The Fab That Chose One Customer: A Timeline of a Foundry Bet

Archi4 min read

semiconductorsmanufacturingsupply-chain

Contents
  1. A contract offer with a deadline attached
  2. Q1 2021: two offers, one clean math problem
  3. Q2 2021: the capacity crunch changes the frame
  4. Q3 2021: a split decision
  5. Q4 2021 to mid 2022: the qualification grind
  6. Late 2022: the second customer arrives
  7. What the decision actually turned on

A contract offer with a deadline attached#

In early 2021, a mid-sized foundry in Taiwan's Hsinchu science park had a capacity problem most rivals would envy: more demand than wafers. A consumer electronics brand wanted to lock two of the foundry's older 28-nanometer lines for a single wearable chip family, three years, take-or-pay. The phone call that mattered came a week later, from a fabless startup designing automotive radar sensors on a newer node the foundry had not yet fully qualified.

This is a timeline of how that choice played out, reconstructed from public filings, trade press reports, and interviews with engineers close to the decision. Names of the specific companies are withheld; the pattern is common across the industry.

Q1 2021: two offers, one clean math problem#

The wearable contract was the easy number. Roughly 40,000 wafers per quarter at a fixed price, with penalties if the brand walked. Utilization of the 28nm lines was already at 95 percent, so the deal would mostly reallocate existing capacity rather than add revenue. Finance liked the predictability. Operations liked that the process was mature and the yield curve flat.

The radar chip was harder to price. Automotive qualification takes 18 to 24 months and involves failure-rate targets measured in parts per billion, plus a temperature range that the foundry's existing 28nm flow only partly covered. The startup wanted a guaranteed allocation at a lower per-wafer price in exchange for a longer commitment, five years, and a co-development agreement on the process.

Q2 2021: the capacity crunch changes the frame#

By spring, a global chip shortage had pushed lead times on mature nodes past 40 weeks. The wearable brand came back with a sweeter offer: it would pay an upfront capacity reservation fee equal to about a quarter of the line's annual revenue. The foundry's board asked for a decision within 30 days.

Engineers pushed back. Their argument, summarized in an internal memo that later surfaced in trade press coverage, was that the radar work would produce a qualified automotive process the foundry could sell to a dozen customers, not one. The wearable contract produced revenue but no new capability. It also carried a risk: if the brand's product cycle turned, the foundry would be stuck with idle lines and a penalty clause written to protect the buyer.

Q3 2021: a split decision#

The foundry took neither offer whole. It signed the wearable brand for 12 months instead of three years, at a higher per-wafer price, and told the startup it would fund the automotive qualification internally, keeping the process proprietary. That meant spending an estimated $180 million on test vehicles, reliability chambers, and a dedicated line, money the foundry had not budgeted for the year.

Two board members resigned over the call, according to local business press. The chief executive's public line was measured: the company would serve both markets, with automotive as a longer-term position.

Q4 2021 to mid 2022: the qualification grind#

Automotive qualification ran into trouble almost immediately. The first reliability batch showed temperature-cycle failures above the target rate, traced to a packaging interface rather than the transistor layer. The fix took two quarters and a new supplier qualification. Meanwhile, the wearable brand moved part of its order to a competitor, citing better pricing, and the foundry's 28nm utilization fell to 78 percent by mid 2022 as the broader shortage eased.

The bet looked bad for about nine months.

Late 2022: the second customer arrives#

A European tier-one automotive supplier signed on for the newly qualified radar process in October 2022, followed by two more customers in the first half of 2023. By then the industry's capacity crunch had flipped into a glut on consumer nodes, and 28nm pricing had softened. The foundry's automotive line was running near full, at prices roughly 30 percent above what the wearable contract had offered.

What the decision actually turned on#

The foundry did not pick the long shot because it was visionary. It picked a narrower version of the long shot because the short-term deal had a clause that would have locked two lines for three years at a moment when nobody knew how long the shortage would last. The lesson is less about automotive than about contract length: in a cyclical industry, the option to change your mind is worth real money, and it is usually the thing a desperate buyer asks you to give up first.

The wearable brand, for its part, found capacity elsewhere and shipped its product on schedule. Both companies got what they needed. Only one of them had to write a $180 million check to find out.

More articles