Edge detection has made significant progress with the help of deep Convolutional Networks (ConvNet). ConvNet based edge detectors approached human level performance on standard benchmarks. We provide a systematical study of these detector outputs, and show that they failed to accurately localize edges, which can be adversarial for tasks that require crisp edge inputs. In addition, we propose a novel refinement architecture to address the challenging problem of learning a crisp edge detector using ConvNet. Our method leverages a top-down backward refinement pathway, and progressively increases the resolution of feature maps to generate crisp edges. Our results achieve promising performance on BSDS500, surpassing human accuracy when using standard criteria, and largely outperforming state-of-the-art methods when using more strict criteria. We further demonstrate the benefit of crisp edge maps for estimating optical flow, generating object proposals and semantic segmentation. In addition, the proposed refinement architecture can be easily generalized to saliency detection task, achieving state-of-art results on five commonly used saliency detection benchmark.

Heavy metal/ferromagnetic layers with perpendicular magnetic anisotropy (PMA) have potential applications for high-density information storage in racetrack memories and nonvolatile magnetic random access memories. Writing and erasing of information in these devices are carried out by domain wall (DW) motion and deterministic magnetization switching via electric current generated spin orbital torques (SOTs) with an assistance of in-plane bias field to break the symmetry. Improvements in energy efficiency could be obtained when the switching of perpendicular magnetization is controlled by an electric current generated SOTs without the in-plane bias fields. Here, we report on reversible electric-current-driven magnetization switching through DW motion in Pt/Co/Cr trilayers with PMA at room temperature due to the formation of homochiral Neel-type domain, in which an in-plane effective Dzyaloshinskii-Moriya interaction field exists. Fully deterministic magnetic magnetization switching in this trilayers is based on the enhancement of SOTs from a dedicated design of Pt/Co/Cr structures with two heavy metals Pt and Cr which show the opposite sign of spin Hall angles. We also demonstrated that the simultaneously accompanying Joule heating effect also plays a key role for field-free magnetization switching through the decrease of the propagation field.

Magnetic trilayers having large perpendicular magnetic anisotropy (PMA) and high spin-orbit torques (SOTs) efficiency are the key to fabricate nonvolatile magnetic memory and logic devices. In this work, PMA and SOTs are systematically studied in Pt/Co/Cr stacks as a function of Cr thickness. An enhanced perpendicular anisotropy field around 10189 Oe is obtained and is related to the interface between Co and Cr layers. In addition, an effective spin Hall angle up to 0.19 is observed due to the improved antidamping-like torque by employing dissimilar metals Pt and Cr with opposite signs of spin Hall angles on opposite sides of Co layer. Finally, we observed a nearly linear dependence between spin Hall angle and longitudinal resistivity from their temperature dependent properties, suggesting that the spin Hall effect may arise from extrinsic skew scattering mechanism. Our results indicate that 3d transition metal Cr with a large negative spin Hall angle could be used to engineer the interfaces of trilayers to enhance PMA and SOTs.