Integrating Software/Hardware Approaches for Compressive Sensing to Enhance the Performance of Ultra-High-Speed Electron Cameras

Sivananthan Laboratories will develop the compressive sensing (CS)/inpainting algorithms and associated software to optimize the acquisition speed, image resolution/sensitivity, and data transfer/storage requirements for an ultra-high-speed camera used in (scanning) transmission electron microscopy (S/TEM) applications. Sivananthan Laboratories will leverage its CS expertise through collaboration with Direct Electron, LP to integrate the efficiencies provided by the CS/inpainting algorithms into DE’s next generation compressive sensing readout S/TEM camera, to create an integrated imaging solution generating frame rates in excess of 100k fps (determining the upper speed limit is the goal of this phase I application). Achievement of this goal will address all of the sub-topics in section 15e of the DOE-SBIR call for proposals, and enable microscopy methods to investigate critical scientific challenges in batteries and energy storage systems, biofuels and structural biology, advanced materials and new synthesis methods, and energy conversion technologies such as heterogeneous catalysts. The main challenge for imaging materials and processes by (S)TEM is optimizing the electron dose/rate budget – how do we acquire the spatial/temporal resolution we need before the electron beam changes the observations? CS and other advanced mathematical frameworks for recovering images can uniquely satisfy the dose/rate budget requirements by optimizing sampling for the highest resolution at the minimum electron dose. While the mathematical framework of CS provides guarantees for recovering data from compressed measurements, the sampling schemes must be incorporated into the detector hardware so that the observation can be used to practically recover the original signal/image. In our approach, therefore, images are compressed during acquisition (i.e., the conversion of electrons to pixel intensity) using a custom direct detector, allowing for faster framerates and optimized data transfer/storage. This approach is resilient to many types of noise, which is particularly useful for the low-dose and rapid acquisitions needed for state-of-the-art (S)TEM observations. Integrated CS acquisitions have been demonstrated by the PI previously at the Pacific Northwest National Laboratory (PNNL). The goal of this proposal, therefore, is to use this knowledge to design faster, CS enabled (S)TEM detectors to make the most use of the benefits CS provides – rather than the CS being an after-market addition, hardwiring the compressive sampling approach into an integrated imaging system will allow the frame rate of the detector to surpass 100k fps without overwhelming the ability of the microscope user to transfer, store and analyze the complete dataset.