Search

In this work, Mn-doped BiFeO3 (BFMO) is employed as the functional layer of the memristor, and the resistive properties of the BFMO-based conductive filament-type memristors are modulated by altering the bottom electrode material (SrRuO3 and TiN) of the device and designing the oxygen-deficient and oxygen-reserving interfaces. Compared to Au/BFMO/SRO memristors with the oxygen-deficient interface, the cycling stability, ON/ OFF ratio, retention properties, endurance performances, and multivalue characteristics of Au/BFMO/TiN memristors are improved significantly by introducing oxygen-reserving interfaces. In addition, BFMO memristors with oxygen-reserving interface show lower nonlinearity factors of long-term potentiation (LTP) and long-term depression (LTD) and demonstrate higher recognition accuracy of 98.82% in convolutional neural network (CNN) for the Mixed National Institute of Standards and Technology (MNIST) handwriting recognition. Furthermore, considering different line resistances between the two BFMO-based memristors, a 128 $\times $ 128 array was involved to investigate the influences of the line resistance problem on IR-drop and network recognition accuracy.

The line resistance (LR) in a large-scale memristor crossbar array can cause serious IR-drop problem, degrading the hardware deployment capability of neural networks (NNs). In this work, two innovation schemes from the level of software are proposed to mitigate the hardware IR-drop problem by intentionally modulating the NN activation function before deploying. The methods are evaluated over typical activation functions and various line resistances on MLP and LeNet-5 for MNIST recognition. Results show the methods can significantly improve the tolerance of NNs to IR-drop and recover the accuracy in some extent. The methods require no extra hardware overhead and reduce the complexity of peripheral circuits, which make them more achievable and attractive.