Abstract
Logic-in-memory (LIM) hardware offers a promising approach to mitigate data transfer bottlenecks in von Neumann architectures. While memristive devices are viable candidates for such systems, switching voltage variability remains a key challenge. Here, we present tin hexathiophosphate (SnP2S6 or SPS) memristors that leverage intrinsic nanopore-assisted titanium ion migration to achieve reduced cycle-to-cycle variability. Through a guided filament formation mechanism, the device exhibits SET and RESET voltage variability of 6.62% and 9.17%, respectively, without requiring doping or additional processing steps. Density functional theory calculations suggest that the intrinsic nanopores and sulfur vacancies in SPS facilitate titanium ion transport and provide relatively stable pathways for filament formation. Beyond reliable switching, the SPS memristors support reconfigurable logic operations, enabling 14 distinct 2-input Boolean logic functions within a single device. Furthermore, a simplified proof-of-concept dynamic object tracking system integrated with a binary convolutional neural network demonstrates 96% classification accuracy in an all-digital domain without reliance on analog-digital conversion peripherals. These results suggest that SPS memristors could provide a potential platform for compact and reconfigurable LIM architectures with relevance to embedded and edge artificial intelligence applications.