Benchmarking Keyword Spotting Efficiency on Neuromorphic Hardware
This work provides an incremental efficiency analysis for neuromorphic hardware in keyword spotting applications, relevant to developers of low-power edge AI systems.
The researchers benchmarked a two-layer neural network keyword spotter on Intel's Loihi neuromorphic chip against conventional hardware like CPU, GPU, and low-power devices, finding that Loihi achieved lower energy cost per inference while maintaining equivalent accuracy, with its advantage increasing for larger networks.
Using Intel's Loihi neuromorphic research chip and ABR's Nengo Deep Learning toolkit, we analyze the inference speed, dynamic power consumption, and energy cost per inference of a two-layer neural network keyword spotter trained to recognize a single phrase. We perform comparative analyses of this keyword spotter running on more conventional hardware devices including a CPU, a GPU, Nvidia's Jetson TX1, and the Movidius Neural Compute Stick. Our results indicate that for this inference application, Loihi outperforms all of these alternatives on an energy cost per inference basis while maintaining equivalent inference accuracy. Furthermore, an analysis of tradeoffs between network size, inference speed, and energy cost indicates that Loihi's comparative advantage over other low-power computing devices improves for larger networks.