Tunable Ion Hopping in Tantalum Oxide Resistive Switching Memory by Zirconium Doping

Artificial intelligence (AI) has received great attention for powerful performance over broad areas. However, conventional devices are unsuitable for multiply-accumulate operation (MAC), which is essential for the artificial neural network. This due to limitations in the architecture and device physics. In this context, in-memory computing technology is emerging. There have been numerous reports on establishing neural network with next generation memory devices, [1] but only few are fully hardware implemented due to imperfections of the devices. [2] Although resistive switching memory (RRAM) based on binary oxides is a promising candidate, it suffers from critical drawbacks: Stochastic switching process, sudden formation or rupture of conducting path. Inconsistent ionic motion in the switching medium is assumed to be related with the unreliable performance of the devices. Metal cation doping of binary oxides can facilitate or regulate the resistive switching process, by forming additional bonds with oxygen. We adopted doping as a method for tuning analog switching properties of RRAMs and studying the switching principles towards further understanding.<br/>We found that zirconium (Zr) doping effectively suppress sudden changes of cell conductance, which can be highly related with ionic hopping modulation. RRAM based on mixture of tantalum oxide (TaO_x) and Zr as a metal cation dopant was fabricated. We specified switching type of our devices to be filamentary, as local conducting paths with nanometer scale dimension were observed by conductive atomic force spectroscopy. The thickness of the switching medium is ~ 40 nm, confirmed by AFM. TaO_x RRAM exhibited gradual current hysteresis along voltage sweep of 0 ~ 5 ~ 0 V, resistance modulated from high resistance state (HRS) of 10 pA to low resistance state (LRS) of 5 nA, determined at 0.3 V read voltage. When voltage sweep of 0 ~ -5 ~ 0 V is applied, the devices return to HRS, showing bipolar resistive switching behavior.<br/>Doping concentrations of Zr in TaO_x is observed to be from 3.79 ~ 7.25 at %, determined by X-ray photoelectron spectroscopy. TaO_x RRAMs with 0, 3.57, 5.35 at % Zr doping concentrations exhibited similar HRS of 10 ~ 16 pA, whereas LRS significantly decreased (0 % : 5.78 nA, 3.57 % : 907 pA, 5.35 % : 283 pA). It is expected that Zr doping does not influence the intrinsic resistance of the cell, but takes role in the resistive switching process, regulating ionic hopping and formation of conducting filament. Incorporation of Zr in the TaO_x lead to multiple bonding with oxygen, where oxygen vacancies are the constituent of the conducting path.<br/>For artificial neural network applications, analog resistive switching properties of the devices were investigated by voltage pulse scheme. Zr doped TaO_x RRAMs exhibit gradual potentiation and depression of 200 conductance levels switching window within 0.1 ~ 1.4 nA. Gradual conductance modulation for 200 states is hardly achieved in previously reported RRAMs. [3] The devices exhibit low power consumption of micro-Watts, which is comparable to previously reported RRAMs.<br/>In conclusion, we realized in-memory computing unit with Zr doped TaO_x RRAM that meets two criteria, superior analog switching properties and simple cell structure. Metal cation doping of resistive switching memory opened a new method to study switching phenomena and proved the higher potential towards hardware implementation of neural network and AI.<br/><br/><b>&lt; References &gt;</b><br/><br/>[1] F. Cai <i>et al.</i>, “A fully integrated reprogrammable memristor–CMOS system for efficient multiply–accumulate operations,” <i>Nat. Electron.</i> 2, 290 (2019)<br/>[2] P. Yao <i>et al.</i>, “Fully hardware-implemented memristor convolutional neural network,” <i>Nature</i>, 577, 641 (2020)<br/>[3] T. Shi <i>et al.</i>, “A Review of Resistive Switching Devices: Performance Improvement, Characterization, and Applications,” <i>Small Struct.</i>, 2, 2000109 (2021)