Tuneable Conductance Quantization in Crossbar Cu/Ta₂O₅/Pt Resistive Random Access Memory Devices

Given the current demands of data-intensive and cost-efficient computing, there exists a need of a memory device that could offer low operating power in-memory computation capabilities and can redefine von Neuman computer architecture while still exhibiting compatibility with the current CMOS fabrication infrastructure. In the past decade, resistive random access memory (ReRAM), has emerged as a promising candidate for the logic-in-memory devices due to its excellent properties such as scalability, low power operation, multiple logic states, fast switching and CMOS compatibility. In addition, quantized conduction in ReRAM devices arising due to size quantization, presents an opportunity to realize high-bit multilevel resistive switching for ultra-high-density memory systems with in-memory computation applications. For device showing quantized conduction, the conductance (<i>G</i>) of the device can be expresses as <i>G=nG₀</i>, where n is integer or half integer, and <i>G₀</i> is the fundamental unit of conductance (<i>G₀</i> ~ 77.5 µS). Here, we report reliable and repeatable resistive switching behavior in crossbar Cu/Ta₂O₅/Pt ReRAM devices based on electrochemical metallization (ECM) , wherein quantized conductance has been observe, investigated, and tuned in reset switching events.<br/>The Cu/Ta₂O₅/Pt devices in 8×8 crossbar geometry were fabricated on SiO₂/Si substrates using RF magnetron sputtering. Platinum bottom electrodes of thickness ~ 20 nm and line width and separation ~ 250 µm were patterned using a metal shadow mask, followed by ~100 nm thick Ta₂O₅thin film. After the deposition of the Ta₂O₅ layer, about 45 nm thick copper line electrodes as top electrode were patterned again using the shadow mask orthogonal to the bottom electrode, forming an array of 64 individual RRAM devices. The devices showed robust bipolar resistive switching, where the set switching voltage was found in the range of ~ 0.6-0.8V and reset voltage ~ (-) 0.5 to (-) 0.8V for 150 switching cycles at compliance current (<i>I_C</i>) of 200 µA. The mean resistances of the low and high resistance states were ~8.4 kΩ and ~828 kΩ, with variability of ~ 0.39 and 6.6 respectively. The reset currents were always found to be smaller than the compliance current confirming the ECM mechanism of the resistive switching, wherein the conducting filament (CF) is considered to have formed of Cu atoms which leads to low operating power per switching event. The variation in compliance current resulted in change in lateral dimension of the CFs leading to different low resistance states.<br/>Furthermore, integral and half integral quantized conduction (QC) sates were clearly observed in the crossbar Cu/Ta₂O₅/Pt devices, where the number and magnitude (<i>n</i>) of the quantized states were found to ~6 and 3.5, respectively at I_C of ~ 200 µA. The distribution of the conductance was Gaussian which were centered at1 G₀, 1.5 G_0,2 G₀, 2.5 G_0,3 G₀, 3.5 G₀. The magnitude and number of the QC states were found to increase from ~2.5 to 3.5 and from 4 to 6, respectively as the <i>I_C</i> increased from 50 to 200μA. The increase in number and magnitude of QC steps with I_C was explained considering the fact that thicker CF obtained at higher <i>I_C</i>_, when undergoes a gradual rupture during reset process, results in larger number of QC steps compared to a thinner CF. The diameter of the CF was calculated theoretically considering resistance contributions from Ohmic, Maxwell and Quantum point contact to the resistance of the low resistance states, and found to increase with increasing compliance current. In future work, the synaptic behaviour of Cu/Ta₂O₅/Pt devices for artificial learning will be explored via spike-time-dependent-plasticity, potentiation and depression characterisations. The observation of QC states in the ECM based crossbar Cu/Ta₂O₅/Pt device, and <i>I_C</i> controlled tuneability of magnitude and number may be useful for next generation high storage density resistive memory and electronic synapses for neuromorphic computing.