Characteristics of the Thermodynamics of Novel Ag-Cation Controllable Diffusive Memristors for Artificial Neuron Application

<b><u>Introduction</u></b>: Diffusive memristors are essential elements for neuromorphic computing applications due to their ideal electrical characteristics such as high off-state resistance and a steep turn-on slope [1]-[2]. However, it is difficult to have characteristics with a large on/off ratio and a fast switching speed at the same time with conventional devices. Here, we reported a method to solve this problem using combination of Al₂O₃ metal oxide solid electrolyte layer and engineered defective graphene monolayer. We have also verified this characteristic through thermodynamic analysis for artificial neuron application.<br/><br/><b><u>Device Description</u></b>: The devices for this study were fabricated using the same method as presented in our previous paper [3]. Silver active metal layer was deposited by e-beam evaporation on a SiO₂/Si wafer, followed by patterning with lift-off method. For the filament-control graphene interlayer, chemical vapor deposition (CVD) of graphene on copper foils was done, and then, CVD graphene was transferred onto a SiO₂ / Si substrate with Ag bottom electrode using poly (methyl methacrylate) (PMMA) support layer. After the transfer of the graphene layer, the PMMA was removed with acetone. A hydrogen electrochemical reaction process and vacuum anneal were performed after the transfer process to form the defective area in the graphene and remove PMMA residue on the graphene film, and then the Al₂O₃ layer was deposited by atomic layer deposition (ALD). Finally, Pd inert metal was defined as the top electrode (TE). To analyze the characteristics of the thermodynamics of Ag-cation controllability with our system, MATLAB, and COMSOL Multiphysics were performed with the Ag-based threshold switching device measured parameters.<br/><br/><b><u>Results and Discussion</u></b>: The DC current–voltage characteristics of Ag/Graphene-based novel diffusive memristor were performed. The initial states of devices were normally in the off-state and their resistances were about 10¹² ohms at 0.1 V. When a graphene layer is placed between the active metal (Ag) and oxide electrolyte layer (Al₂O₃), the off-current was found to maintain its low value until the conductive filament is fully generated and the memristor is changed from off-state to on-state. The off-current of the novel memristor was ~10^-13 A. The on-current was~ 10^-3 A, with self-compliance characteristics, as reported in our previous paper [3]. To elucidate the self-compliance characteristics, we investigated the dynamics of conductive filaments with limited Ag cation concentration using our defective graphene. We utilized MATLAB and COMSOL Multiphysics to calculate the relationship between chemical potential and filament radius during the turn-on process. The filament dynamics during turn-on were analyzed based on the minimization of free energy, which comprises thermal energy, electrostatic energy, surface chemical energy, and volume chemical energy. We calculated each term in the free energy equation using our device structures, and the total free energy was obtained by summing up these terms. The radius of the conductive filament was found to have the minimum value at the intersection of the thermal energy and volume chemical energy terms. This indicates that the radius of the filament can be fixed, resulting in the achievement of self-compliance characteristics at this fixed point. This limitation can be related to the filament dissolution time which makes the fast reset speed for artificial neuron applications.<br/><br/>[1] Z. Wang, <i>et al.</i>, Nat. Mater. 16, 101–108 (2017). [2] F. Ye, <i>et al</i>., Adv. Mater., 2204778 (2022). [3] M. Song, <i>et al.</i>, Nano Lett. 23, 2952–2957 (2023).