From Python to FPGA: Modern Toolchains for Machine Learning Acceleration

 quantbeckman fpga and python

The DaCe FPGA Python programming paper (Python → FPGA/HLS workflows) Python FPGA Programming with Data-Centric Multi-Level Design

The 2017 arXiv paper (I checked: likely PyTorch/ML acceleration related),

PyLog / PyTorch‑inspired FPGA ML compilation (Sitao Huang 2021 PyLog paper),

PYNQ (Python + Zynq FPGA boards).

Recent advances in Python-based FPGA programming have made it increasingly feasible to use high-level languages for hardware acceleration in machine learning and data-intensive applications. The DaCe framework demonstrates how Python can be used to generate efficient FPGA designs through data-centric, multi-level transformations, while PyLog builds on this approach by enabling PyTorch-inspired computation graphs to be compiled onto FPGA hardware.

 Similarly, frameworks like PYNQ bring Python to Zynq-based FPGAs, allowing developers to prototype and deploy hardware-accelerated algorithms without writing low-level HDL. These developments complement broader trends in ML frameworks, such as the methods described in the 2017 arXiv paper, where Python-driven workflows simplify and optimize hardware-accelerated computation. Together, these works illustrate a growing ecosystem where Python serves as a powerful, high-level interface for FPGA-based machine learning and computational acceleration.