Random telegraph signal (RTS) noise is ubiquitous in electronic and electro-optical devices, having been observed in MOSFETs and photodiode arrays. For imaging arrays, in particular, RTS noise (blinking pixels or "blinkers") deteriorates system performance through poor nonuniformity correction (NUC) stability and degrades image quality with blinking pixel behavior that can distract human operators and confuse computer vision algorithms. To date, there exists no universally accepted identification method or description of RTS noise in photodetectors, nor a conventional analysis approach to determine its origin. Current approaches typically focus on spectral properties (RTS noise is characterized by a Lorentzian power spectrum), which can be expensive to compute through Fourier methods, and analysis is usually performed on only a small sample of pixels. Here, we propose a method to identify and characterize blinkers by training a hidden Markov model (HMM) to extract the principal parameters governing blinking behavior, including the underlying state space, the state transition probabilities, and the distribution of state output levels. We find evidence to support classifying blinking behavior with HMM parameters; the variation of the model parameters with extrinsic variables, such as the temperature and applied bias, give some indication of the underlying physical mechanisms. Specifically, we find the timescale of the blink current is longer than typical electron{phonon, electron{electron, and electron{photon interactions, which leads to the suggestion that the blinking mechanism may be related to trap occupation dynamics.
|