Ranjan Abhijeet, Farooq Tamkeen, Chi Chong-Chi, Sung Hsin-Ya, Salinas Padilla Rudis Ismael, Lin Po-Hung, Wu Wen-Wei, Lu Ming-Yen, Mishra Rahul, Lai Chih-Huang
Department of Materials Science and Engineering, National Tsing Hua University, Hsinchu 30013, Taiwan.
Centre for Applied Research in Electronics, Indian Institute of Technology Delhi, New Delhi 110016, India.
Nano Lett. 2025 Apr 30;25(17):7089-7096. doi: 10.1021/acs.nanolett.5c01100. Epub 2025 Apr 17.
Neuromorphic computing aims to replicate the brain's efficient processing through artificial neurons and synapses, requiring binary and multilevel switching. We present a PtMn/(Co/Pd)/Ta device that uniquely enables dual spin-orbit torque (SOT) switching modes─binary and multilevel (analog)─within the same geometry and stack structure, eliminating the need for device modifications. Binary SOT switching is achieved via domain wall nucleation and propagation at moderate current levels (∼65 mA), while multilevel switching occurs via domain nucleation mode without significant propagation after a high-current treatment (∼85 mA). The transition between two modes originates from structural changes after the current treatment. These modes allow for neuronal and synaptic functionalities, with the device achieving 96% accuracy in digit/letter recognition on the MNIST data set using an artificial neural network (ANN). The device's robust perpendicular magnetic anisotropy (PMA), dual-mode switching under a small in-plane field (), and simplified fabrication underscore its promise as an energy-efficient solution for neuromorphic computing.
神经形态计算旨在通过人工神经元和突触复制大脑的高效处理过程,这需要二进制和多级开关。我们展示了一种PtMn/(Co/Pd)/Ta器件,它能够在相同的几何结构和堆叠结构中独特地实现双自旋轨道扭矩(SOT)开关模式——二进制和多级(模拟),无需对器件进行修改。二进制SOT开关是通过在中等电流水平(约65 mA)下的畴壁成核和传播实现的,而多级开关则是在高电流处理(约85 mA)后通过畴成核模式发生的,且没有明显的传播。两种模式之间的转变源于电流处理后的结构变化。这些模式实现了神经元和突触功能,该器件使用人工神经网络(ANN)在MNIST数据集上的数字/字母识别准确率达到了96%。该器件强大的垂直磁各向异性(PMA)、在小面内磁场下的双模式开关以及简化的制造工艺突出了其作为神经形态计算节能解决方案的潜力。