Operational Limits of MEMS Nanosensing: Correlation Visibility and Irreversible Attribution under Finite Accessible Correlation MEMS nanosensing keeps getting “more sensitive,” yet the same hard failures keep resurfacing: temperature drift, packaging stress, nonlinearity, adsorption/contamination, and long-term instability. This work reframes these not as mere physical limits, but as structural consequences of how we *observe*, *evaluate*, and *decide*. Using the Operational Projection (OPT) view, measurement is modeled as an operation that inevitably discards information, introducing operational irreversibility. Building on this, Irreversible Attribution (IAF) clarifies that “signal vs artifact” judgments are not fixed at an instant; they become stabilized through practical chains of evaluation, tooling, and operational decisions—making reversal prohibitively costly. To make this controllable, the paper defines correlation visibility β as a quantitative margin for how far attribution can be delayed: β = Qsig / Qevalwhere Qsig represents signal-side evidence (multi-channel constraints, reference consistency, physical invariants) and Qeval represents evaluation-side burden (assumptions, degrees of freedom, calibration/model flexibility, heuristic thresholds). Large β means attribution can be delayed safely; small β means evaluation dominates and the system becomes “over-explainable” and fragile. Three design principles are organized as countermeasures:1) Evaluation placement (where judgment is performed in the pipeline),2) Attribution delay (keeping multiple hypotheses live until β improves),3) Correlation sealing (restricting harmful correlation pathways and bounding evaluation degrees of freedom). A concrete resonant MEMS example shows implementability using existing components: dual operational projections (PLL tracking + spectral sweep) plus a reference resonator for common-mode cancellation, enabling staged operation where attribution is delayed near β≈1 during exploration and fixed under β≫1 in operation. Appendix A extends the same logic across common MEMS challenges (temperature/packaging, 1/f and phase noise, adsorption, nonlinearity, sensor arrays and dimensional explosion), emphasizing that the core difficulty is not “can we measure,” but “when, where, and how do we irreversibly fix decisions.” This perspective generalizes beyond MEMS to any high-sensitivity, high-dimensional measurement system where evaluation itself can dominate what is believed to be “the signal.”
Koji Mochizuki (Thu,) studied this question.