Foundry utilization: reconcile wafer capacity and node mix
Foundry utilization asks how much of a defined production capability was used during a period. A larger share of revenue from advanced nodes does not establish that the whole factory network is full. Revenue mix combines prices and physical output, while utilization needs a compatible physical numerator and capacity denominator.
Keep three measures apart
TSMC’s annual reporting provides separate operating information on capacity and technology revenue mix. Those company disclosures are useful inputs, but this article’s reconstruction is an editorial worksheet. It is not TSMC’s official utilization calculation and makes no claim about its current factory loading.
Separate installed capacity, production-ready capacity and actual wafer output. Equipment installation does not necessarily mean a fab is qualified for customer production. A new line may undergo engineering runs before contributing saleable wafers. Conversely, maintenance can reduce the capacity actually available during the quarter without reducing the installed equipment base.
A hypothetical area-equivalent calculation
Consider two fictional production lines. Line A can process 60,000 units of 300 mm wafers in a month and produces 54,000. Line B has capacity for 90,000 units of 200 mm wafers and produces 72,000. Simple wafer counts would combine dissimilar physical units.
For an illustrative area conversion, one 200 mm wafer equals (200 / 300)² = 4/9 of a 300 mm wafer. Line B therefore has capacity of 40,000 and output of 32,000 in 300 mm area equivalents. Combined loading is (54,000 + 32,000) / (60,000 + 40,000) = 86%.
| Worksheet item | Line A | Line B |
|---|---|---|
| Physical wafer diameter | 300 mm | 200 mm |
| Stated monthly capacity | 60,000 | 90,000 |
| Area-equivalent capacity | 60,000 | 40,000 |
| Area-equivalent output | 54,000 | 32,000 |
| Line utilization | 90% | 80% |
This geometric conversion is not a yield or economic-value conversion. Edge losses, process complexity, tool bottlenecks and wafer starts versus finished shipments remain separate questions. Never use the area ratio to infer equivalent chip output or profitability across nodes.
Investigate the revenue mix independently
Suppose the advanced line raises its average selling price by 20% while physical output stays flat. Its revenue share can rise despite unchanged loading. Alternatively, advanced-node output can expand while mature-node output contracts. A group revenue increase can therefore coexist with idle capacity in another process family.
Build a second table with node family, wafer basis, revenue share, price evidence and ramp status. Mark price information unknown when it is not disclosed. Do not backsolve exact wafer volumes from revenue shares without a defensible price assumption, and do not present an assumed price as a company disclosure.
Choose a testable follow-up
If the thesis depends on fixed-cost absorption, inspect the utilization population that bears those costs. A mature fab and a newly ramping overseas fab may have different depreciation and qualification profiles. Group averages can hide the specific facility causing the margin movement. Request or locate management’s explanation rather than allocating all margin change to utilization.
Compare like quarters and document shutdown days. A lower denominator following planned maintenance can mechanically raise measured loading even when output falls. The final worksheet should contain a physical-load result, an economic interpretation and an explicit list of unresolved capacity definitions. None of these establishes a production forecast automatically.
FAQ
Is revenue divided by capacity a utilization measure?
It is revenue per capacity unit. Changes can reflect pricing and product mix, so label it differently from a physical loading ratio.
Can one company be compared with another directly?
Only when wafer basis, time period and capacity status are compatible. Otherwise compare the mechanisms and disclose the unresolved measurement differences.