Kong Tie-Lin, Bie Jie, Chen Zhuo, Fa Wei, Chen Shuang
National Laboratory of Solid State Microstructures and Department of Physics, Nanjing University, Nanjing, Jiangsu 210023, China.
Kuang Yaming Honors School, Nanjing University, Nanjing, Jiangsu 210023, China.
ACS Appl Mater Interfaces. 2025 Jun 18. doi: 10.1021/acsami.5c04408.
Memristors are nonlinear resistors with memory, capable of multiple nonvolatile resistance states. They promise to break through the von Neumann bottleneck, enhance computing speed, and reduce device scaling, ultimately enabling advanced artificial intelligence (AI) computing. Ferroelectric memristors, which modulate resistance through electric-field-induced polarization switching, are considered leading candidates for neuromorphic computing and hold great promise for advancing AI. Understanding their mechanism is key to improving real-world performance. A time-dependent current transport model for ferroelectric memristors with ultrathin ferroelectric layers, i.e., ferroelectric tunnel junctions (FTJs), integrating the Thomas-Fermi screening theory, nonequilibrium Green's Function (NEGF), and polarization reversal dynamics, has been developed to estimate their current response. An optimized processing is proposed to save computational effort. It is assumed that the up and down polarization states of the ferroelectric film are likely to occur at a given voltage. These two states are treated equally to calculate the corresponding potential profiles. Based on these potential results, standard currents of FTJs in these two states are computed by using the time-consuming NEGF method. A particular multidomain polarization switching model is proposed to estimate proportions of two polarization states in the FTJ ferroelectric film at a specific voltage. Based on this model, not only the coercive field but also the polarization reversal speed of a thin film can be estimated. Then, the current response to input voltage is computed as a linear combination of each standard current weighted by its corresponding proportion. An ionic two-dimensional van der Waals (2D vdW) material, CuInPS (CIPS), regarded as an ideal ferroelectric material, is taken to construct model FTJs to test our proposed time-dependent current transport model. Finally, the current response of CIPS-based FTJs to continuously varying input voltage is estimated to well measure their synaptic functions for neuromorphic computing. Our developed model provides an effective approach to not only quickly compute current-voltage curves of FTJs but also accurately simulate their synaptic functions without experiments, accelerating the research and development of these devices.
忆阻器是具有记忆功能的非线性电阻器，能够呈现多种非易失性电阻状态。它们有望突破冯·诺依曼瓶颈，提高计算速度，并减少器件尺寸，最终实现先进的人工智能（AI）计算。铁电忆阻器通过电场诱导的极化切换来调制电阻，被认为是神经形态计算的主要候选者，在推动人工智能发展方面具有巨大潜力。了解其机制是提高实际性能的关键。一种用于具有超薄铁电层的铁电忆阻器（即铁电隧道结，FTJ）的时间相关电流传输模型已被开发出来，该模型整合了托马斯 - 费米屏蔽理论、非平衡格林函数（NEGF）和极化反转动力学，用于估计其电流响应。还提出了一种优化处理方法以节省计算量。假设铁电薄膜的上下极化状态在给定电压下可能出现。对这两种状态同等对待以计算相应的电势分布。基于这些电势结果，通过使用耗时的NEGF方法计算这两种状态下FTJ的标准电流。提出了一种特定的多畴极化切换模型来估计在特定电压下FTJ铁电薄膜中两种极化状态的比例。基于该模型，不仅可以估计矫顽场，还可以估计薄膜的极化反转速度。然后，将对输入电压的电流响应计算为每个标准电流与其相应比例加权后的线性组合。一种离子二维范德华（2D vdW）材料CuInPS（CIPS），被视为理想的铁电材料，被用来构建模型FTJ以测试我们提出的时间相关电流传输模型。最后，估计基于CIPS的FTJ对连续变化输入电压的电流响应，以很好地测量它们在神经形态计算中的突触功能。我们开发的模型提供了一种有效的方法，不仅可以快速计算FTJ的电流 - 电压曲线，还可以在无需实验的情况下准确模拟其突触功能，加速了这些器件的研发。
