FURI | Spring 2026

RTL Implementation and Verification of an FP4 Systolic Array for Edge AI Workloads

Today, AI models require enormous computational resources for training and inference, making efficient deployment more difficult. This research project implements a systolic array using 4-bit floating-point (FP4) arithmetic, performing matrix multiplication with significantly reduced hardware complexity compared to higher-precision data types. The design supports dual FP4 formats with INT8 accumulation. This energy-efficient matrix multiplication hardware will advance edge AI, enabling AI capabilities on devices where power and size constraints are limited.