BiKA: Kolmogorov-Arnold-Network-inspired Ultra Lightweight Neural Network Hardware Accelerator
#BiKA #Kolmogorov‑Arnold Network #multiplication‑free accelerator #binary thresholds #comparators #accumulators #Ultra96‑V2 #systolic array #resource utilization #edge device inference #hardware efficiency #arXiv preprint
📌 Key Takeaways
- Introduces BiKA, a Kolmogorov‑Arnold‑Network (KAN) inspired accelerator that eliminates multiplications entirely.
- Replaces nonlinear activation functions with binary, learnable thresholds, enabling a lightweight compute pattern.
- Prototype implementation on Xilinx Ultra96‑V2 FPGA shows 27.73% hardware resource savings over binarized accelerators and 51.54% savings over quantized systolic array designs while keeping accuracy competitive.
- Targets edge devices where power, area, and memory constraints are critical.
- Sets a new direction for designing hardware‑friendly neural networks that avoid expensive arithmetic operations.
📖 Full Retelling
The BiKA hardware accelerator, devised by researchers Yuhao Liu, Salim Ullah, and Akash Kumar, presents a new multiply‑free neural network design that substitutes traditional nonlinear functions with binary, learnable thresholds. The prototype, implemented on a Xilinx Ultra96‑V2 FPGA, was demonstrated in a February 2026 arXiv preprint (arXiv:2602.23455) submitted to the Computer Science Hardware Architecture and Artificial Intelligence categories, aiming to reduce power and area consumption for edge‑device neural network inference by leveraging only comparators and accumulators.
🏷️ Themes
Edge computing, Artificial intelligence hardware, Lightweight neural networks, FPGA acceleration, Quantization and binarization, Multiply‑free architectures
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Hardware Architecture arXiv:2602.23455 [Submitted on 26 Feb 2026] Title: BiKA: Kolmogorov-Arnold-Network-inspired Ultra Lightweight Neural Network Hardware Accelerator Authors: Yuhao Liu , Salim Ullah , Akash Kumar View a PDF of the paper titled BiKA: Kolmogorov-Arnold-Network-inspired Ultra Lightweight Neural Network Hardware Accelerator, by Yuhao Liu and 2 other authors View PDF HTML 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 computation pattern. The recently proposed Kolmogorov-Arnold Network 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. Subjects: Hardware Architecture (cs.AR) ; Artificial Intelligence (cs.AI) Cite as: arXiv:2602.23455 [cs.AR] (or arXiv:2602.23455v1 [cs.AR] for this version) https://doi.org/10.48550/arXiv.2602.23455 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Yuhao Liu [ view email ] [v1] Thu, 26 Feb 2026 19:20:55 UTC (306 KB) Full-text links: Access Paper: View a PDF of the paper titled BiKA: Kolmogorov-Arnold-Network-inspired Ultra Lightweight Neural Network Hardware Accelerator, by Yuhao Liu and 2 other authors View PDF HTML TeX Sou...
Read full article at source