A Novel Architecture of a fault-tolerant Memory System based on Missing Data Imputation

In the modern era of edge computing, each computing device is used as a resource of distributed computing elements as well as distributed storage system. Therefore, every computing device plays a crucial role in a distributed system. A huge amount of data are being collected, processed, and stored at every moment. Different kinds of AI engines are being trained by these data. The consistency of data, that is stored in distributed memories, is essential for the AI engines. The absence of data, due to memory failure, is very much effective for feature extraction. Memory failure may happen due to several causes like struck-at-fault, soft-error, etc. To get consistent data, memories should be fault-tolerant, where memory can work properly in faulty conditions. This can be done either by bypassing or by predicting the data which was stored in a faulty location. In this paper, the design of a novel architecture of memory is proposed which can predict the missing data due to memory failure.