Bilayered Oxide Heterostructure-Mediated Capacitance-Based Neuroplasticity Modulation for Neuromorphic Classification

To overcome the limitations of memristors in neuromorphic computation, memcapacitors are gaining attention owing to their scalability, low power dissipation, and sneak-path-free nature. This study focuses on the progressive capacitive switching of a bilayered metal-oxide WOx/ZrOx heterojunction memcapacitor. To gain a better understanding of the interfacial switching behavior, density functional theory simulations are used to analyze the defects and oxide formation energy of the heterostructure. The memcapacitive characteristics are studied using the capacitance-voltage curves under different voltage-sweeping conditions and impedance analysis. The memcapacitive characteristics can be attributed to the trapping of carriers in the depletion region of the WOx/ZrOx heterojunction, which is modulated by the relocation of oxygen vacancies under the electric field. The device exhibits a wider dynamic range of capacitance values than other metal-oxide memcapacitors reported, and demonstrates versatile synaptic functions, such as potentiation/depression behavior, paired-pulse facilitation, experience-dependent plasticity, and learning-relearning behavior. Furthermore, an accuracy of 99.01% is achieved in handwritten digit classification using the capacitive state as the weight through a computing-in-memory emulator. The results affirm the applicability of the WOx/ZrOx memcapacitor in future capacitive neural networks. A bilayered oxide WOx/ZrOx heterojunction memcapacitor demonstrates a wide dynamic capacitance range and multiple synaptic functions. The density functional theory simulation along with the impedance analysis and voltage sweeping rate-dependent capacitance measurement is used to characterize the interface and defect dynamics. 99.01% accuracy of dataset classification is conducted through multilayer perceptron and LeNet-5 neural network using CIMulator platform.image