Feature Extraction Accelerator for Streaming Time Series.

We present an FPGA-based accelerator architecture that can rapidly extract features from streaming time series. The accelerator currently extracts 25 features, and leaves approximately 30% of the resources unused on an AMD/Xilinx Alveo U280 FPGA. The FPGA-based accelerator can extract the same set of features significantly faster than a GPU or multi-core CPU while consuming far less power. Additional features can be extracted if desired, as long as doing so does not exceed the resource capacity of the FPGA.