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

TL;DR

BiKA, a multiply-free architecture that replaces nonlinear functions with binary, learnable thresholds, introducing an extremely lightweight computational pattern that requires only comparators and accumulators, provides a promising direction for hardware-friendly neural network design on edge devices.

Abstract

Lightweight neural network accelerators are essential for edge devices with limited resources and power constraints. While quantization and binarization can efficiently reduce hardware cost, they still rely on the conventional Artificial Neural Network (ANN) computation pattern. The recently proposed Kolmogorov-Arnold Network (KAN) presents a novel network paradigm built on learnable nonlinear functions. However, it is computationally expensive for hardware deployment. Inspired by KAN, we propose BiKA, a multiply-free architecture that replaces nonlinear functions with binary, learnable thresholds, introducing an extremely lightweight computational pattern that requires only comparators and accumulators. Our FPGA prototype on Ultra96-V2 shows that BiKA reduces hardware resource usage by 27.73% and 51.54% compared with binarized and quantized neural network systolic array accelerators, while maintaining competitive accuracy. BiKA provides a promising direction for hardware-friendly neural network design on edge devices.

Figures (8)

• Figure 1: Difference between the computation of single neuron in MLP (up) and KAN (down)
• Figure 2: Difference between the ANN, BNN, KAN, and BiKA
• Figure 3: Converting one learnable nonlinear function to a series of weighted learnable thresholds
• Figure 5: Mixed all learnable thresholds with the sum of $\alpha$
• Figure 7: Computation in BiKA neurons when set quantization parameter, $m$, as one.