An Efficient VLSI Architecture of Recurrent Neural Network

An Efficient VLSI Architecture of Recurrent Neural Network

연구 분야: Artificial Intelligence

학회: 2022 IEEE 6th Conference on Information and Communication Technology (CICT)

Recurrent Neural Network (RNN) is a kind of artificial neural network (ANN) in which the output of the previous time step is provided as an input in the current time step. For sequence prediction problems, RNNs are widely used. For each dataset, we require unique RNN architecture. In this paper, folded RNN architectures are proposed that can be trained and tested on the datasets such as daily temperature, shampoo sales, and monthly mean sunspot numbers, where 80% of the data is used for training RNN and 20% of the data is used for testing RNN. Also, this paper proposes a flexible RNN architecture that can be used for all these three datasets with respect to our requirement. The implementations of these works have been done in 45 nm CMOS technology.