Several recent studies have proposed the utilization of emerging technology devices, such as ReRAM, spintronic, and phase change memory in hardware-implemented neural network designs. However, the current poor maturity of the manufacturing process of memristive devices limits the implementation of synapses to low precision weights and to smaller size crossbars, which could be an issue for complex, higher dimensions machine vector learning tasks (e.g., object recognition, classifications, etc). Face to these challenges, efficient hardware implementations use binarization for weights and activation functions in the attempt to reach better energy efficiency, reduce the utilization of memory and the execution time. Moreover, to compensate the immaturity of the emerging devices technology and achieve better convergence, accuracy, and speed for learning and inference process, the neural network has to be designed either with an increased degree of redundancy, or with error correction capabilities. To avoid the inherent hardware cost of the redundancy and counteract the aforementioned issues, we propose an approach combining the concept of Ensemble Neural Networks paradigm with analog in-memory hardware implementation with spin-orbit torque (SOT) spintronic devices. These devices are among the most power-efficient emerging technologies. The architectural performances, power, and accuracy are verified on several datasets, showing that these combined approaches allow not only a very good resilience to high bit error rates but also a great reduction in execution time and number of memory accesses with a further reduction of$\times 100$for the energy consumption thanks to the SOT spintronic-based device.

The disruptive potential of the IEEE 1687 standard does not come from a single innovation, but rather from its capacity of providing a unified framework where heterogeneous approaches can co-exist and interact. In this Special Session, we will present the complementary research activities performed in the TIMA laboratory covering different aspects of the standard (Mixed-Signal instrument testing, Embedded Aging Monitors and Test Access Securization), and their coordination thanks to the Manager-for SoC Test (MAST) software environment.

This paper presents the construction of a control loop for the concealment of defective pixels in an image or an image sensor. The proposed control loop is based on defective pixel detection methods, and the concealment of the detected defective pixels. Two detection algorithms have been used in the process. The first is based on distance and the second on the evaluation and analysis of local dispersion parameters of the pixel blocks in the input image. The concealment process uses the median filter. The proposed control loop takes as input an image and a defective pixel number threshold into an image, and outputs the final corrected image, after passing through defective pixel detection and correction phases if necessary.

This paper presents an online and integrated solution for diagnosis and healing of the image sensors based on a defective pixel detection and concealment method. The proposed defective pixel detection method is based on the evaluation and analysis of the median, range and other local dispersion parameters of the pixel blocks in the images taken by the to-be-tested image sensor. Once a defective pixel is detected, the concealment process uses a median filter, which consists of the substitution of the defective value by the median value of the pixel block. For an error detected in the image sensor, a diagnostic process can be used to determine the origin of the problem. The global detection and concealment process was implemented on an FPGA platform to evaluate the computational complexity and to confirm the feasibility of integrating the proposed solutions in image sensors.

This paper presents imager self-healing method based on detection and correction of defective pixels in the produced image file. The proposed method uses a neighborhood analysis with simple arithmetic operations including distance between the to-be-tested pixel and its neighbor pixels. A 2-dimensional 3 by 3 gray-scale image matrix around the to-be-tested pixel is used to estimate an expected pixel value and a weighted average. This average value is used as an adaptive threshold of the difference value between expected and actual pixel values. The performances in terms of sensibility, specificity, predictive values, and phi-coefficient of the produced results on a set of 144 distorted images (24 references images x 6 distortion types), are compared to another dead pixel detection methods performances. Experimental results demonstrated that our proposal produces the best results.

In this article, we apply different machine learning (ML) techniques for building objective models, that permit to automatically assess the image quality in agreement with human visual perception. The six ML methods proposed are discriminant analysis, k-nearest neighbors, artificial neural network, non-linear regression, decision tree and fuzzy logic. Both the stability and the robustness of designed models are evaluated by using Monte-Carlo cross-validation approach (MCCV). The simulation results demonstrate that fuzzy logic model provides the best prediction accuracy.