PROJECT TITLE :

Influence of Compact Memristors’ Stability on Machine Learning

ABSTRACT:

Memristors are gaining popularity due to their high integration and parallel computation capabilities, which have the potential to accelerate the development of machine learning. Because memristors are prone to internal and external variabilities, these variabilities degrade memristor performance and, as a result, memristive neural network performance. The impact of memristor stability on machine learning is investigated in this work. Following the model's four variation parameters, the variations of maximum memristances, conductive filaments' change speeds, initial conductive filaments' lengths, and minimum memristances, two typical machine learning methods, a feed-forward network and data clustering, as representatives of supervised and unsupervised learnings, are tested. The shifting speeds of conductive filaments' length play a vital role in a feed-forward network, according to the findings. Furthermore, the smaller feed-forward network has a tendency to perform worse. Variations in the maximum and lowest memristances have a determining effect on performance in data clustering. While the lengths of conductive filaments vary, the speeds and initial lengths of conductive filaments have a random effect. Furthermore, the size of neural networks has little effect on the migration pattern of clustering centers. We expect that the research presented in this paper will help to clarify the role of memristors in machine learning and provide instructions for designing and fabricating memristive neural networks.