Department of Materials Science and Engineering, Korea Advanced Institute of Science and Technology (KAIST), 291 Daehak-ro, Yuseong-gu, Daejeon, 34141, Republic of Korea.
Memory Division, Samsung Electronics Co. Ltd., Pyeongtaek-si, Gyeonggi-do, South Korea.
Nat Commun. 2022 May 19;13(1):2811. doi: 10.1038/s41467-022-30432-2.
Neuromorphic computing targets the hardware embodiment of neural network, and device implementation of individual neuron and synapse has attracted considerable attention. The emulation of synaptic plasticity has shown promising results after the advent of memristors. However, neuronal intrinsic plasticity, which involves in learning process through interactions with synaptic plasticity, has been rarely demonstrated. Synaptic and intrinsic plasticity occur concomitantly in learning process, suggesting the need of the simultaneous implementation. Here, we report a neurosynaptic device that mimics synaptic and intrinsic plasticity concomitantly in a single cell. Threshold switch and phase change memory are merged in threshold switch-phase change memory device. Neuronal intrinsic plasticity is demonstrated based on bottom threshold switch layer, which resembles the modulation of firing frequency in biological neuron. Synaptic plasticity is also introduced through the nonvolatile switching of top phase change layer. Intrinsic and synaptic plasticity are simultaneously emulated in a single cell to establish the positive feedback between them. A positive feedback learning loop which mimics the retraining process in biological system is implemented in threshold switch-phase change memory array for accelerated training.
神经形态计算针对神经网络的硬件体现，个体神经元和突触的设备实现引起了相当大的关注。忆阻器出现后，突触可塑性的仿真已经显示出有希望的结果。然而，涉及通过与突触可塑性相互作用进行学习过程的神经元内在可塑性很少得到证明。突触和内在可塑性在学习过程中同时发生，这表明需要同时实现。在这里，我们报告了一种神经突触器件，该器件在单个细胞中同时模拟突触和内在可塑性。 阈值开关和相变存储器在阈值开关相变存储器器件中合并。基于底部阈值开关层证明了神经元内在可塑性，类似于生物神经元中发射频率的调制。通过顶部相变层的非易失性切换也引入了突触可塑性。内在和突触可塑性在单个单元中同时进行仿真，以在它们之间建立正反馈。 在阈值开关相变存储阵列中实现了一个正反馈学习循环，该循环模拟生物系统中的再训练过程，用于加速训练。
