Enhancing Energy-efficiency by Solving the Throughput Bottleneck of LSTM Cells for Embedded FPGAs
This work addresses energy-efficient inference for IoT sensor data processing on embedded devices, representing an incremental improvement in hardware optimization.
The paper tackled the throughput bottleneck of LSTM cells for embedded FPGAs, proposing an optimization that achieved 17534 inferences per second and 3.8 μJ per inference in traffic speed prediction, resulting in at least 5.4× faster throughput and 1.37× more energy efficiency than existing approaches.
To process sensor data in the Internet of Things(IoTs), embedded deep learning for 1-dimensional data is an important technique. In the past, CNNs were frequently used because they are simple to optimise for special embedded hardware such as FPGAs. This work proposes a novel LSTM cell optimisation aimed at energy-efficient inference on end devices. Using the traffic speed prediction as a case study, a vanilla LSTM model with the optimised LSTM cell achieves 17534 inferences per second while consuming only 3.8 $μ$J per inference on the FPGA XC7S15 from Spartan-7 family. It achieves at least 5.4$\times$ faster throughput and 1.37$\times$ more energy efficient than existing approaches.