. The operations lead pulled eighteen months of actual shipments and found growth closer to 11 percent, with two large customers accounting for most of it. She also checked their purchase orders against their own public guidance. One customer had been double-ordering during the shortage and was now sitting on inventory.

Action: build the capacity case on the bottom 80 percent of demand, then model the top 20 percent as upside. A take-or-pay floor should sit near the conservative case, never the optimistic one.

2. Separate the wafer from the die#

Wafer counts are a poor unit for planning when die sizes change. The team had a die shrink planned for year two that would raise die per wafer by roughly 40 percent. Committing to a wafer floor at the old die size would have locked in more capacity than the product needed.

Action: convert every forecast into finished units, then back into wafers at each process node. Recheck the floor against the post-shrink die size before signing.

3. Price the flexibility you give up#

Spot pricing during the shortage ran two to three times the contract rate. That made the contract look cheap. But the deposit had a real cost: at a plausible internal hurdle rate, tying up that cash for two years was the equivalent of several percent on every wafer.

Action: score the deal on total cost, including the cost of capital on the deposit and the opportunity cost of the product line that money would have funded.

4. Check the foundry's own constraints#

Guaranteed wafers still depend on substrate, assembly and test capacity downstream. The team asked the foundry which packaging houses it used and whether those had their own backlogs. Two of the three were quoting lead times longer than the fab itself.

Action: trace the full path from wafer to tested part. A capacity contract that ignores assembly and test is only half a guarantee.

5. Build the exit math before the entrance#

The contract allowed renegotiation only after eighteen months. The team modeled what happens if demand falls 30 percent: the company would pay for unused wafers while cutting staff elsewhere. It also modeled the reverse, a demand spike with no contract, and found the spot market would have cost more than the deposit.

Action: write both downside scenarios on one page. If the downside is survivable and the upside is not, the contract earns its place.

6. Ask what the customer will actually pay#

Industrial buyers sign multi-year supply agreements and dislike requalifying parts. The team surveyed five key accounts and found most would accept a modest price increase for guaranteed supply. That changed the math: part of the contract premium could be passed through.

Action: test price tolerance with real customers before treating the contract cost as a fixed loss.

What the team decided#

The company signed, but not the original offer. It negotiated the floor down to the conservative demand case, moved the deposit to a smaller upfront sum with quarterly top-ups, and added a clause allowing wafer volumes to convert to a newer node once the die shrink qualified. The foundry accepted most of it because the company was asking for less capacity, not a lower price.

Eighteen months later, demand came in near the conservative case. The company paid for a small amount of unused capacity in one quarter. It also avoided the spot market entirely during a second shortage and kept every industrial account.

A short version for teams facing the same call#

, Build the floor from verified shipments, not sales targets., Plan in finished units and convert to wafers at each node., Include the cost of the deposit in the per-wafer price., Confirm assembly and test capacity match the fab promise., Write the downside on one page before signing., Find out which customers will pay for guaranteed supply.

The pattern generalizes. Capacity contracts are not bets on growth. They are bets on your own ability to forecast honestly and on a foundry's willingness to share the risk it is asking you to carry alone.

More articles