Implementation of Low-Energy LSTM with Parallel and Pipelined Algorithm in Small-Scale FPGA

Regression and classification are necessary for biometric systems and carried out using machine learning. A method for regression and classification is long short-term memory (LSTM). We proposed and implemented algorithms for low-energy LSTM for the regression of microwave-sensor signals into a small-scale FPGA. We found that our FPGA-based parallel (including unrolled)-pipelined algorithm decreased the computation time by 95% compared with the FPGA-based sequential algorithm. In addition, the amount of energy consumption with the proposed algorithm was reduced by 92% and 91% compared with that with a high-end GPU and CPU, respectively.

[1]  Matin Hashemi,et al.  LSTM-Based ECG Classification for Continuous Monitoring on Personal Wearable Devices , 2018, IEEE Journal of Biomedical and Health Informatics.

[2]  Jason Cong,et al.  FPGA-based accelerator for long short-term memory recurrent neural networks , 2017, 2017 22nd Asia and South Pacific Design Automation Conference (ASP-DAC).

[3]  Hirofumi Taki,et al.  Accurate heartbeat monitoring using ultra-wideband radar , 2015, IEICE Electron. Express.

[4]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[5]  Berin Martini,et al.  Recurrent Neural Networks Hardware Implementation on FPGA , 2015, ArXiv.