Large-scale graphene films are available, which enables the integration of graphene field-effect transistor (G-FET) arrays on chips. However, the transfer characteristics are not identical but diverse over the array. Optical microscopy is widely used to inspect G-FETs, but quantitative evaluation of the optical images is challenging as they are not classified. Here, we implemented a deep-learning-based semantic image segmentation algorithm. Through a neural network, every pixel was assigned to graphene, electrode, substrate, or contaminants, with exceeding a success rate of 80%. We also found that the drain current and transconductance correlated with the coverage of graphene films.

可以使用大规模的石墨烯薄膜，从而可以在芯片上集成石墨烯场效应晶体管（G-FET）阵列。但是，传输特性并不相同，而是在整个阵列上各不相同。光学显微镜被广泛用于检查G-FET，但是光学图像的定量评估具有挑战性，因为它们尚未分类。在这里，我们实现了一种基于深度学习的语义图像分割算法。通过神经网络，每个像素被分配给石墨烯，电极，基底或污染物，成功率超过80％。我们还发现，漏极电流和跨导与石墨烯薄膜的覆盖率相关。