Spin-Transfer Torque Magnetic RAMs (STT-MRAMs) are on their way to commercialization. However, obtaining high-quality test and diagnosis solutions for STT-MRAMs is challenging due to the existence of unique defects in Magnetic Tunneling Junctions (MTJs). Recently, the Device-Aware Test (DA-Test) method has been put forward as an effective approach mainly for detecting unique defecting STT-MRAMs. In this study, we propose a further advancement based on the DA-Test framework, introducing the Device-Aware Diagnosis (DA-Diagnosis) method. This method comprises two steps: a) defining distinctive features of each unique defect by characterization and physical analysis of defective MTJs, and b) utilizing march algorithms to extract distinctive features. The effectiveness of the proposed approach is validated in an industrial setting with real devices and data measurement.