The QN908x integrates a co-processor (FSP) as hardware accelerator to offload the MCU from routine computations in data fusion and machine learning algorithms; the processing efficiency can be improved 5~10x. The FSP is on the AHB bus. Once it is programmed and the input data is ready, it starts working and generates an interrupt once the operation is complete. The interrupt triggers the MCU or DMA engine to fetch output data. To reduce current consumption for data transfers, the FSP has direct access to system memory to read input data and write the resultant output. The FSP can perform matrix operations, non-linear operations like sine and cosine, statistics like max and min values, and much more.