Stimulus-Programmable Multi-Regime Depression in Ag-Modified TaOx Memristive Synapses
Please login to view abstract download link
Memristive devices with analog switching characteristics are promising candidates for hardware neuromorphic computing; however, their performance is often limited by fixed and non-ideal synaptic update dynamics. In particular, the lack of tunability in potentiation/depression behavior constrains both learning efficiency and achievable accuracy in physical neural systems. Here, we report Ag-modified TaOx memristive devices exhibiting stimulus-programmable depression curves (DC) with high polarity symmetry. Under pulsed operation, three distinct regimes were observed in the DC controlled by pulse amplitude: (i) a low-amplitude sigmoidal response that presents a gradual saturation, (ii) an intermediate regime characterized by steep initial updates, and (iii) a high-amplitude regime showing gradual, non-saturating conductance evolution. This non-monotonic and tunable behavior enables access to multiple synaptic update modes within a single device. In addition, the hysteresis switching loops exhibit lateral extensions (“wings”) associated with metastable intermediate resistance states, which remain stable over extended timescales. These observations indicate that different stimulation conditions can access distinct dynamical regimes within the same physical system, offering an additional degree of flexibility in synaptic operation. In this context, tuning the programming conditions may provide a pathway to implementing hybrid learning strategies directly at the device level, without the need for additional circuit complexity.