Titanic overconfidence -- dark uncertainty can sink hybrid metrology for semiconductor manufacturing

Hybrid metrology for semiconductor manufacturing is on a collision course with dark uncertainty. An IEEE technology roadmap for this venture has targeted a linewidth uncertainty of +/- 0.17 nm at 95 % coverage and advised the hybridization of results from different measurement methods to hit this target. Related studies have applied statistical models that require consistent results to compel a lower uncertainty, whereas inconsistent results are prevalent. We illuminate this lurking issue, studying how standard methods of uncertainty evaluation fail to account for the causes and effects of dark uncertainty. We revisit a comparison of imaging and scattering methods to measure linewidths of approximately 13 nm, applying contrasting statistical models to highlight the potential effect of dark uncertainty on hybrid metrology. A random effects model allows the combination of inconsistent results, accounting for dark uncertainty and estimating a total uncertainty of +/- 0.8 nm at 95 % coverage. In contrast, a common mean model requires consistent results for combination, ignoring dark uncertainty and underestimating the total uncertainty by as much as a factor of five. To avoid such titanic overconfidence, which can sink a venture, we outline good practices to reduce dark uncertainty and guide the combination of indeterminately consistent results.

💡 Research Summary

The paper addresses a critical yet often overlooked source of error in semiconductor manufacturing metrology—“dark uncertainty,” a hidden component of total measurement uncertainty that arises when results from different measurement techniques are inconsistent. The IEEE technology roadmap for advanced semiconductor nodes has set an ambitious target of ±0.17 nm (95 % coverage) for critical‑dimension (CD) linewidth measurements by 2028 and recommends hybrid metrology, i.e., statistically combining results from disparate measurement methods (e.g., CD‑SEM, optical critical‑dimension (OCD), CD‑SAXS, CD‑AFM) to achieve this goal.

The authors point out that most prior studies assume that the results to be combined are mutually consistent, an assumption that is rarely satisfied in practice. In real inter‑tool, inter‑method, and inter‑lab comparisons, the spread of measured values often exceeds the quoted uncertainties of the individual measurements. This excess variability is termed “dark uncertainty,” analogous to dark matter: it is invisible in standard uncertainty budgets but can dominate the total error budget.

To illustrate the impact of dark uncertainty, the authors revisit published inter‑comparison data for approximately 13 nm linewidths measured by imaging (TEM, SEM) and scattering (CD‑SAXS, OCD) techniques. They apply two contrasting statistical models:

1. Common‑Mean Model (Fixed‑Effect / Consistency Model) – This model forces all measurements to share a single true value and combines them using inverse‑variance weighting. It treats the reported uncertainties as the only source of variance, thereby ignoring any excess spread. When applied to the data, the model yields a combined uncertainty of roughly ±0.16 nm (95 % coverage), seemingly meeting the roadmap target. However, this estimate is severely optimistic because it does not account for the observed inconsistency among the methods.

2. Random‑Effects Model (Hierarchical / Meta‑Analysis Model) – This model acknowledges that each measurement may have its own underlying mean and that the observed variability includes both the reported (within‑study) uncertainties and an additional between‑study variance component, which the authors identify as dark uncertainty. Using either restricted maximum likelihood or Bayesian inference, the model estimates a between‑method variance of about 0.5 nm² (σ_dark ≈ 0.5 nm). The resulting combined uncertainty is ±0.8 nm (95 % coverage), roughly five times larger than the common‑mean estimate.

The paper then dissects the origins of dark uncertainty into four broad categories:

• Inter‑person variation – differing judgments, assumptions, or inadvertent errors made by metrologists.
• Inter‑tool variation – unmodeled instrument‑specific effects such as electron‑beam tilt in SEM, lens aberrations in optical microscopes, or beam shape variations.
• Inter‑laboratory variation – environmental differences, local standards, and procedural divergences across sites.
• Inter‑method variation – ambiguities in the definition of the measurand (e.g., whether linewidth is taken at the top, bottom, or an average across a trapezoidal profile) that affect imaging and scattering techniques differently.

Because many of these factors are either unknown or difficult to quantify, they manifest as excess variance rather than as explicit uncertainty terms. The authors argue that standard uncertainty evaluation (e.g., GUM) typically omits this “under‑water” portion, leading to systematic under‑estimation of total uncertainty.

To mitigate dark uncertainty and enable reliable hybrid metrology, the authors propose a set of practical guidelines:

• Explicit measurand definition – reach consensus on the exact physical quantity being measured and document any averaging conventions.
• Comprehensive calibration and correction – routinely characterize and correct instrument‑specific biases (e.g., beam tilt, optical misalignment) and include the residual correction uncertainty in the budget.
• Replication and repeatability studies – perform multiple replicate measurements on the same sample to quantify repeatability variance, which provides a baseline for detecting excess (dark) variance.
• Hierarchical statistical analysis – adopt random‑effects or Bayesian hierarchical models that can absorb inconsistent results and yield a transparent estimate of dark uncertainty.
• Transparent documentation – record all assumptions, judgment calls, and correction procedures to allow external review and reproducibility.

The authors conclude that while hybrid metrology holds promise for achieving the stringent CD uncertainty targets required for next‑generation semiconductor nodes, ignoring dark uncertainty can produce “titanic overconfidence,” potentially leading to costly design failures or yield losses. By recognizing, quantifying, and properly accounting for dark uncertainty, manufacturers can combine disparate measurement results in a statistically sound manner, thereby delivering trustworthy metrology that truly meets roadmap specifications.