No abstract

Abstract:This paper proposes an energy-efficient reconfigurable architecture for deep neural networks (EERA-DNN) with hybrid bitwidth and logarithmic multiplier. To speed up the computing and achieve high energy efficiency, we first propose an efficient network compression method with hybrid bit-width weights scheme, that saves the memory storage of network LeNet, AlexNet and EESEN by 7x-8x with negligible accuracy loss. Then, we propose an approximate unfolded logarithmic multiplier to process the multiplication operations efficiently. Comparing with state-of-the-art architectures EIE and Thinker, this work achieves over 1.8x and 2.7x better in energy efficiency respectively.

This paper proposes an Energy-Efficient Reconfigurable Architecture (E-ERA) for Recurrent Neural Networks (RNNs). In E-ERA, reconfigurable computing arrays with approximate multipliers and dynamically adaptive accuracy controlling mechanism are implemented to achieve high energy efficiency. The E-ERA prototype is implemented on TSMC 45 nm process. Experimental results show that, comparing with traditional designs, the power consumption of E-ERA is reduced by 28.6%∼52.3%, with only 5.3%∼9.2% loss in accuracy. Compared with state-of-the-art architectures, E-ERA outperforms up to 1.78X in power efficiency and can achieve 304 GOPS/W when processing RNNs for speech recognition.

No abstract

No abstract