In this paper, a neuromorphic crossbar circuit with binary memristors is proposed for speech recognition. The binary memristors which are based on filamentary-switching mechanism can be found more popularly and are easy to be fabricated than analog memristors that are rare in materials and need a more complicated fabrication process. Thus, we develop a neuromorphic crossbar circuit using filamentary-switching binary memristors not using interface-switching analog memristors. The proposed binary memristor crossbar can recognize five vowels with 4-bit 64 input channels. The proposed crossbar is tested by 2,500 speech samples and verified to be able to recognize 89.2% of the tested samples. From the statistical simulation, the recognition rate of the binary memristor crossbar is estimated to be degraded very little from 89.2% to 80%, though the percentage variation in memristance is increased very much from 0% to 15%. In contrast, the analog memristor crossbar loses its recognition rate significantly from 96% to 9% for the same percentage variation in memristance.

In this paper, we propose a new twin crossbar architecture of binary memristors for low-power image recognition. In the new twin crossbar, we use two identical memristor arrays instead of using the previous complementary memristor arrays of M ⁺ and M ^- . Thereby, we can apply the discrete cosine transform (DCT) algorithm to reduce the number of low-resistance state (LRS) cells in the two identical M ⁺ arrays. With the reduced number of LRS cells in two M ⁺ arrays, the power consumption in the crossbar can be significantly saved compared to the previous complementary crossbar that is not suitable to DCT. When the number of discarded coefficients in the DCT matrix is 56.25%, 67.19%, 76.56%, and 84.38%, the power consumption of the new twin crossbar is reduced by 51.7%, 61.3%, 69.9%, and 77.4%, respectively, compared to the previous complementary one.

In this paper, we propose a new crossbar architecture of memristors with bipolar inputs for an image recognition application. The performance of the proposed crossbar array with bipolar inputs is based on the simplified Exclusive-NOR function to measure the similarity between the input pattern and the stored patterns. In the proposed architecture, only one crossbar array is used instead of two crossbar arrays. The number of memristors in the proposed architecture is reduced by 50%, when compared to the complementary and the twin architectures. The proposed crossbar architecture with bipolar inputs consumes 16.7% and 7.2% less power than do the complementary and twin architectures. In addition, using only one crossbar array in the proposed architecture can improve the fault tolerance of the crossbar circuit. When 10% of memristors are defective, the proposed crossbar architecture shows a recognition rate improved by 5%, 7% and 4% over that of the memristor binarized neural network, the complementary architecture of the memristor crossbar and the twin architecture of the memristor crossbar when recognizing 10 images.

Wire resistance in metal wire is one of the factors that degrade the performance of memristor crossbar circuits. In this paper, an analysis of the impact of wire resistance in a memristor crossbar is performed and a compensating circuit is proposed to reduce the impact of wire resistance in a memristor crossbar-based perceptron neural network. The goal of the analysis is to figure out how wire resistance influences the output voltage of a memristor crossbar. It emerges that the wire resistance on horizontal lines causes the neuron’s output voltage to vary more than the wire resistance on vertical lines. More interesting, the voltage variation caused by wire resistance on horizontal lines increases proportionally to the length of metal wire. The first column has small voltage variation whereas the last column has large voltage variation. In addition, two adjacent columns have almost the same amount of voltage variation. Under these observations, a memristor crossbar-based perceptron neural network with compensating circuit is proposed. The neuron’s outputs of two columns are put into a subtractor circuit to eliminate the voltage variation caused by the wire resistance. The proposed memristor crossbar-based perceptron neural network is trained to recognize the 26 characters. The proposed memristor crossbar shows better recognition rate compared to the previous work when wire resistance is taken into account. The proposed memristor crossbar circuit can maintain the recognition rate as high as 100% when wire resistance is as high as 2.5 Ω. By contrast, the recognition rate of the memristor crossbar without the compensating circuit decreases by 1%, 5%, and 19% when wire resistance is set to be 1.5, 2.0, and 2.5 Ω, respectively.

This paper performs a comparative study on the statistical-variation tolerance between two crossbar architectures which are the complementary and twin architectures. In this comparative study, 10 greyscale images and 26 black-and-white alphabet characters are tested using the circuit simulator to compare the recognition rate with varying statistical variation and correlation parameters. As with the simulation results of 10 greyscale image recognitions, the twin crossbar shows better recognition rate by 4 % on average than the complementary one, when the inter-array correlation = 1 and intra-array correlation = 0. When the inter-array correlation = 1 and intra-array correlation = 1, the twin architecture can recognize better by 5.6 % on average than the complementary one. Similarly, when the inter-array correlation = 1 and intra-array correlation = 0, the twin architecture can recognize 26 alphabet characters better by 4.5 % on average than the complementary one. When the inter-array correlation = 1 and intra-array correlation = 1, the twin architecture is better by 6 % on average than the complementary one. By summary, we can conclude that the twin crossbar is more robust than the complementary one under the same amounts of statistical variation and correlation.

Semantic segmentation of aerial and satellite images plays a crucial role in a wide range of applications and services, catering to the increasing needs of environmental resource management, urban planning, and traffic safety. Many efficient semantic segmentation methods have been proposed by exploiting deep learning techniques with convolution neural networks (CNNs) architectures and self-attention mechanisms to achieve superior accuracy if compared with conventional machine learning-based approaches. In this paper, we introduce a hybrid bilateral segmentation network (HBSeNet), a novel semantic segmentation architecture. Inspired by the success of dual-path network models that replace conventional single-branch encoder-decoder architectures, we construct a model with the core idea of combining the context path and the spatial path to optimize both the accuracy and the complexity of deep learning model in the field of remote sensing image segmentation. Moreover, HBSeNet innovates with auxiliary modules designed to enhance its performance, such as sequential atrous convolution, information synthesis module, and bridge for efficient multi-scale feature extraction, fusion, and integration. In simulations, our model achieves a global accuracy of 92.04%, a mean intersection-over-union (mean IoU) of 83.57%, and a mean boundary-F1-score (mean BFScore) of 90.23% when evaluated on the ISPRS Potsdam dataset, surpassing the state-of-the-art segmentation models such as DeepLabV3+, SwinCNN, and ST-UNet.