Anticipative Objective Tracking With Short-Term Memristive Synapses

Short-term synaptic plasticity in biological systems is critical in performing temporal tasks, such as classifying and tracking moving objects. Emerging devices possess significant advantages over CMOS technology in achieving short-term synaptic plasticity. However, systematic validation of short-term synaptic computation using these emerging devices remains an open question. Here, we fabricate a TiOx-based memristor to mimic the short-term behavior and explore its computing ability for system applications. The memristors exhibit high uniformity (< 3.47% temporal variance) and stable transient pulse response (< 1% cycle-to-cycle variance). We achieved the short-term depression by introducing a fixed resistor and validated it by a constructed model. Then, we introduce the devices into a continuous attractor network, enhancing the attractor mobility and further implementing anticipative tracking tasks in head orientation and tennis moving. This work paves the way for building more functional neuromorphic systems utilizing short-term devices.