Research ArticleMATERIALS SCIENCE

Design of defect-chemical properties and device performance in memristive systems

See allHide authors and affiliations

Science Advances  08 May 2020:
Vol. 6, no. 19, eaaz9079
DOI: 10.1126/sciadv.aaz9079
  • Fig. 1 Permittivity values for pure and doped SiO2 as a function of OH concentration (water content) and dopant concentration.

    Highly pure SiO2 resembles water-free undoped Suprasil W. The permittivity values for the sputtered samples were measured in vacuum, at ~35% relative humidity (RH) and > 90% RH. We calculated the residual OH concentration in SiO2 films according to (42). This calculation procedure was also used for doped SiO2. For all materials, the permittivity is increasing with increasing OH (humidity) concentration, following linear relation εs=3.8073+2.72*1022cm3ion*N (42), where N is the concentration of OHdefects per cubic meter. The increase in Al content leads to a much higher polarizability of the material, whereas Ga doping has lower impact on ε. The concentrations and permittivities of the glasses Suprasil, Homosil, and Infrasil are extracted from (43), (44), and (42), respectively. The method for determining dopant concentration by dielectric measurements has a sensitivity even in 1-ppm range (fig. S4).

  • Fig. 2 Schema of the atomistic configuration in SiO2 amorphous films around a substitutional atom.

    (A) Al ion replaces a Si ion. Because of the electron configuration of the Al, the local charge surplus can bind Cu+/Cu2+ ions in the vicinity of the Al atom. (B) Ga substitution. Ga is forming on [GaO3]0 complex, and Cu ions interact only with the E′ center of the [SiO3]. This bond is weaker compared to that between Cu and [AlO4], resulting in a less pronounced polarization and smaller influence on the permittivity of the SiO2 matrix as function of the concentration.

  • Fig. 3 Potential and energy distribution in nanoelectrochemical memristors.

    (A) Symmetric cell with Pt electrodes. (B) Asymmetric cells with Cu and Pt as electrodes. Details on the calculation of the Debye length and the corresponding values are given in Supplementary text S3. (C) Charge separation and formation of ion-enriched and depleted zones in asymmetric memristive devices with two mobile-charge species and the corresponding electric field distribution [or SiO2:(Al,Cu)2]. In the particular case, the electric field drops only within the small zones of higher charge concentration. In case of pure material, the field will drop across the entire film thickness [see (B)]. (D) Schematic sketch of the energy landscape for pure and doped samples with and without applied external voltage. The difference between doped and pure SiO2 is presented by the change in the principle shape of the energy potential curve. Charge separation can additionally occur under applied external bias in cases that jump distances a1 and a2 and/or activation energies ΔGa1Ga2 for redox reactions and transport, respectively, differ. a.u., arbitrary unit.

  • Fig. 4 SET kinetics for SiO2-based devices.

    In the low-voltage regime, the SET kinetics are limited by the nucleation time, and the slope of the kinetics is steep. For higher voltages, the time-determining mechanism is strongly dependent on the doping and its influence on the switching properties. The undoped device (gray) show the typical trend for an ECM switching kinetic with a pronounced steep nucleation regime followed by a flatter slope representing a limitation by the ion movement and/or redox processes (45). For the highly doped SiO2:(Al,Cu)1, in devices with the highest permittivity, the kinetics are much slower and additionally limited by the RC time given by the high capacity and the shunt resistance in the setup. With a 1-megohm resistance, the switching speed is limited to times of around 1 ms at 5 V (blue); see also Supplementary text S6. When measuring with 50-ohm impedance, the kinetics are similar up to 2 V. Above 2 V, the kinetics with the low impedance are becoming faster than with high impedance. The SET kinetic shows a plateau at voltages lower than 2 V. We account this to the stronger interaction between the doped SiO2 (especially the AlO4− clusters) and the Cu2+ ions (formation of defect associations; see also Fig. 2) involved in the switching.

  • Fig. 5 Potentiation of SiO2:(Al, Cu)1 memristive device.

    Red curve is the current response to the input voltage signal (black curve). As long as the switching filament has not short-circuited the device, the capacity is loaded and discharges in every pulse and following delay time (first region, marked in blue). In the second region (yellow), the filament continues to grow. As long as charge is stored in the capacity, a driving force for further filament growth is present. In the event of a fully formed filament (third region, marked in green), the capacity is shorted, and no further charging/discharging can be distinguished. The device’s capacity plays a significant role in the potentiation process for neuromorphic applications influencing STP component, which can not only learn and forget in short time ranges but also transition to long-term memory. The electrical circuit during potentiation is present at the right side in the figure.

  • Fig. 6 Potentiation and depression response of SiO2-based memristive devices.

    (A) The pulse scheme applied to the cells. (B) Undoped SiO2, (C) highly (Al,Cu)-doped SiO2, and (D) (Ga,Cu)-doped SiO2. The delay between the pulses was varied between 10 μs, 1 ms, and 100 ms (from bottom to top) to simulate different presynaptic firing rates. A resistor of 100 kilohms connected in series limits the current to protect the devices. Insets are given where applicable to show the potentiation (black) and depression (red) in more detail. In comparison, the undoped SiO2 shows no stable potentiation with the chosen series resistor. The depression is dependent on delay times during potentiation. Short delays form more stable filaments, whereas long delays form “softer” filaments, easier to depress. The potentiation and depression kinetics are mirroring the SET kinetics observed for same memory cells as shown in Fig. 4. The experiment demonstrates how device performance and functionalities can be adjusted to the materials’ design.

Supplementary Materials

  • Supplementary Materials

    Design of defect-chemical properties and device performance in memristive systems

    M. Lübben, F. Cüppers, J. Mohr, M. von Witzleben, U. Breuer, R. Waser, C. Neumann, I. Valov

    Download Supplement

    This PDF file includes:

    • Sections S1 to S8
    • Figs. S1 to S9
    • Tables S1 and S2
    • References

    Files in this Data Supplement:

Stay Connected to Science Advances

Navigate This Article