Electronic Supplementary Information for GHz operation of nanometer-scale metallic Ag2S memristors

Ag thin films with a nominal thickness of 80 nm were vacuum evaporated onto a Si substrate. The thin Ag2S layers were grown by depositing sulfur onto the Ag surfaces in a clean environment. First, analytic grade sulfur powder was loaded in a quartz tube, melted and cooled back in order to ensure a homogenous source. The thin film sample was then loaded in the tube to a distance of 2 cm from the sulfur. After loading both the sulfur and the sample, the tube was evacuated to 10−5 mbar pressure. Then the temperature was ramped up to 60 ◦C and the sublimation of the sulfur was performed in a static vacuum for 2–10 minutes. Finally, the temperature was rapidly ramped down. The samples were characterized by He-RBS (Rutherford Backscattering Spectrometry) and ERDA (Elastic Recoil Detection Analysis) [1]. They exhibit inhomogeneous sulfur concentration profiles consistent with the presence of an Ag2S surface layer [2]. Nanoscale contacts were created by gently touching the sample surface with a mechanically sharpened PtIr or Nb tip. For coarse adjustment a screw thread mechanism was used, whereas for the fine positioning a three dimensional piezo scanner was applied. Using this technique numerous contacts were created with reproducible I-V characteristics. The bias voltage was applied to the junctions by utilizing the analog outputs of a National Instruments data acquisition card. The output voltage was divided and filtered in order to ensure low noise on the contact. The current was measured using a variable range I-V converter and processed by the data acquisition card. The current-voltage characteristics of the junctions were recorded during repeated bias voltage sweeps. The bias scheme of the I-V measurements carried out with a superconducting Nb tip at 4.2 K is shown in Fig. S1(a). The subsequent resistive switchings are obtained by voltage sweeps of 800 mV amplitude and alternating sign while the ON and OFF state resistances as well as the non-linear current contributions induced by the superconducting tip are detected by low amplitude (±10 mV) triangular voltage pulses. The first high bias sweep is performed in order to verify the switching and to initialize the device in its OFF state. This is followed by the acquisition of the low bias IV data repeated 5 times for averaging purposes in order to further improve the signal to noise ratio. The next high bias sweep covers one and a half of the total hysteresis loop preparing the device in its ON state before the second sequence of the low bias measurements probing the ON state transport properties are executed. In order to confirm that upon the first OFF to ON switching and ON state measurements the device preserved its reversible configuration, we restored and tested the OFF state again by applying another high bias loop and a low bias sequence, respectively. The reproducibility of the hysteretic high bias I-V traces shown in Fig. S1(b) demonstrate that the two dominant device configurations determining the ON and OFF states did not change over the above sweeping sequences. Fitting the numerical derivative of the low bias I-V data displayed in Fig. S1(c) against the BTK theory provides information about the effective transmission values of the corresponding configurations. The numerical accuracy of the fitting procedure on T is illustrated in Fig. S2. After obtaining the best fitting transmission values by a numerical least square method, T was detuned from its optimum to the values indicated