Die Yield And Cost Calculator: Step-by-Step Silicon Costing
Reviewed by the OnlineFree.app team · Updated
Key points
- Cost per good die is wafer cost divided by good dies, so yield moves the answer more than wafer price.
- The Die Yield And Cost Calculator uses the Murphy model with area in cm² and D0 in defects/cm².
- At 800 mm², 0.1 defects/cm² and $17,000 per wafer, one good die costs roughly $607.
- Splitting 800 mm² into eight 100 mm² chiplets cuts bare silicon cost to about $245 per package.
- Chiplet savings shrink once packaging, assembly yield loss and known-good-die test are added back.
What die yield actually tells you
Die yield is the share of dies on a finished wafer that pass test, and cost per good die is the processed wafer cost divided by the number of dies that pass. The Die Yield And Cost Calculator estimates both from three inputs — die area in mm², defect density in defects per cm², and wafer cost in USD per 300 mm wafer — and updates every output the moment you change one of them.
The yield model is Murphy's: Y = ((1 − e^(−A·D0)) / (A·D0))², with area converted to cm² and D0 in defects per cm². The practical message is that the area penalty grows faster than linearly. Doubling a die's area more than doubles the chance that a killer defect lands on it, which is why the model's curve bends so sharply near the reticle limit.
That gap is why a cheap wafer can still produce expensive silicon. If more than half your dies fail, you are effectively paying roughly double the wafer price for every part you ship, and no procurement win at the fab changes that arithmetic. Yield, not price, is usually the lever worth arguing about.
Three inputs, four outputs, one screen
Enter the design area first, in square millimetres. The field defaults to 800 mm² and accepts 25–3000 mm². Keep 858 mm² in view — that is the 26 × 33 mm reticle field, and anything above it needs stitching or a different exposure strategy, which is a separate cost conversation entirely.
Defect density (D0) is the input people guess at most often. The default is 0.1 defects/cm²; a mature leading-edge process typically lands somewhere in the 0.05–0.25 band, and the field goes up to 2.0 so you can model a line that is still ramping. Use your fab's reported figure if you have one, then re-run at the low and high end of the plausible range to see how much the answer actually depends on it.
Wafer cost defaults to $17,000 per 300 mm wafer. Enter the total processed-wafer cost from your foundry quote or internal cost model, not just the substrate or mask portion. The calculator assumes 3 mm edge exclusion, so it works with an effective radius of 147 mm on a standard 300 mm wafer).
Four numbers come back: predicted yield as a percentage, gross dies per wafer from the classic πR²/S − 2πR/√(2S) approximation, good dies per wafer after yield is applied, and cost per good die, plus cost per mm² of good silicon. If yield is so low that good dies floors to zero, the cost field shows a dash rather than a meaningless number.
Worked example: an 800 mm² die
At the defaults — 800 mm², 0.1 defects/cm², $17,000 per wafer — the calculator returns about 47% predicted yield, 61 gross dies per wafer, 28 good dies, and roughly $607 per good die, or about $0.76 per mm² of good silicon. Those are the numbers worth putting in a design review, not the wafer price.
Push the area to the 858 mm² reticle limit and yield falls to roughly 45%, gross dies to about 56, good dies to about 25, and cost per good die to roughly $680. A 7% increase in area costs about 12% more per good die — that sensitivity is why floorplanning arguments get so heated.
Now change only D0, to 0.2 defects/cm². At the same 800 mm², yield drops to about 25%, good dies to about 15, and cost per good die jumps to roughly $1,133. Doubling defect density nearly doubles your silicon cost while the wafer price has not moved at all.
Why chiplets beat one big die on cost
The second half of the tool splits the same total area into 2, 4 or 8 identical chiplets and reprices each scenario. Because total silicon area is held constant, you are comparing like with like: same wafer, same D0, same total mm², different granularity.
At 800 mm² total and D0 = 0.1, two 400 mm² chiplets yield about 68% each — roughly $183 per good chiplet, so about $366 for the pair. Four 200 mm² chiplets yield about 82%, roughly $71 each and about $283 per package. Eight 100 mm² chiplets yield about 91%, roughly $31 each and about $245 per package, against $607 for the monolithic die.
Two effects stack here. Smaller dies catch fewer killer defects, so yield rises, and smaller dies pack more efficiently into the round wafer, so less edge silicon is wasted. Notice that gross die count rises faster than area shrinks: 61 dies at 800 mm² versus about 613 at 100 mm², not the 488 you would get from simple proportionality.
What the comparison leaves out
Everything above is bare silicon cost. Chiplets only exist once they are packaged, and advanced 2.5D or 3D assembly, interposers and known-good-die test are billed per package — often dwarfing the silicon saving on a small part. Add your packaging quote to the per-package number before concluding anything. As of 2026 the calculator deliberately excludes those line items so the silicon arithmetic stays visible.
Assembly yield also compounds in the wrong direction. If each attach step is 99.5% good, eight chiplets give 0.995⁸ ≈ 96.1% package assembly yield, so you must start more packages than you ship. Interconnect PHY area, power and latency are real engineering costs the model cannot see.
And Murphy's model is a model: a single-parameter fit that ignores defect clustering, systematic failure modes and edge effects. Fabs often prefer negative binomial or critical-area analysis, and standards bodies such as SEMI define the metrology conventions behind those numbers. Treat the output as a fast order-of-magnitude check, verify against your own yield data, and remember the whole set of free online tools on OnlineFree.app runs in your browser without uploading your figures anywhere.
Frequently asked questions
How does the Die Yield And Cost Calculator estimate die yield?
It applies the Murphy yield model, Y = ((1 − e^(−A·D0)) / (A·D0))², with die area A converted to cm² and D0 in defects per cm². That yield is multiplied by an estimated gross die count to get good dies per wafer, and wafer cost is divided by that count. It is a single-parameter estimate, not a replacement for your fab's measured yield.
What defect density should I enter?
Start at 0.1 defects/cm², the tool's default, then try 0.05 and 0.25 to bracket a mature leading-edge process. If your fab reports yield, back-solve D0 or use its published figure. Higher defect density punishes large die far more than small die, so the sensitivity test matters more than the exact input.
Why does splitting a design into chiplets lower cost per good die?
Yield improves as area per die falls, so eight small chiplets each pass test far more often than one large die of the same total area. The Die Yield And Cost Calculator holds total silicon constant and multiplies the chiplet count, isolating that gain. Packaging, assembly and test cost are excluded and frequently erase it.
Does the calculator include packaging and assembly cost?
No. Cost per good die covers processed silicon only — wafer cost divided by good dies. Chiplet packaging, interposers, known-good-die test and assembly yield loss sit outside the model, so the chiplet scenarios look cheaper than a finished package would be. Add your OSAT or packaging quote before deciding anything.
What is the largest die area the calculator accepts?
The area field accepts 25 to 3000 mm², with 800 mm² as the default. Designs beyond 858 mm², the 26 × 33 mm reticle field, need stitching or multi-exposure approaches, so treat anything above that boundary as a planning exercise rather than a manufacturable single die.