VIKIN: A Reconfigurable Accelerator for KANs and MLPs with Two-Stage Sparsity Support

TL;DR

VIKIN is a reconfigurable hardware accelerator that efficiently supports both KANs and MLPs with two-stage sparsity, improving inference speed and energy efficiency for AI workloads on edge devices.

Contribution

The paper introduces VIKIN, a unified hardware accelerator with novel pipeline and sparsity support for simultaneous KAN and MLP inference, addressing hardware limitations and workload diversity.

Findings

Achieves 1.28x acceleration replacing MLPs with KANs

Provides 1.25x speedup and 4.87x energy efficiency over edge GPU

Maintains low latency overhead for higher-accuracy KAN models

Abstract

Recently, multi-layer perceptrons (MLPs) widely used in modern AI applications suffer from limited real-time performance due to intensive memory access overhead. Kolmogorov--Arnold Networks (KANs) have attracted increasing attention as an alternative architecture with similar structures to MLPs but improved parameter efficiency. However, the lack of dedicated hardware support limits the practical performance benefits of KANs. Moreover, since many edge workloads still rely heavily on MLPs, accelerators designed exclusively for KANs become inefficient and impractical. In this work, we present VIKIN, a reconfigurable accelerator that efficiently supports both KAN and MLP inference using unified hardware. VIKIN introduces a pipeline execution mode and two-stage sparsity support for efficient KAN processing, while enabling parallel-mode acceleration to improve MLP throughput under the same sparsity framework. Experiments on real-world datasets demonstrate that replacing MLPs with KANs on VIKIN achieves $1.28\times$ acceleration with $19.58\%$ reduced accuracy loss. For a higher-accuracy KAN model requiring $3.29\times$ more operations, VIKIN incurs only $1.24\times$ latency overhead compared with the baseline KAN model. In addition, VIKIN achieves $1.25\times$ speedup and $4.87\times$ higher energy efficiency than a representative edge GPU when executing KAN workloads.