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

TL;DR

BiKA introduces a multiply-free, comparator-based neural network accelerator inspired by Kolmogorov-Arnold Networks, significantly reducing hardware resource usage for edge devices while maintaining competitive accuracy.

Contribution

It presents a novel, ultra lightweight hardware architecture that replaces nonlinear functions with binary thresholds, enabling efficient deployment on resource-constrained devices.

Findings

Reduces hardware resource usage by up to 51.54%

Maintains competitive accuracy with traditional accelerators

Demonstrates effectiveness on FPGA prototype

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.