Abstract
An RRAM-based computing system (RCS) is widely used in neuromorphic computing systems due to its fast computation and low cost. The immature fabrication processes cause high rate of hard faults and limited endurance of RRAMs restrict the life of RCS. These can degrade the accuracy of RCS remarkably. The training of neural network consumes lot of time due to large network size and more number of layers. We propose a hierarchical fault-tolerant and cost effective framework for RCS. The proposed flow can tolerate the faults in the next training phase using the inherent fault-tolerant capability of the neural network without using extra hardware. Proposed framework consists of significant classification and remapping, transfer learning, threshold training and online fault detection. The proposed solution is validated and evaluated on CIFAR-10 dataset on VGG11 and Alexnet networks. Proposed solution resulted in training time improvements of 40% and number of RRAM writes are reduced by 97%.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: IEEE Transactions on Circuits and Systems II: Express Briefs
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
