Analog computing and a hybrid approach to the element base of artificial intelligence applications

Strong demand for artificial intelligence technology is driving the development of complex high-performance applications with less power consumption. Analog computing is of high-performance and has simplified system, which simulates the physical processes occurring in nature. The universality of the digital coding allows getting a fairly accurate calculation result and provides saving without loss and additional restoration. The benefits of digital and analog computing systems can be enhanced by their hybridization. The type and level of hybrid computing depend on the complexity of the task set. Hardware implementation of a neural network offers promising solutions for computing tasks that require compact and low-power computing technologies. Artificial neural networks or ANNs, like biological neurons, are characterized by their capacity of learning and memorizing information, depending on architecture and weight of the latter. The literature review shows that stable weight storage can be achieved using digital weights and analog multipliers to reduce footprint. The proposed methodology for the network architecture provides optimal conditions for maintaining synaptic weights, increasing processing speed by the parallel weight perturbation.