摘要
Antiferromagnets (AFMs) offer exceptional promise for next-generation spintronic devices due to their ultrafast dynamics and resilience to external perturbations. However, while single-crystalline AFMs have been capable of being electrically manipulated, controlling polycrystalline AFM spins remains a major challenge due to their aperiodic nature. In this work, a N & eacute;el tensor is introduced as a rank-two symmetric tensor that statistically captures the spin correlations in polycrystalline AFMs, a fundamental departure from the conventional N & eacute;el vector approach. Using machine learning techniques, hidden statistical patterns in AFM spin structures are extracted, and establish the N & eacute;el tensor torque, an emergent symmetry-breaking mechanism at the FM/AFM interface. This torque enables field-free spin-orbit torque (SOT) switching in heavy-metal/FM/AFM trilayers. Furthermore, it is experimentally demonstrated that the N & eacute;el tensor can be trained and memorized, allowing the system to retain its switching polarity-an unprecedented feature in AFM spintronics. This work unveils previously hidden statistical correlations in polycrystalline AFMs, bridging the gap between theoretical models and practical spintronic applications. The findings lay the foundation for non-volatile, reconfigurable spintronic memory and neuromorphic computing, establishing the N & eacute;el tensor as a new degree of freedom for AFM-based SOT switching.