Analysis and Mitigation of Coupled Nonidealities in Monolithic 3-D RRAM In-Memory Computing

Current designs of monolithic 3-D (M3D)-integrated resistive random access memory (RRAM)-based computing-in-memory (CIM) systems often lack comprehensive modeling of nonidealities and overlook the coupled effects among multiple nonideal factors. This may result in inaccurate performance estimation and suboptimal compensation strategies. This study establishes a novel model for temperature, current-resistance (IR) drop, and device-level variability in M3D-integrated RRAM structures from both fabrication and operational perspectives. Furthermore, a unified simulation framework is proposed to systematically investigate the coupled impact of these factors. The results reveal that temperature and IR-drop exhibit a partially counteracting relationship, while device-level variability, when coupled with either or both, significantly amplifies computational degradation. This study further proposes compensation recommendations tailored to different dominant nonideal scenarios, aiming to avoid excessive or isolated compensation. These recommendations provide a theoretical foundation for nonideality-aware compensation design in M3D-integrated CIM systems. Moreover, these findings are not only applicable to M3D-integrated RRAM structures but also provide valuable insights for traditional 2-D RRAM arrays.