Mechanism Elucidation of Resistance Relaxation Phenomenon in Pt/Nb-doped SrTiO₃ Junctions for AI Applications

Resistive switching (RS) devices of a Schottky junctions consisting of a doped wide bandgap oxide such as SrTiO₃(STO) and a high work function metal such as Pt have been the subject of considerable interest in terms of both non-volatile and volatile resistance relaxation nature. When used as general non-volatile memory, the continuously tunable write/erase characteristics are advantageous for multibit application [1]. In this sense, resistance state retention characteristics are important and improving resistance volatility is crucial. On the other hand, when used in new computational paradigm for the breakthrough of von Neumann-type computing, controllability of resistance relaxation phenomenon is critically important. It has been reported that the resistance relaxation phenomenon after RS in Pt/Nb(0.5 wt%)-doped STO can be used to mimic synaptic plasticity for AI applications [2]. The development of nanoscale devices with this neuromorphic functionality is the basis for hardware implementation of artificial neural networks. For practical application, details of the resistive relaxation phenomenon is required to be understood. Schottky parameters (SPs) such as barrier height (SBH) and depletion width (<i>W</i>_D) can be determined by combining current-voltage (<i>I-V</i>) and capacitance-voltage (<i>C-V</i>) measurements. To date, this has been conducted only for two extreme states, high-resistance states (HRS) and low-resistance states (LRS) [3]. Changes in SPs with subsequent resistance relaxation have not been reported.<br/>In this study, sequential <i>I-V</i> and <i>C-V</i> measurements of a Pt/Nb:STO junction were performed on the same junction to extract time evolution of SPs during the relaxation. This is the first report on the time dependence of SPs as a function of resistance during the relaxation. <i>I-V</i> and <i>C-V</i> measurements were performed by the 2-terminal method using a semiconductor parameter analyzer and impedance analyzer, respectively. Since the resistance value changes logarithmically after RS, the switching unit was used to instantly switch to the circuit for <i>C-V</i> measurements after setting the junction to HRS or LRS using the circuit for <i>I-V</i> measurements.<br/>The relaxation phenomenon until 1000 s after RS was sequentially captured. All SPs such as SBH and <i>W</i>_D linearly depend on the logarithm of time as the change in resistance during the relaxation was. After setting to LRS, SBH was estimated to be changed from 0.57 to 0.63 eV and from 0.81 to 0.87 eV by the <i>I-V</i> and <i>C-V</i> measurement, respectively, showing an increase of 0.06 eV independently of measurement methods. Furthermore, the ideal factor <i>n</i> of the Schottky barrier was also decreased from 3.2 to 2.9 with increasing SBH. Considering the time dependence of <i>n</i>, the donor concentration (<i>N</i>_D) decreased from 7.5×10¹⁹ cm^-3 to 6.5×10¹⁹ cm^-3 and <i>W</i>_D increased from 19.0 nm to 20.8 nm after LRS. This change is considered to be caused by the trapping of electrons in the defects, which were once de-trapped by the application of forward bias to set the device to LRS. The revealed continuous change in the SPs suggests that resistance relaxation is brought about by electron trapping/de-trapping to/from oxygen defects rather than oxygen vacancy diffusion that likely gives discontinuity. This is consistent with the result of impedance spectroscopy measurement that no peak was observed in imaginary part of the dielectric constant for the frequency range of 10 Hz-5 MHz.<br/>The present study showed that time dependent SPs of metal/oxide junctions undergoing relaxation can be obtained by the simple method. We are now investigating the time dependence of SPs on doping concentration, electrode type, etc., to establish a method to control the relaxation characteristics of Pt/Nb:STO junction for AI applications.<br/><br/>[1] E. Mikheev, <i>Nat. Commun.</i> 5, 3990 (2015).<br/>[2] T. F. Tiotto <i>et al</i>., <i>Fron. Neurosci.</i> 14, 627276 (2021).<br/>[3] C. Park <i>et al</i>., <i>J. Appl. Phys.</i> 103, 054106 (2008).