Barmpatsalos, Nikolaos; (2025) Investigation of ReRAM Devices for Emerging Memory and Neuromorphic Computing Applications. Doctoral thesis (Ph.D), UCL (University College London).

Preview	Text Nikolaos Barmpatsalos - 17143404 - PhD Thesis - Final Final.pdf - Accepted Version Download (13MB) | Preview
Preview	Text Barmpatsalos_10215639_Declaration Form_sigs_removed.pdf Download (293kB) | Preview

Abstract

Resistive random-access memory (ReRAM) based on silicon oxide (SiOx) is a promising candidate for next-generation non-volatile memory due to its simple structure, scalability, and CMOS compatibility. This thesis investigates the physical mechanisms, resistance switching (RS) behaviour, and modeling of SiOx-based and sodium bismuth titanate (NBT)-based ReRAM devices, combining electrical and structural characterisation, device engineering and programming, and compact modeling. The effect of device design and sputtering conditions on pristine, electroforming, RS electrical behaviour and scaling potential was studied. Increasing the SiOx thickness increased the electroforming voltage (and thus current overshoot) which subsequently influenced the RS behaviour and variability. Surface roughness at the bottom electrode Mo/SiOx interface impacted filament formation, likely by influencing the oxide’s columnar microstructure. Additionally, higher RF power lead to higher electroforming voltage, whereas higher Ar:O₂ ratio lead to lower electroforming voltage and pristine resistance—attributed to changes in film density and oxygen incorporation during growth. Intermediate resistance states (IRS) were programmed using voltage sweep, constant voltage bias, and voltage pulses, with constant bias yielding the smoothest state transition. The non-linearity of IRS was probed and a simple multi-level cell (MLC) scheme was proposed. Variability remained a key challenge for reliable multi-bit operation, but suitable software mitigation methods were presented. A novel hybrid filament/interface-type switching NBT-based ReRAM was investigated. It demonstrated fast switching speed, better endurance, and lower variability than SiOx-based ReRAMs. Long-term potentiation/depression (LTP/LTD) programming was explored, proving the suitability of NBT devices for neuromorphic computing applications. Finally, the Memdiode model was modified to capture the RESET current plateau feature observed in SiOx-based devices. A constant characteristic time of the state variable progression successfully replicates this gradual behaviour between HRS, LRS, and IRS. The model modification provides a foundation for circuit-level simulation of gradual switching dynamics, valuable for neuromorphic applications.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item