Detecting Hidden Trap States from Adjacent Dwell Times in FinFET Random-Telegraph Noise

Random-telegraph noise from a single defect has been usually characterized by the distributions of its filled and empty dwell times. If a charge state hides several structural configurations, different Markov models can produce exactly the same two distributions, making the order in which the dwells occur an important source of information. We ask when the correlation between a dwell and the one that follows it is sufficient to detect that hidden structure. For a reversible model with two hidden sub-states per level, the covariance of adjacent dwells is a product of three factors. It vanishes only when the joint dwell distribution factorizes, with or without transitions inside a level. With more sub-states, the answer changes. When (n_E − 1)(n_F − 1) ≥ 2, where n_E and n_F count the hidden sub-states of the empty and filled levels, reversible models can have the same one-dimensional dwell laws, exactly zero correlation, and a joint distribution of higher rank. We construct such models explicitly and show that zero-correlation models with higher-rank joint distributions also occur on a published six-state defect topology. A Pearson chi-squared test on quantile-binned pairs of adjacent dwells detects these dependencies, but its cost depends mainly on the separation between sub-state time constants. Across 300 constructions, the median record length required for 90% power at five-sigma significance decreases from approximately 7 × 10⁸ dwell pairs when sub-state exit rates span less than a factor of two to approximately 6 × 10³ when they span more than a factor of sixteen. Simulated continuous trajectories match these independent-pair benchmarks to within 1.5 percentage points of power. A BSIM-CMG sample card is used only to confirm that the threshold-shift parameter acts as a rigid gate-voltage offset over the simulated range.