A study of a top-down error correction technique using Recurrent-Neural-Network-based learning
Masanori Natsui, Naoto Sugaya, Takahiro Hanyu · 2016
A new error correction scheme based on a brain-inspired learning algorithm, called Recurrent Neural Network (RNN), is proposed for resilient and efficient intra-chip data transmission. RNN has a feature to find partially-clustered time-series data stream and predict the next input data from previous input data stream. By utilizing this feature, a novel top-down error correction approach which considers the “context” included in the data stream and predicts original data by an acquired knowledge can be realized. In this paper, the performance of a RNN/BCH-hybrid error correction scheme for reducing the effect of false-positive detection is demonstrated through an experimental evaluation using a general purpose microprocessor.