Machine learning enhanced in situ electron beam lithography of photonic nanostructures

We report on the deterministic fabrication of quantum devices aided by machine-learning-based image processing. The goal of the work is to demonstrate that pattern recognition based on specifically trained machine learning (ML) algorithms and applying it to luminescence maps can strongly enhance the capabilities of modern fabrication technologies that rely on a precise determination of the positions of quantum emitters like, for instance, in situ lithography techniques. In the present case, we apply in situ electron beam lithography (EBL) to deterministically integrate single InGaAs quantum dots (QDs) into circular Bragg grating resonators with increased photon extraction efficiency (PEE). In this nanotechnology platform, suitable QDs are selected by 2D cathodoluminescence maps before EBL of the nanoresonators aligned to the selected emitters is performed. Varying the electron beam dose of cathodoluminescence (CL) mapping, we intentionally change the signal-to-noise ratio of the CL maps to mimic different brightness of the emitters and to train the ML algorithm. ML-based image processing is then used to denoise the images for reliable and accurate QD position retrieval. This way, we achieve a significant enhancement in the PEE and position accuracy, leading to more than one order increase of sensitivity in ML-enhanced in situ EBL. Overall, this demonstrates the high potential of ML-based image processing in deterministic nanofabrication which can be very attractive for the fabrication of bright quantum light sources based on emitters with low luminescence yield in the future.