Fast ptychographic reconstruction for low dose, binary 4D STEM

Abstract text (incl. figure legends and references)

The popularity of 4D-STEM and ptychography has expanded significantly in recent years since the development of fast pixelated detectors such as the pnCCD[1], which enable new applications of 4D-STEM to beam sensitive materials which require low dose capabilities. A strong focus has developed on dose efficient method of imaging highly beam sensitive materials including, battery materials, [2]. However despite the data sets becoming very low dose and therefore the number electron recorded being small the size of the datasets remains a challenge for fast data processing.

It has previously been demonstrated by O"Leary et al. that high quality ptychographic reconstructions can be obtained from sparse binary CBED patterns such that within individual diffraction patterns electron events recorded by the detector do not overlap and cumulative doses as low as 1000eÅ-2 [3]. It was demonstrated that despite the very low bit-depth in the diffraction information that high quality reconstructions could be obtained by single side band reconstruction.

We simplify the analysis of these datasets by only storing coordinates at which electrons are recorded, which in the case of the low dose data is very few. We then aim to demonstrate the advantage of this significantly reduces dataset for further processing and perform the ptychography reconstruction on this significantly reduced dataset.

The JEOL 4D Canvas used here is a charge integrating detector with a high sensitivity, to reduce the data to a binary list of electron detection events with a pre-set threshold level for the integrated signal to be registered as an electron detection event. Instead of storing the entire pixel array we transform the dataset into an event-stream of 4D-coordinates where electrons are detected. Fig 1. Shows an example of a low dose image (a) and relevant 4D-STEM diffraction pattern (b) with a summary of the initial lines of the 4D-hitlist (c).

The slowest step in the ptychographic reconstruction process usually the Fourier Transform of the square modulus of 4D dataset which is recorded by the detector. This can be very fast in the case of sparse binary data, especially when the number of electrons detected is low; the speed of the Fourier Transform scales linearly with the number of events recorded.

In this presentation we will demonstrate the memory and time improvement potential in this workflow for low dose ptychographic reconstructions and demonstrate that there is no decreased performance in terms of the quality of reconstruction. We demonstrate the promising capabilities of this to enable a shift towards real-time ptychographic reconstructions.

[1] H. Ryll et al., J. Instrum. 11 (2016)

[2] J.G. Lozano, G.T. Martinez, L. Jin, P.D. Nellist, P.G. Bruce, Nano Lett. 18 (2018) 6850–6855.

[3] C.M. O"Leary et al , Appl. Phys. Lett. 116 (2020)..

[4] The authors acknowledge use of characterization facilities within the David Cockayne Centre for Electron Microscopy, Department of Materials, University of Oxford and in particular the Faraday Institution (FIRG007, FIRG008), the EPSRC (EP/K040375/1 "South of England Analytical Electron Microscope") and additional instrument provision from the Henry Royce Institute (Grant reference EP/R010145/1).

Figure 1. a) Low dose ADF image of Au nanoparticle. b) Single diffraction pattern acquired on JEOL 4D Canvas equipped with a pnCCD® (S)TEM Camera® System. c) Edit of 4D-hitlist of electron interaction events.