The LibFPGA blog

Practical FPGA design, one problem at a time: FIFOs and clock domains, timing closure, RISC-V soft cores, and the tools that make the hard parts routine.

Latest

Binary Neural Networks on FPGAs: When a Look-Up Table Becomes the Neuron

Here is a genuinely strange idea: you can build a neural network with no multipliers at all. No floating point, no DSP blocks doing multiply-accumulate, not even integer multiplies. Just XNOR gates and a circuit that counts bits. And on an FPGA the payoff is beautiful, because a single look-up table, the basic atom of the chip, can become a neuron. This is the story of binary neural networks (BNNs), the most FPGA-native corner of AI hardware, and why the match is so good it feels like cheating.

Jul 11, 2026 · 17 min read