Logic Implementation Based on Selector-Only Memory for High-Efficiency In-Memory Computing

Abstract

In-memory computing (IMC) and logic-in-memory (LIM) architectures aim to reduce data-movement overhead by enabling computation directly within memory arrays. These approaches are considered promising for improving energy efficiency and throughput in data-intensive systems. Among various device platforms, memristor-based stateful logic has been actively investigated due to its compact structure and ability to perform logic operations through resistance modulation. However, its sensitivity to device variability and the need for selectors remain major obstacles for efficient logic implementation. In this work, we present a stateful logic scheme based on Selector-Only Memory (SOM), which operates through threshold switching rather than controlled resistance tuning. The polarity-dependent variation of threshold voltage in SOM devices enables digital-like state transitions and provides a stable mechanism for realizing logic implications, the resulting switching outcomes are represented by distinct threshold-defined states. Fundamental IMP and RNIMP functions were experimentally validated, and these primitives were sequentially combined to implement higher-order Boolean logic. The device-level threshold distributions, cycle-to-cycle stability, and drift characteristics were experimentally analyzed to identify operating margins that ensure reliable logic behavior. To evaluate the impact of variability on logic accuracy, Monte Carlo simulations were performed using experimentally extracted threshold voltage statistics. The results show that logic errors primarily originate from threshold variation, external-resistor selection, and drift-induced boundary shifts. Despite these sources of variation, SOM-based logic maintains stable performance within an optimized operating window defined by high- and low-voltage-state and appropriate pulse conditions. Overall, this study demonstrates that SOM provides a compact and reliable platform for threshold-driven logic-in-memory operation, with logic operations governed solely by threshold voltage. The combination of experimental verification and error simulations establishes SOM as a feasible device structure for efficient logic-in-memory architectures.