Ferroelectric Domain Wall Memristor

Abstract

A domain wall-enabled memristor is created, in thin film lithium niobate capacitors, which shows up to twelve orders of magnitude variation in resistance. Such dramatic changes are caused by the injection of strongly inclined conducting ferroelectric domain walls, which provide conduits for current flow between electrodes. Varying the magnitude of the applied electric-field pulse, used to induce switching, alters the extent to which polarization reversal occurs; this systematically changes the density of the injected conducting domain walls in the ferroelectric layer and hence the resistivity of the capacitor structure as a whole. Hundreds of distinct conductance states can be produced, with current maxima achieved around the coercive voltage, where domain wall density is greatest, and minima associated with the almost fully switched ferroelectric (few domain walls). Significantly, this "domain wall memristor" demonstrates a plasticity effect: when a succession of voltage pulses of constant magnitude is applied, the resistance changes. Resistance plasticity opens the way for the domain wall memristor to be considered for artificial synapse applications in neuromorphic circuits.

Keywords: ferroelectric domain wall; memristor.

Figure 1

a) Vertical piezoresponse force microscopy map of a 5 × 5 µm² region of the 500 nm thick ion‐sliced lithium niobate single crystal, showing the combined vertical amplitude and phase response of the domain structure written using a stationary biased atomic force microscopy tip. b) Conductance map of the same region with −2.3 V DC bias applied to the bottom electrode. c) A plot of current against domain diameter, as measured at the top surface of the LiNbO₃ thin film. Points represent mean currents obtained from the 100 datapoints at each domain size which showed the largest current values (error bars mark two standard deviations). d–f) Phase‐field simulation of d) domain evolution during tip‐induced switching in a 500 nm LiNbO₃ film and corresponding e) electrostatic potential and f) electron density, estimated by using a degenerate electron gas approximation.

Figure 2

a) Schematic showing a tungsten carbide needle electrically contacting microscopic Ga–In–Sn eutectic alloy electrodes to allow switching and conductance measurements at the mesoscale. The three images to the right are (top‐to‐bottom) typical current, piezo‐response amplitude and phase across the boundary between a partially switched and completely unswitched region. The fine‐scale microstructure and localized conducting domain walls are evident. The scale bar for the maps is 1 µm. b) 3D schematic showing typical domain structure in a lamella (arrows represent polarization directions) used for transmission electron microscopy investigations. c) Scanning transmission electron microscopy high angular annular dark field image (STEM‐HAADF) of a lamella cut from a partially switched region. The high density and angle of the domain walls can clearly be seen. The variations in contrast are associated with variations in the microstructure that occur through the thickness of the lamella: dark areas are associated with a uniform domain state (either “up” or “down”) throughout the entire lamellar thickness (cream regions in the schematic shown in (b), while lighter areas are associated with both “up” and “down” domains and the domain wall being sampled by the electron beam in the through‐thickness direction (in the schematic shown in b, these would be areas in which the blue domain wall is sampled in the through‐thickness direction). Scale bar 50 nm.

Figure 3

a) Three different conduction states induced during initial switching studies, spanning twelve orders of magnitude in resistance. The current measured for the R₃ state hit the compliance limit of the source‐measure unit, implying that even greater resistance changes during switching are likely to occur. b) Steady‐state current (measured at 10 kV mm⁻¹) in a high conductance state (yellow/orange) compared with the equivalent switching current based on the maximum possible switching charge discharged over the 900 s measurement period (purple). Clearly, the dc currents observed cannot be accounted for by domain wall movement. c) A schematic hysteresis loop compared with the conductance measured as a function of switching field. Currents were measured at 5 V (equivalent field of 10 kV mm⁻¹) applied for 1000 ms and the switching field was applied stroboscopically (in between each dc current measurement) in 10 ms pulses. Arrows indicate the relation between the current peaks (states of highest conductance) and the coercive field in the P–E loop. Note the slight imprint which causes off‐centring of the P–E response and an asymmetry in the switching fields needed to generate the most highly conducting states.

Figure 4

a) Conductance variations with time when switching field pulses (1000 ms duration) of progressively increasing magnitude (in 4 V steps) are applied every 10 s: in this case, at least ten different stable remnant states have been demonstrated within an order of magnitude change in conductance b) Finer steps of increasing pulsed switching voltage, varied between 0 and 50 V (in 500 mV steps), generated 100 different conductance states within a two orders of magnitude window in conductance. For each state, the time dependence of the conductance is presented, monitored for 10 s after the completion of the switching pulse used to create it. Inset: higher resolution plot of the states created by switching pulses at the upper end of the voltages used.

Figure 5

a) A plot of current measured under a 5 V steady‐state potential difference (10 kV mm⁻¹ field) against the number of 10 ms pulses applied, with the voltages of these pulses as labeled. The domain state was reset before each set of measurements made with a given pulse voltage magnitude. b) A 3D plot showing steady‐state currents as a function of the cumulative number of pulses applied, for different pulse voltage magnitudes. The double peak features that emerge mirror those in Figure 3c, but here the plasticity effects are clearly evident. c) Shows (b) from a bird's‐eye view. Colored dots indicate the switching pulse fields associated with data in (a).