Coherent X-ray diffraction imaging (CDI) is a powerful technique for operando materials characterization at the nanoscale. Uniquely, in this technique, the image resolution is not determined by the resolution of the optics used in the experiment but by the maximum angle through which the x-rays are scattered by the sample.  The challenge lies in recovering the phases of those scattered x-rays, which are not measured by the detector, to retrieve the image of the sample from the data. The phase retrieval process can be computationally expensive, and the current methods has inherent deficiencies. First, this method does not include any physical knowledge of the material or the phenomena being studied.   Furthermore, the resulting image is also limited by the signal-to-noise ratio of the data; so while signal may exist at greater angle, and hence higher resolution, it may be insufficient for the phase retrieval to return a reliable image at that resolution.

As part of an ongoing LDRD project, we are using deep convolutional neural networks to invert raw X-ray diffraction data to reconstruct the real-space image in a fraction of second, where previously the process could take minutes or hours.