Multilevel switching memristor by compliance current adjustment for off-chip training of neuromorphic system

CMOS-compatible compute-in-memory accelerators based on integrated ferroelectric synaptic arrays for convolution neural networks

“…The high accuracy of the VGG-8 network based on a ferroelectric synaptic transistor array could be achieved because of weight update characteristics of the FeTFTs, such as high linearity ( A pot = 0.8139; A dep = 1.1464), 64-level conductance states, G max / G min of 33.1, and small variation characteristics. In addition, the potentiation and depression characteristics with low conductance variation resulted in a high recognition accuracy of 91% for CIFAR-10 images with VGG-8 network using off-chip training method ( 57 ). The change in the conductance of synaptic devices after training can cause the degradation of the accuracy ( 58 ).…”

Implementation of Convolutional Neural Networks in Memristor Crossbar Arrays with Binary Activation and Weight Quantization

“…Convolutional neural networks (CNNs) have been utilized for image and speech recognition tasks due to their ability to extract features from large datasets. , A typical CNN structure consists of convolutional and pooling layers for filtering and fully connected (FC) layers for classification, where the rectified linear unit (ReLU) function is commonly used as the activation function in each layer. Particularly, the feature images obtained from input images through convolutional and pooling layers are then passed to the FC layers for pattern classification, which enables CNNs to achieve high accuracy in image recognition on popular datasets like the Modified National Institute of Standards and Technology (MNIST) and CIFAR-10. , To implement hardware CNNs with synaptic devices, it is crucial for electronic devices to possess nonvolatile characteristics and support varying states of weights to minimize performance degradation. − Additionally, the computational requirements of CNNs are significant, making it challenging to deploy them on numerous electronic devices. This challenge has led to growing interest in developing hardware implementations of CNNs.…”

Optimization of Random Telegraph Noise Characteristics in Memristor for True Random Number Generator

“…Memristor realized by a metal-insulator-metal structure is one of the most emerging nonvolatile memories thanks to the advantages of low-power operation, metal-insulator-metal structure, and fast-switching operation and has been utilized not only for stand-alone memory device but also for various computing applications including analogue computing for neuromorphic system, and stochastic computing such as physical unclonable function. [18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] TRNG is also one of these applications and several studies have employed random telegraph noise (RTN) in a memristor device as an entropy source for TRNG. [36][37][38][39][40][41][42][43][44][45][46] RTN is an intrinsic randomness characteristic of electronic devices and is mainly caused by the capture and emission of electrons in a trap site.…”