A New Low-dose STEM Imaging Mode with Probability Driven Intra-pixel Beam Blanking

Transmission electron microscopy (TEM) has proven itself as an invaluable tool in understanding material properties, physical phenomena, and biological processes. Annular dark field scanning TEM (ADF-STEM) has emerged as a prevalent imaging mode due to its ease, interpretability, and flexibility with collecting multiple simultaneous signals. However, for beam sensitive samples (e.g. organic materials, zeolites), it becomes difficult or impossible to image the sample in its pristine state without modifications due to the beam or potentially catastrophic damage. Here we propose a new image formation method aimed at maximizing the information return per electron. Several approaches have emerged to mitigate beam damage effects, including dose rate control (e.g. multi-frame imaging), rapid beam blanking, or advanced scan strategies including interlacing and undersampling. As dose is lowered, the acquisition system becomes limited by Gaussian noise. Whilst this is improved by digital pulse counting imaging [1], it is over-simplistic to consider ADF-STEM imaging as a continuous signal, when in actuality it is an event stream of quantized electrons. Conventional ADF-STEM is, in essence, a scattering event rate sampled over a fixed time (the pixel dwell time). We propose a new event-driven approach whereby the time to detect a set number of electrons, n, is measured at every scan-point. For example, a heavy, strongly scattering atom, such as lead, will scatter n electrons onto the ADF detector faster than a low atomic number element, e.g. oxygen. This enables imaging with a bespoke number of detected electrons per pixel. The information gained can be described by a Bayesian model, with most of the image information contained within the first few electrons in each pixel time. To realize such a measurement experimentally, we combine the electron event detection of the turboTEM ‘Pulse’ digitizer [2] with the rapid blanking of the IDES electrostatic dose modulation (EDM) system [3]. We show experimentally that by obtaining the maximum signal information from each electron, we can control and drastically reduce dose. We demonstrate reduced beam damage for imaging organelles and graphene in human cells and solid-state materials whilst retaining the benefits of ADF STEM imaging. We quantitatively compare both the reliability and the information efficiency of the scattering probability images for ADF-STEM imaging [4]. Fig. 1 shows plots of the time-to-threshold detection events for increasing threshold choice. The n=1 image saves more than 95% of the beam-dose while retaining a bright-field like contrast suitable for life science applications and for basic sample navigation. A higher SNR image is captured at, say, n=10 while still offering a dose saving of ∼70%. Similar benefits are shown for oxides in Fig. 2, where the n=6 threshold image yields a dose saving of around 60% [4]. Fractional time-to-threshold images (top) of human macrophage cells exposed to graphene recorded with target detection values of 1, 3, 10, 25 electrons respectively from left to right. Experimentally observed exposure fraction (bottom) as a function of the electron detection target. Symmetry averaged ADF STEM image of SrTiO3 (left) and the simultaneously acquired probability observation (right) (n=6, dwell-time = 20 μs). Note the near identical appearance, even though the event-driven image uses an average of only 39.8% of the dose of the conventional image.