HfZrOx-Based Ferroelectric Synapse Device With 32 Levels of Conductance States for Neuromorphic Applications
HfZrOx-Based Ferroelectric Synapse Device With 32 Levels of Conductance States for Neuromorphic Applications
复制标题
DOI：
10.1109/led.2017.2698083
复制
发表时间：
2017-06-01
影响因子：
4.9
通讯作者：
Hwang, Hyunsang
中科院分区：
文献类型：
--
作者：
Oh, Seungyeol;Kim, Taeho;Hwang, Hyunsang
我们提出了一种基于 HfZrOx (HZO) 的铁电突触器件，其具有相当于多级电导态的多级剩余极化状态。通过优化脉冲条件，我们获得了增强和抑制的 32 级剩余偏振态。此外，使用所获得的多个剩余极化状态来模拟铁电场效应晶体管。仿真结果表明，通过应用最佳的增强和抑制脉冲条件可以获得线性和对称的电导状态。使用所提出的模式识别设备模拟神经网络。使用基于HZO的铁电器件的突触参数和神经网络模拟器，我们已经确认MNIST数据集的模式识别准确率为84%。它表明基于 HZO 的突触装置具有未来高密度神经形态系统的潜力。
We propose a HfZrOx (HZO)-based ferroelectric synapse device with multi-levels states of remnant polarization that is equivalent to multi-levels conductance states. By optimizing the pulse condition, we obtained 32 levels of remnant polarization states for both potentiation and depression. Furthermore, a ferroelectricfield-effect transistor is simulated using the obtained multiple remnant polarization states. The simulation results show that linear and symmetric conductance states can be obtained by applying optimum potentiation and depression pulse conditions. A neural network was simulated using the proposed devices for pattern recognition. Using synapse parameters of the HZO-based ferroelectric device and a neural network simulator, we have confirmed that the pattern recognition accuracy of the MNIST data set is 84%. It shows that the HZO-based synapse device has potential for future high-density neuromorphic systems.