Analytical and physical modelling of Y2O3- based memristive devices for neuromorphic computation

Abstract

This work is the amalgamation of analytical and physical modelling of Y2O3-based memristor crossbar array for neuromorphic computation. Firstly, in the case of analytical modelling, the advancement has been implemented in the existing analytical model and remove the limitations to make the new memristive device more dedicated towards the neuromorphic applications by introducing two novel internal state variables namely as forgetting rate and retention. The newly introduced parameters greatly helped in investigation of the working of the system and improved the synaptic behavior in terms of potentiation and depression processes by enabling re-stimulation process effectively. The developed analytical model is fully capable to emulate the highly dense memristive crossbar array-based neural network of biological synapses and can implement the learning capability of the neurons. These biological synapses control the communication efficiency between neurons by varying the synaptic weight between to neurons during effective communication process. During electrical stimulation of the memristive devices, the memory transition is exhibited along with the number of applied voltage pulses, which is analogous to the real human brain functionality. Further, to obtain the forgetting and retention behaviors of the memristive devices, a modified window function is proposed by incorporating two novel internal state variables. The obtained results confirm that the effect of variation in electrical stimuli on forgetting and retention is similar to that of the biological brain. The modelled data is well fitted with the fabricated Y2O3- based memristive crossbar array to evaluate the performance of the memristive array system and helps to understand the synaptic behavior in the neuromorphic computation. The developed analytical memristive model can further be utilized in the memristive system to develop real-world applications in neuromorphic domains. On the other hand, a physical electro-thermal modelling of nanoscale Y2O3-based memristor devices has also been carried out to understand and analyze the effect of symmetric and asymmetric electrodes variation on the resistive switching (RS) properties and device synaptic properties. The physical modelling of memristor device is carried out in a semiconductor physics-based tool i.e., COMSOL Multiphysics and MATLAB Livelink with a well-defined MATLAB script. The RS responses for the reported physical model show low values of coefficient of variability (CV) i.e., 6.69 and 7.11% in SET and RESET voltages, respectively, during cycle-to cycle variation which eventually confirms the stability of the modelled device. The physics-based simulation is carried out by considering minimum free energy of the materials at an applied certain voltage. The simulated results also exhibit a stable pinched hysteresis loop in the RS responses in symmetric and asymmetric electrodes combinations with an efficient ON/OFF ratio. Moreover, the simulated devices show the synaptic plasticity functionalities in terms of potentiation and depression processes with almost ideal linearity factor for symmetric and asymmetric electrodes combinations. Therefore, the presented work efficiently depicted the electrode material's suitability with the Y2O3 switching layer which shows better device performance and can also help the researchers to develop a perfect Y2O3-based memristor device for neuromorphic, digital, and logic applications.