Hardware-Efficient Locality-Preserving Graph Processing via Ferroelectric Fowler-Nordheim Tunneling Crossbar Array

Graph analysis provides deep insights into complex real-world systems, yet conventional digital-based hardware is inefficient for graph data processing and faces significant scalability bottlenecks. Although metallic-cell-at-diagonal crossbar arrays (mCBAs) have been proposed as hardware platforms for graph processing by exploiting controlled sneak current, the relationship between device-level transport characteristics and the distance-dependent signal attenuation remains largely unexplored. In this work, we demonstrate an mCBA hardware composed of ferroelectric Fowler-Nordheim tunneling diodes (FFN-diodes) with high ON/OFF and rectifying ratios. The FFN-diodes' strong bias-dependent nonlinearity induces a sharp signal decay during multi-hop propagation, which naturally suppresses long-range noise while preserving local connectivity. Using experimental measurements and SPICE simulations, we quantitatively correlate the device characteristics with the effective attenuation factor governing graph reachability and benchmark the hardware operation against Katz centrality. Furthermore, a locality-preferred hierarchical community detection algorithm is implemented, demonstrating enhanced modularity and structural coherence. These results establish a direct link between device physics and graph-algorithm performance, positioning FFN-diode-based mCBAs as an efficient platform for locality-aware graph processing.