4.3 Resistive switching in asymmetrically shaped
Figure 4.6: a) Current-voltage traces of the voltage ramp cycles during the elec-tromigration of asymmetrically shaped silver nanocontacts. The measurement was performed in vacuum at room temperature. b) Evolution of the device conductance during the electromigration process (left panel) and during the self-breaking (right panel). c) I-V characteristic of the tunneling contact after the self-breaking (black dots). By tting the Simmons model (red line) the main parameters could be ex-tracted (annotation). d) A representative hysteric I-V trace of the nanofabricated Ag-Ag2S-Ag nanojunction. The inset shows the SEM image of the device. The white scale bar in the lower right corner indicates 200nm and the convention of the bias voltage polarity is also shown. Positive voltage means positive biasing of the bottom at part. e) 20 consecutive I-V traces acquired on stable nanojunction congurations.
ROF F = 279±15 Ω, RON = 155±5 Ω [2].
I(V) curve of the formed Ag2S memristor. It is highly similar to the ones obtained in the STM and MCBJ setups. The switching direction, indicated by the arrows, agrees with the expectations of asymmetry induced lament evolution, i.e. the set transition occurs at positive voltage. The formation and destruction of the conductive
channel may take place at a few 10nm long region of Ag2S. The device showed about 150 switching cycles with 1.5 OFF/ON resistance ratio. Afterwards it turned into unresolvable high resistance state. Figure 4.6.c shows 20 consecutive I-V traces, which indicates high reproducibility both in ON and OFF resistances. The switching occurs between RON = 155±5 Ω and ROF F = 279±15 Ω.
The measurements in Figure 4.6 are considered as the rst proof of principle results on asymmetrically shaped on-chip Ag-Ag2S-Ag resistive switches. At the same time these are the rst results in our lab, where resistive switching was achieved in an on-chip environment instead of the STM-based memristor junctions. These devices had great mechanical stability, the tunnel contact could be sustained for hours, while in the former STM arrangement the contacts were stable only for few minutes [184]. Therefore the devices showed good resistance reproducibility in the consecutive switching curves, however, the poor endurance, vanishing of switching ability after few 100 of switching cycles, refers to the intrinsic instability of the active region.
After the rst set of devices our attention turned into the samples fabricated with shadow mask. Applying mask would greatly simplify and make more reliable the sulfurization process. However, most of the samples exhibited only weak eect or nothing when they were exposed to the sulfur regardless of the mask structure. In order to get feedback about the structural changes of the samples during the sulfur-ization, the resistance was measured in-situ. Figure 4.7 shows the two typical in-situ resistance measurements performed on two dierent devices on the same chip. It was a general observation that these samples reacted with the sulfur much less sen-sitively than the rst set of devices without mask. The rst type of samples (Figure 4.7.a) did not show any reasonable resistance change even after 30 minutes and they did not show resistive switching behavior. On the other hand the device in Figure 4.7.b exhibited a slow resistance increase and nally a sudden jump was observable.
Removing from the sulfur rich environment at this point, the resistance remained in the range of few kΩ. However these devices mostly showed unstable resistance jumps with large ON/OFF ratio, presented in Figure 4.7.c-d. The switching direction was not well determined. These observations suggest that the cleanness of the silver plays crucial role in the reliable device fabrication. A thin passivation layer on the surface can hinder the formation of Ag2S region at the constriction. The slight dierence in the surface contamination can cause the large variety of device behavior during and after the sulfurization.
Devices without any mask were fabricated again by myself to test the repro-ducibility. Using subsequent electromigration and sulfurization processes resistive switching could be induced, but it has very low yield presumably due to the surface
Figure 4.7: In-situ resistance measurements during the sulfurization of as-prepared silver devices using PMMA mask. In most cases the resistance of the samples either a) did not show any changes or b) a very slow one in the sulfur rich environment. c-d) Representative hysteric I(V) curves with high ON/OFF ratio. The arrows indicate the switching direction, which is opposite as the geometrical asymmetry induced case.
contamination. However, neither in stability nor in reproducibility of resistances they did not exceed the prototype. Figure 4.8 shows 10-10 switching cycles of the two most stable devices. The switching direction, indicated by the arrows, agrees with the previous results. However, after 40-60 cycles the switching behavior disap-peared, the hysteresis of the I(V) traces closed. We had to increase the amplitude of the I-V measurement to sustain the switching eect, but the resistance of the contacts increased continuously and nally their switching capability failed.
Beside the ECM induced resistance transitions, atomic switching was also ob-served in sulfurized silver contacts. As it was presented in right panel of Figure 4.1 if the size of the junction is in atomic scale the resistance changes in discrete jumps.
Figure 4.9 shows 20 consecutive I-V curves of an atomic switch device. Resistance jumps occur at wide range of voltages and the device preserves its new resistance state until the back-switching at the opposite polarity [183, 187]. The switching di-rection of a specic atomic switch is xed, but varies randomly for dierent junctions
Figure 4.8: 10 consecutive I-V traces of the two stable Ag2S based memristors. The re-sistance transition occurs between a) RON=2.8 kΩ, ROF F=4.1 kΩand b) RON=150Ω, ROF F=190Ω. The arrows indicate the switching direction.
according to the local geometry [183]. The atomic switches had outstanding stability, more than thousand switching cycles could be measured without any degradation.
This observation is in contrast with the thermal instability of pure silver nanowire during the electromigration, presented in Figure 4.6. The much higher stability of the atomic switch may arise from the stabilization eect of the Ag2S matrix. It conrms that the low endurance of the on-chip Ag2S memristor devices is not the consequence of the poor mechanical stability, rather the intrinsic instability of the switching process.