To apply novel materials in future artificial neural networks, it is essential to fully understand the fundamental properties and mechanisms that control their building blocks, i.e., artificial synapses. One of the most promising devices for this purpose is the memristor, a memory resistor that exhibits energy-independent changes in resistance determined by the history of applied voltage and current. Previous research on memristive switching devices has primarily focused on conventional inorganic bistable materials, which typically provide only two conductive states. However, multistate devices are required to mimic the behavior of biological synapses and their defining feature, synaptic plasticity [1]. Hybrid semiconductor materials possess complex charge‑transport mechanisms involving both electronic and ionic conductance, which can lead to the formation of multiple resistive states dependent on the history of applied voltage and current. Combined with low‑cost fabrication methods, facile tunability, and mechanical flexibility, these materials offer promising avenues for overcoming the current limitations of devices used and studied today [2].

In this work, we report how the material composition and structure of the device fundamentally affect the current-voltage characteristics when evaluating the memristive behavior of materials and devices. To prepare such devices, high-quality thin layers of MAPI perovskite were used together with a commonly applied PMMA buffer layer, the only difference is whether it was filtered before deposition or not. This seemingly minor modification in the preparation of the memristive devices leads to a completely different behavior of the devices. In the case of filtered samples, we achieve a significantly higher ON/OFF ratio, approximately 2500, while for unfiltered samples it is only 43. In addition, in samples with unfiltered PMMA, redox reactions appear when measured at higher applied voltages. It is also worth noting that although samples with unfiltered PMMA show a lower on/off ratio, the samples nevertheless show stabilized j-V curves, while filtered samples undergo strong modification during continuous measurement. These results show that for any practical application, a detailed understanding of the structure–property relationship is essential; otherwise, these materials will not be suitable for commercialization.