Memristor-Based Circuit Demonstration of Gated Recurrent Unit for Predictable Neural Network
Zhang Zhang, Qi Chen, Tingting Han, Chao Li, Yulin Liu, Gang Liu
Abstract
Analysis of time-series data can be used to recognize long-term trends and make correct forecasts. Compared with artificial neural network (ANN), gated recurrent unit (GRU) can process time-series signals and has a wide range of applications in natural language processing, speech recognition, machine translation, and so on. However, GRU models suffer from bottlenecks in hardware implementation due to the large number of parameters and circuit complexity. Here, we build a memristor-based GRU unit with full circuit function yet fewer input–output parameters. Inclusion of the as-designed GRU unit into predictable neural network allows the recognition and prediction of handwritten characters with the accuracies of 93% and 92%, respectively. The implementation of GRU network with memristor circuit extends its capability in machine learning and artificial intelligence.