Every defect excursion at a 200mm specialty fab carries a price tag. Most operations teams have a rough sense of this, but in our experience, the mental model is usually off by a factor of two or three. The true cost isn't just the scrapped lot. It's the compounding effect of diagnostic labor, delayed lot releases, rework cycles, and the frequency at which these events recur. Attach real numbers to each component and the picture changes fast.
Building the Cost Model: Four Components
A defensible excursion cost model has four distinct buckets. Treat them separately before summing, because each responds differently to process improvement levers. Missing any one of them understates total cost in ways that make engineering headcount requests harder to justify.
1. Direct Yield Loss per Excursion Event
At 200mm analog and power device fabs, a single unresolved excursion event typically drives a yield penalty in the range of 0.5% to 2.5% on the affected lots. The variance is wide because specialty process flows differ significantly in die density, device sensitivity, and the number of process layers that interact with the defect signature.
To translate that range into dollars, you need a per-lot value. A 200mm lot running a mid-complexity CMOS process, 25 wafers, with die value at final test around $8-$15 per die, typically carries a gross lot value between $80,000 and $160,000 before yield loss. At a 1.5% yield penalty, that's $1,200 to $2,400 per lot, per excursion event. Not catastrophic in isolation. The problem is frequency.
2. Diagnostic Labor Cost
This is the component that most ops teams systematically undercount. An active excursion investigation at a specialty fab typically pulls 3 yield engineers into the loop, running parallel hypothesis tracks: process equipment data review, inline inspection correlation, and test data binning. In our tracking of similar investigations, an average cycle runs 4 engineering days before a root cause is confirmed or ruled out.
At a fully-loaded engineering cost of $250/hour, 3 engineers for 4 days runs approximately $24,000 per event. That number surprises people. It shouldn't. Specialty fab yield engineers aren't generalists; their time has a high replacement cost and an even higher opportunity cost when they're pulled off predictive work into reactive fire-fighting.
3. Scrapped Lot and Rework Costs
Not every excursion results in a full lot scrap. But a meaningful fraction do, particularly when the defect signature reaches final electrical test before identification. In a facility running 8 to 20 excursion events per quarter, we've seen scrap rates on investigated lots ranging from 15% to 40% depending on how early in the process the excursion was detected.
Rework, where applicable, adds its own layer. A wafer requiring a redo of a single critical layer carries rework costs in the $3,000 to $8,000 range per wafer, depending on the process step involved. For 25-wafer lots where partial rework is attempted on 10 wafers, that's $30,000 to $80,000 added to the event cost. And rework cycles occupy tool time that would otherwise run new product starts.
4. Frequency: Where the Model Gets Uncomfortable
The frequency term is the multiplier. Eight to twenty excursion events per quarter is a realistic range for a 200mm specialty fab running 15,000 to 30,000 wafer starts per month. Fact: at the upper end of that range, you're looking at roughly 80 events per year. At the lower end, 32.
Multiply by total cost per event and the annual burden becomes the number that gets into budget conversations.
Worked Example: Mid-Sized Specialty Fab
Let's anchor this to a concrete scenario. A 200mm fab running 20,000 wafer starts per month, primarily analog and mixed-signal, with a product mix that includes power management ICs and sensor ASICs. Lot value averaged at $110,000. Excursion frequency: 12 events per quarter (48 per year).
| Cost Component | Per-Event Estimate | Annual (48 events) |
|---|---|---|
| Direct yield loss (1.5% × $110K lot value) | $1,650 | $79,200 |
| Diagnostic labor (3 engineers × 4 days × $250/hr) | $24,000 | $1,152,000 |
| Scrapped lots (25% scrap rate, partial) | $27,500 | $1,320,000 |
| Rework (10 wafers × $5,000 avg, 30% of events) | $15,000 | $216,000 |
| Total | ~$68,150 | ~$2,767,200 |
The range across the full parameter space runs from roughly $400K/year at the low end (fewer events, faster diagnosis, low scrap rate) to $3.2M/year at the high end. The worked example sits in the middle of that range. Most specialty fabs we've engaged with fall somewhere between $800K and $2.5M when they run this model for the first time.
Sensitivity Analysis: What Diagnostic Cycle Time Does to Total Cost
The most operationally tractable variable in the model is diagnostic cycle time. You can't always prevent excursions, but you can systematically compress how long it takes to identify root cause. Here's what that lever does to annual cost at the 48-event frequency:
| Diagnostic Cycle Time | Labor Cost / Event | Scrap Rate Impact | Estimated Annual Cost |
|---|---|---|---|
| 8 days (baseline) | $48,000 | 35% (late detection) | ~$3.2M |
| 4 days (mid) | $24,000 | 25% (moderate) | ~$2.1M |
| 2 days (target) | $12,000 | 15% (early ID) | ~$1.3M |
| 1 day (optimized) | $6,000 | 8% (near-inline) | ~$800K |
The relationship isn't linear. Faster diagnosis does two things simultaneously: it cuts the direct labor cost, and it reduces the scrap rate because fewer lots advance past the affected process window before containment. A 4-day cycle to a 2-day cycle saves roughly $800K/year in this model. Going from 4 days to 1 day saves closer to $1.3M. The biggest gains come from early-stage pattern recognition, before investigations even start.
Practical note: The scrap rate reduction effect of faster diagnosis is typically larger than the labor savings. If your excursion cost reduction program is focused only on engineering headcount efficiency, you're looking at the wrong lever first.
Applying the Model in Practice
This framework is designed to be directional, not actuarial. You don't need three significant figures to make resource allocation decisions; you need order-of-magnitude clarity. Here's how to use it without getting lost in precision theater.
Start with your excursion log for the last two quarters. Count events. Estimate average diagnostic cycle time from event open to root cause confirmed. Look at lot disposition records: how many resulted in full scrap vs. rework vs. pass-through with yield penalty. That's enough to populate the model with real inputs.
The output won't be exact. It will be credible, which is what you need when justifying investment in defect pattern analytics, faster inline review cycles, or dedicated yield engineering capacity. A $2M/year number, even if the true figure is somewhere between $1.5M and $2.5M, lands differently in a resource discussion than a verbal assertion that "excursions are expensive."
In our experience running this analysis with specialty fab teams, the diagnostic labor component is consistently the number that surprises people. Direct yield loss feels concrete because it appears in yield reports. Diagnostic labor is invisible in most ERP systems because engineers log time against projects, not against excursion event IDs. That accounting gap makes the true cost invisible until you build the model deliberately.
What the Model Doesn't Capture
Two costs are real but hard to quantify without customer-specific data: expedite and schedule disruption costs, and customer relationship costs from excursion-related delivery misses. At specialty fabs serving automotive or industrial customers with tight JIT windows, a late lot tied to an unresolved excursion can trigger penalties or qualification re-reviews. That's a different conversation, but worth flagging when you present this model to executive teams.
Keep the model bounded to what you can measure. The four components above are sufficient to make the case for systematic diagnostic improvement. Everything else is context for the business stakes, not inputs to the cost calculation.
