Auto-thresholding for unbiased electron counting

As interest in fast real-space frame-rate scanning transmission electron microscopy for both structural and functional characterization of materials increases, so does the need for precise and fast electron detection techniques. Electron counting, with monolithic, segmented, or 4D detectors, has been explored for many years. Recent studies have shown that a retrofittable signal digitizer for a monolithic or segmented detector can be a sustainable and accessible way to enhance the performance of existing detectors, especially for imaging at fast scan speeds. Since such signal digitization uses a threshold on the gradient of the detector signal to identify electron events, appropriate threshold choice is key. Previously, this threshold has been set manually by the operator and is therefore inherently susceptible to human bias. In this work, we introduce an auto-thresholding approach for electron counting to determine the optimal threshold by maximizing the difference in identified counts from a stream with real electron events and a stream with only noise. This leads to easier operation, increased throughput and eliminates human bias in signal digitization. When pixel dwell time becomes shorter than scintillator response time, digitization of the detector signal is needed to avoid artefacts in STEM images. Optimizing the threshold for this digitization process automatically is crucial to achieve high-quality quantitative digitized images, free of human bias for what threshold yields the best digitization.