BiKA: Kolmogorov-Arnold-Network-inspired Ultra Lightweight Neural Network Hardware Accelerator

Researchers propose BiKA, a new ultra-lightweight neural network accelerator inspired by Kolmogorov-Arnold Networks that uses binary thresholds instead of complex computations. The FPGA prototype demonstrates 27-51% reduction in hardware resource usage compared to existing binarized and quantized neural network accelerators while maintaining competitive accuracy.

• →BiKA introduces a multiply-free architecture using only comparators and accumulators for neural network computation.
• →The design reduces hardware resource usage by 27.73% compared to binarized neural networks and 51.54% compared to quantized networks.
• →BiKA replaces complex nonlinear functions with binary, learnable thresholds for extreme computational efficiency.
• →FPGA prototype testing on Ultra96-V2 validates the approach's effectiveness for edge device deployment.
• →The research opens new possibilities for hardware-friendly neural network design in resource-constrained environments.