equanimous
Norse clairvoyant shapeshifter goddess
Sub-mW Neuromorphic SNN audio processing applications with Rockpool and Xylo
Hannah Bos, Dylan MuirSpiking Neural Networks (SNNs) provide an efficient computational mechanism for temporal signal processing, especially when coupled with low-power SNN inference ASICs. SNNs have been historically difficult to configure, lacking a general method for finding solutions for arbitrary tasks. In recent years, gradient-descent optimization methods have been applied to SNNs with increasing ease. SNNs and SNN inference processors therefore offer a good platform for commercial low-power signal processing in energy constrained environments without cloud dependencies. However, to date these methods have not been accessible to ML engineers in industry, requiring graduate-level training to successfully configure a single SNN application. Here we demonstrate a convenient high-level pipeline to design, train and deploy arbitrary temporal signal processing applications to sub-mW SNN inference hardware. We apply a new straightforward SNN architecture designed for temporal signal processing, using a pyramid of synaptic time constants to extract signal features at a range of temporal scales. We demonstrate this architecture on an ambient audio classification task, deployed to the Xylo SNN inference processor in streaming mode. Our application achieves high accuracy (98%) and low latency (100ms) at low power (<4muW inference power). Our approach makes training and deploying SNN applications available to ML engineers with general NN backgrounds, without requiring specific prior experience with spiking NNs. We intend for our approach to make Neuromorphic hardware and SNNs an attractive choice for commercial low-power and edge signal processing applications.
CONCLUSION We demonstrated a general approach for implementing audio processing applications using spiking neural networks, deployed to a low-power Neuromorphic SNN inference processor Xylo. Our solution reaches high accuracy (98 %) with <100 spiking neurons, operating in streaming mode with low latency (med. 100 ms) and at low power (<200 µW dynamic inference power). Our software pipeline Rockpool (rockpool.ai) provides a modern machine-learning (ML) neural network approach to building applications, with a convenient high-level API for defining networks. Rockpool supports definition and training of SNNs via several automatic differentiation backends. Rockpool also supports quantization, mapping and deployment to
SynSense (formerly aiCTX) was established in 2017, and is the world’s leading supplier of neuromorphic intelligence and application solutions. SynSense focuses on the research and development of neuromorphic intelligence, based on 20+ years of world’s leading experience of University of Zürich and ETH Zürich. Centering on the application scenes of edge computing,SynSense has the full-stack service, and is the only neuromorphic technology company that involves both sensing and computing in the world.
SynSense makes a significant breakthrough in commercial neuromorphic chips, which is a crucial step towards cognitive intelligence and intelligent connectivity. SynSense will build the cognition ecology of intelligent connectivity,lead the development of neuromorphic intelligence in the world, and create well-being for the humans in the future.