tinyML Talks recorded May 12, 2021
“TinyML FPGA implementation for condition monitoring”
CEO of Infxl LLC, Colleyville, TX
Marketing Manager at Microchip Technology GmbH, Munich
We have reduced the size of the deep neural net inference engine by minimizing the intra-network connectivity, eliminating the need for floating-point data, and replacing the multiply-accumulate operation with just accumulation. The resultant small-footprint, low-latency deep nets are suitable for embedded applications in general. They are especially suited for processing data from IoT sensors (inertial, vibration, temperature, flow, electrical, biochemical, etc.) in battery-powered endpoint applications in wearables, robots, and automotive, particularly for predictive maintenance, real-time condition monitoring, and process automation use cases. The trained deep nets are delivered in the form of compact and simple C code that is suitable for MCU, DSP, and FPGA implementations. We present FPGA size and performance results on an IoT condition monitoring use case.