Xufeng Kou

Low-symmetry materials unlock rich Berry curvature physics and anomalous transport phenomena that are forbidden in high-symmetry quantum systems through rigorous crystalline symmetry constraints. Current approaches rely on external fields or complex heterogeneous stacking to break crystalline symmetries, which hinders scalable device integration. In this work, we demonstrate surface premelting engineering to create wafer-scale low-symmetry graphene on germanium-110 [Ge(110)]. Controlled premelting forms striped germanium surface patterns that reduce graphene symmetry from D _6h to C _1v , producing a room-temperature nonlinear Hall conductivity of ~11 micrometers per volt per ohm. First-principles calculations attribute this to originating from overtilted massive Dirac cones of hybridized germanium bands and the graphene Dirac cone. Integrated nonlinear Hall rectifiers generate >20 millivolts of output from radio frequency input and drive commercial voltage boosters and light-emitting diodes, establishing a complementary metal-oxide semiconductor–compatible route toward wafer-scale nonlinear quantum devices.

The deployment of large-scale 32-bit neural networks is hindered by the escalating parameter sizes and inherent susceptibility to noise. Here, we present a nonvolatile memory (NVM)-based probabilistic computing architecture, which harnesses both the stochastic nature and in-memory computing capabilities of NVM to enhance computation efficiency and robustness. Utilizing in-plane magnetized spin–orbit torque magneto-resistive random-access memory (i-SOT) cells, we demonstrate ultra-fast (400 ps) all-electrical switching, 1011 data endurance, and voltage-controllable probabilistic states with a low variation of 3.8%. These features enable the implementation of an on-chip SOT probabilistic binary neural network (SOT-PBNN) hardware, achieving near-baseline accuracy (88.2%) on CIFAR-10 classification tasks. Moreover, the stochastic training process endows the SOT-PBNN to attain a sevenfold improvement in classification accuracy over its 32-bit convolutional neural network counterpart under 25% write/read noise, while reducing the parameter size and inference energy by 1–2 orders. Our work establishes a lightweight framework suitable for realizing artificial intelligence platforms on resource-constrained edge devices.

Self-generated spin–orbit torque (SOT) in magnetic materials has emerged as a promising pathway for efficient magnetization control without external spin current sources. Here, we unveil an alternative mechanism for deterministic SOT polarity control based on inversion symmetry engineering via ferromagnet (Co) thickness gradients in [Co/Pt]p multilayers with the repeating periods p. Unlike conventional approaches relying on spin Hall effect asymmetry in heavy metals, we modulate the spatial profile of spin current absorption and reflection across the stack, creating a net damping-like SOT driven by gradient-induced imbalanced spin accumulation. By inverting the Co gradient, we achieved polarity-reversed perpendicular magnetization switching, demonstrating that the SOT originates from structural symmetry breaking within the ferromagnetic layers. Second harmonic measurements confirm robust SOT switching efficiencies up to 294 Oe cm2 MA−1, along with enhanced thermal stability and scalability beyond bilayer systems. Our findings establish the symmetry-governed design principle for spin transport, enabling tunable and thickness-robust spintronic devices.