Hardware acceleration for TinyML on low-power FPGA. Research results

[Yarib]

Dear community,

Here, I share the research results on low-power CNN inference using custom floating-point quantization and its dedicated hardware design. We use TensorFlow Lite micro on the smallest Zynq FPGA.

As a TinyML application, we implement structural health monitoring analytics based on acoustic emissions.

Here you have the links:

CNN Sensor Analytics With Hybrid-Float6 Quantization on Low-Power Embedded FPGAs

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10013365

Kind regards,
-Yarib Nevarez