Computing self-consistent local dose corrections for images to offset proximity effects poses significant problems because the number of computations needed cannot be done in a reasonable amount of time. Recently, we have demonstrated the use of an adaptive neural network method to increase the speed of these computations by several orders of magnitude. We have now implemented these corrections in practical hardware, and have introduced improvements in the algorithm that further reduce the computational complexity and time. The challenges in computing proximity effect corrections are to find algorithms that work well for general feature shapes, and that can be efficiently implemented. This paper will discuss the iterative computation of optimal corrections for electron scattering, and the limits of image resolution that can be obtained. It will describe the neural network algorithm used to obtain equivalent results more efficiently, and the method of adaptively determining the network's parameters. Finally, it will discuss several recent modifications that significantly improve the network's performance.
