Chip Sense / Methodology
Estimator methodology
Formulas and default provenance for the estimate tools. Full write-up with equations: ESTIMATORS.md.
How estimates stay estimates
- Formulas are public teaching identities, not calibrated foundry models.
- Defaults are labeled estimate (or left blank, like WSPM). A
citedkind is reserved for future filing-backed parameters — tools currently ship estimate/user only. - Edits you make are labeled your input.
- Outputs use ~ / ranges and an amber estimate badge.
Open the tools: /tools · yield · export-controls · fab-capacity · packaging-cost · ai-cluster-demand
Yield estimator
Open tool →Formulas
- Poisson: Y = e^(−A·D)
- Murphy: Y = (1 − e^(−A·D)) / (A·D)
- DPW ≈ π(R − E)² / A
- Good dies/wafer ≈ DPW × Y
Defaults: Die area 8 cm², D0 0.15 /cm², 300 mm wafer, 3 mm edge exclusion — teaching estimates.
Export control simulator
Open tool →Formulas
- Stressed companies = union of active rule targets
- Severity score = sum of active rule weights (teaching scale)
Defaults: Public-policy themes (EUV + DUV + service on by default to match map scenario) — theme references, not legal text.
Fab capacity planner
Open tool →Formulas
- Good dies/month ≈ WSPM × utilization × DPW × yield
Defaults: WSPM starts at 0 (you must enter it). DPW/yield/utilization ship as teaching estimates.
Packaging cost model
Open tool →Formulas
- C_unit = substrate + assembly + test
- C_yield-adj = C_unit / Y_pkg
- C_batch = N × C_yield-adj
Defaults: Illustrative CoWoS-class cost split; sanity band ~$500–$2000 / AI GPU package (trade-press range).
AI cluster demand model
Open tool →Formulas
- GPUs = clusters × GPUs/cluster
- HBM stacks = GPUs × stacks/GPU
- Packages ≈ GPUs / Y_pkg
- Logic wafers ≈ GPUs / (DPW × Y_die)
Defaults: 8-GPU cluster shape, illustrative HBM stacks/GB — replace with your SKU.