Silver Nanofilament Formation Dynamics in a Polymer-Ionic Liquid Thin Film by Direct-Write

Abstract

Silver nanofilament formation dynamics are reported for an ionic liquid (IL)-filled solid polymer electrolyte prepared by a direct-write process using a conductive atomic force microscope (C-AFM). Filaments are electrochemically formed at hundreds of xy locations on a ^~40 nm thick polymer electrolyte, polyethylene glycol diacrylate (PEGDA)/[BMIM]PF₆. Although the formation time generally decreases with increasing bias from 0.7 to 3.0 V, an unexpected non-monotonic maximum is observed ^~ 2.0 V. At voltages approaching this region of inverted kinetics, IL electric double layers (EDLs) becomes detectable; thus, the increased nanofilament formation time can be attributed to electric field screening which hinders silver electro-migration and deposition. Scanning electron microscopy confirms that nanofilaments formed in this inverted region have significantly more lateral and diffuse features. Time-dependent formation currents reveal two types of nanofilament growth dynamics: abrupt, where the resistance decreases sharply over as little as a few ms, and gradual where it decreases more slowly over hundreds of ms. Whether the resistance change is abrupt or gradual depends on the extent to which the EDL screens the electric field. Tuning the formation time and growth dynamics using an IL opens the range of accessible resistance states, which is useful for neuromorphic applications.

Graphical Abstract

An ionic liquid (IL)-filled solid polymer electrolyte is presented with potential application in neuromorphic computing. Using a conductive AFM, direct-write of silver nanofilaments shows that a competition between IL EDL screening and nanofilament formation can be tuned as a control mechanism for nanofilament growth dynamics.

1. Introduction

Nanoscale metal filaments are required for applications ranging from optical metamaterials^{[1, 2]}to advanced electrical probes.^[3] One of the most studied applications is conductive bridge random-access memory (CBRAM), which is a type of non-volatile resistive random-access memory (ReRAM) wherein the electrical resistance of a solid-state dielectric is set to either a high or low state.^{[4, 5]} One approach to achieve the two states is to electrochemically form and dissolve conductive metal nanofilaments through the dielectric in a metal-insulator-metal structure (MIM).^[6,7] Although progress has been made in both organic and inorganic resistive memories, i]sues such as large device-to-device variability and poor reliability persist.^[8–10] To address the challenges in polymer based ReRAM, ionic liquids (ILs) – liquid salts that are commonly added to solid polymer electrolytes for various electrochemical applications^[11–15]–are being considered. In a recent report, ILs were added to polymer-based CBRAM to reduce the formation (set) bias and increase endurance;^[16] however, a detailed understanding of the role that ILs play in polymer-based resistive memory is still developing. Such understanding is important to uncover the microscopic mechanisms of conductive nanofilament formation and switching,^[7] and how material structure and composition can be tailored to tune performance.

We have previously reported on the formation and dissolution of silver nanofilaments in a new type of polymer electrolyte consisting of a UV-crosslinkable polymer, polyethylene glycol diacrylate (PEGDA), and an ionic liquid, l-butyl-3-methylimadozolium hexafluorophosphate ([BMIM]PF₆).^[17] The IL enhances uniform nanofilament formation kinetics under constant applied bias and enables modulus tuning over a range of a few hundred MPa to several GPa,^{[11, 12, 17]} making it more versatile than IL-free polymer electrolytes. Also, adding a sufficient amount of IL (> 10 wt%) enables 5× faster nanofilament formation compared to the IL-free electrolyte, but adding too much (50 wt%) slows the nanofilament kinetics by a factor of 7. These observations suggest complex dynamics, which further motivate investigation of the underlying mechanisms.

In this study, we focus on one IL concentration and study the impact of applied bias on nanofilament formation dynamics using C-AFM where the AFM tip serves as an inert mobile top electrode, while a sacrificial bottom electrode supplies the silver for electrodeposition (Figure 1). C-AFM enables the direct-writing of nanofilaments by precisely defining the nucleation sites, yielding more uniform nucleation and growth.^{[18, 19]} Using an automated script, nanofilament formation events at hundreds of xy locations on the film are recorded as a function of applied bias. This spatially dense sampling allows us to fully capture the stochasticity of nanofilament growth and enables statistical analyses of the data, which are particularly crucial for a polymer electrolyte containing microscopic heterogeneities.

Figure 1.

Schematic of the direct-writing of Ag nanofilaments in a polymer electrolyte thin film on a Ag substrate. A conductive AFM (C-AFM) controls the electrochemical formation and dissolution of Ag nanofilaments through the PEGDA/IL thin film between the AFM tip and a sacrificial Ag substrate.

The data reveal an unexpected relationship between formation times and applied bias. Specifically, the formation times do not decrease monotonically with increasing driving force – as expected for oxide-based dielectrics^[20–22]–but instead exhibit a pronounced maximum near 2.0 V. We interpret this behavior as the result of a competition between IL EDL formation and electrochemical filament growth. This competition can be used to control nanofilament morphology over multiple, well-defined resistance states – a control mechanism that should lend itself to the production and engineering of neuromorphic architectures where multiple distinguishable resistance states must be achieved to emulate the connections between neurons in an artificial neural network.^{[9, 23]}

2. Results and Discussion

2.1. Impact of applied bias on nanofilament formation kinetics

Formation times of all ^~500 nanofilaments formed at each bias over a range of 0.7 to 3.0 V are shown in Figure 2 for three IL concentrations. At all IL concentrations, the formation times decrease with increasing bias over the low bias range of 0.7 to 1.8 V. Specifically, for 70/30 PEGDA/IL wt% in Figure 2a, the average formation time decreases from 8.8 s at 0.7 V to 0.5 s at 1.8 V. The trend of decreasing formation time with increasing voltage is expected and can be explained by straightforward electrochemical kinetics. That is, the driving force for both electrochemical redox reactions and silver ion migration increases with increasing formation bias. However, unexpectedly, the average formation time increases abruptly when the bias is increased to just below 2.0 V. In addition to longer formation times, a more scattered distribution of formation times is also observed. For the 30 wt% IL sample, this distribution is highlighted in the insets in Figure 2a showing formation histograms at 0.7, 2.0 and 3.0 V. At voltages larger than 2.1 V, the formation times continue to decrease as expected, and the distribution returns to a standard Gaussian that is also observed for voltages less than 1.8 V. (For a complete set of nanofilament formation and dissolution histograms, see Figure S1, Supplemental Information, SI.) We refer to nanofilament formation kinetics near 2.0 V as “inverted” kinetics, where the average formation time first increases with increasing formation bias and then decreases again, causing unexpected non-monotonic behavior and giving rise to a peak in the average formation time vs. voltage plot. Inverted formation kinetics appear for all three IL concentrations at ca. 2.0 V, with the formation time generally increasing with increasing IL concentration.

Figure 2.

Nanofilament formation times as a function of applied bias for (a) 70/30, (b) 65/35 and (c) 60/40 PEGDA/IL wt% samples. 500 formation events were collected for all samples at each bias with the exception of data at 2.1 V for 65/35 %, where 339 events were collected. The average formation time at each bias is indicated by a horizontal marker and connected by the solid line. Insets in (a) are the histograms of formation time distributions at 0.7, 2.0 and 3.0 V, respectively for the 70/30 PEGDA/IL wt% sample.

2.2. The role of the electrical double layer (EDL) on formation kinetics

The observation that inverted kinetics at ^~ 2.0 V becomes more pronounced with increasing IL concentration motivates further investigation with an eye toward the possible role of the IL. One clue regarding the contribution of the IL comes from their use as high-capacitance gate dielectrics in electrolyte-gated transistors.^{[24, 25]} Here, electric double layers (EDLs) are formed by drift and diffusion of cations and anions to the surface of electrodes with opposite polarity. Specifically, when a positive bias is applied to the anode, cations drift to the cathode surface forming an EDL at the interface between the electrolyte and the cathode and induce an image charge in the electrode that is detected as a charging current. Simultaneously, an anionic EDL forms at the anode surface. In our case, we expect the cations in the IL (i.e., [BMIM]⁺) to accumulate near the C-AFM tip (cathode) when a positive formation bias is applied. Thus, one piece of evidence for IL EDL formation would be a nonzero current at the first measurement time after a positive voltage is applied followed by a temporal decay – assuming that EDL formation is faster than nanofilament formation. Figure S2 confirms this behavior, showing the current decay over 10 ms, which is orders of magnitude faster than nanofilament formation. This behavior is identical in form to the charging current associated with EDL formation in a parallel plate capacitor.^[26]

Up to half of the applied potential can drop across the EDL^{[27, 28]} (depending on the ion concentration and the geometry of the electrodes) leaving relatively little drop in the bulk of the electrolyte to drive ion migration. Thus, residual charge accumulated within any remaining EDL is likely to impact sequential nanofilament formation as the tip is moved from location to location. To address this issue, we modified the algorithm of the automation script to discharge residual EDL by grounding the tip for 100 s both before applying the formation bias at a new location, and after applying dissolution bias. The modified script was tested on the 65/35 PEGDA/IL wt% sample at 50 locations for each of six formation biases over two regions separated by > 100 μm (i.e., 25 points/region) in a 4.8 × 4.8 μm array, with a pitch of 1.2 μm. Formation times with and without the tip set to ground are shown in Figure 3. The data clearly show that grounding the tip decreases the formation time independent of bias. Notice that this approach does not eliminate the contribution from EDL formation during the measurement, but it eliminates any residual EDL that may persist from one location to the next.

Figure 3.

Filament formation times versus bias at 500 locations without grounding (blue, same data as Figure 2b) and 50 locations with grounding (orange) for 65/35 PEGDA/IL wt%; the average formation times are connected using a solid line (without ground) and dotted line (with ground).

Grounding the electrode has an especially strong impact on the formation times at voltages corresponding to the region of “inverted” formation kinetics. For example, the average formation time at 2.1 V is 20× faster with the grounding step, whereas it is only up to 3× faster at other biases. In contrast, at the highest applied bias of 3.0 V, filament formation is less affected by EDL formation as filaments are likely formed before the EDL screening can take effect. Therefore the grounding has relatively little effect on the formation times, as confirmed by the overlapping data at 3.0 V.

While grounding the AFM tip decreases the formation time, the distribution of times at 2.1 V remain widely scattered with a relative standard deviation of 94%, compared to ^~ 25% at 0.7 and 3.0 V. The results show that while discharging the previously formed EDL via grounding decreases the magnitude of the peak, it does not decrease the variability. Based on these results, we conjecture that there is an additional factor contributing to the inverted kinetics region, and we hypothesize that it is due to competition between EDL screening and silver ion electrodeposition. That is, the timescales of EDL formation and nanofilament formation are comparable at applied biases near 2.1 V, resulting in both longer nanofilament formation times and a wider distribution of formation times.

2.3. Quantifying IL EDL formation times as a function of bias

While the data in Figure 3 suggest that there may be important physics occurring at timescales shorter than those involved with nanofilament formation, it is difficult to decouple the electrical response of EDL formation from faradic electron transfer due to electrodeposition (Ag⁺to Ag(s)) in the nanofilament growth data. Thus, we prepared control samples in which silver nanofilaments cannot form by using a Au counter electrode instead of a sacrificial Ag electrode. One control sample, consisting of 65/35 PEGDA/IL wt%, was designed to isolate the effects of IL EDL formation, and a second control sample, consisting of pure PEGDA, eliminates IL EDL formation altogether. Current-time data were collected for 50 s at three locations (separated by more than 100 μm) at biases in the range of 0.7 to 3.0 V with grounding between measurements. These data for the pure PEGDA thin film are shown in Figure S3, and as expected, there is no detectable current above the noise threshold at any of the applied biases, because an EDL cannot form. In contrast, in Figure 4 we observe clear current signatures of EDL formation in the 65/35 PEGDA/IL wt% sample at biases greater than 1.8 V, where three measurements are made at each voltage. Specifically, when the bias is ≥ 2.1 V, up to 0.1 nA of current is captured on timescales ranging from milliseconds to seconds.

Figure 4.

Time-dependent current response at various biases for the 65/35 PEGDA/IL wt% sample on Au. At each bias, three consecutive measurements are plotted in three different colors. All of the current data are included on the plot, and the average of every 10 consecutive data points is extracted and plotted as a solid line to show the trend in the current through the noise. Note that the Y axis is linear. Dotted, vertical lines at each bias indicate ± one standard deviation of the average formation times from Figure 3 for the data that included grounding between measurements.

Because of the low signal-to-noise ratio in these extremely low-current measurements, it is not possible to quantify a time constant for the EDL response; however, we can use these data to identify the voltage at which the EDL contribution becomes detectable. This voltage could reflect the driving force required to overcome the barrier to diffusion for the IL ions through the heavily cross linked polymer chains. That is, while small silver cations may be able to easily traverse the crosslinked PEGDA, the much larger IL ions may only gain appreciable mobility once the voltage is sufficiently large (i.e., ≥ 2.0 V). In addition, the variability between the three measurements at each voltage for biases ≥ 2.0 V likely reflects the stochastic nature of IL EDL formation in the PEGDA/IL system. This is not surprising given the known complexity of IL EDL dynamics in parallel plate geometries.^[29]

Nevertheless, the EDL formation data in Figure 4 combined with the nanofilament formation data in Figure 3 provide insight as to why EDL formation is likely to have a stronger impact on nanofilament formation kinetics in the inverted region than at lower voltages. Note that the vertical dotted lines in Figure 4 indicate ± one standard deviation from the average formation times enabling the direct comparison of nanofilament formation timescales to EDL formation. At low biases, no appreciable EDL formation occurs and, therefore, there is no competition between silver nanofilament and EDL formation. However, at a bias of 2.1 V, where the inverted region is observed in Figure 3, the nanofilament formation time is comparable to the timescale over which the EDL current becomes appreciable. This result suggests that the silver nanofilament formation is competing with EDL formation, and the screening of the field by the EDL likely gives rise to longer formation times. Because EDL formation is disruptive to nanofilament formation, this competition could also account for the larger distribution of formation times at 2.1 V. In contrast to the inverted region, at a bias of 3.0 V the nanofilaments have already formed before the EDL has a chance to respond, giving rise to fast formation times with a tight time distribution.

2.4. Impact of formation bias on nanofilament growth dynamics

We have established how the nanofilament formation times at various locations and IL concentrations depend on bias, and how IL EDL formation can disrupt nanofilament formation. Here, we turn our attention to the nanofilament formation dynamics. Previous reports show how the final structure of a conductive metal nanofilament is strongly associated with its growth dynamics,^{[22, 30, 31]} which are further governed by kinetic and transport properties such as the redox reaction rate and metal ion mobility in the specific dielectric system.^[30] As the principal driving force for silver migration and electrodeposition in our PEGDA/IL system, the magnitude of the applied bias is expected to affect nanofilament growth dynamics, especially when a field-sensitive IL is present. Thus, we analyze the time-dependent current data and compare these results to direct imaging of the nanofilament structure as a function of bias.

Two types of time-dependent current data are detected during formation: abrupt growth and gradual growth, and both are shown schematically in Figure 5a. Here, we define the nucleation time (t_n) as the time for the current to reach a low, but detectable value of 0.5 nA, defined as the nucleation current (i_n). This current is ~100 times higher than system noise, defined as i_o. The growth time (t_g) is defined as the time required for the current to increase from i_n to the formation threshold (4 nA at a current sensitivity of 1 nA/V). The distributions of filament growth time (t_g) at different biases are plotted in Figure 5b, where we define t_g = 0.10 s as the boundary that distinguishes abrupt (t_g < 0.1 s) versus gradual (t_g> 0.1 s) growth.

Figure 5.

Time-dependent current data during nanofilament formation at various applied biases, (a) Schematic of abrupt (top) versus gradual (bottom) growth, (b) Growth time (t_g) distributions for filaments formed at different biases (c) All current versus time measurements are plotted at each bias. Note that the x-axis range is adjusted from plot-to-plot because the average formation time varies with voltage. Data density is represented by color, calculated by dividing the number of data points in each bin over the total number of data points for each bias. The scale of color bar is set from 0 to 50 × 10⁻⁴% and each plot contains 100 × 100 bins (i.e., the time and current scales are equally divided into 100 parts).

We extracted the t_g values in Figure 5b from the time-dependent current data leading up to nanofilament formation, which are shown in Figure 5c at six biases for the 65/35 PEGDA/IL wt% sample. Note that all current signals from hundreds of formation events at different xy locations are included in Figure 5c, where the color contours are used to represent the relative fraction of measurements with a specific current and formation time (i.e. the current data density). For nanofilaments formed at 0.7 and 1.4 V, the growth type is abrupt with an average t_g of 0.01 s; similarly abrupt transitions are observed at 3.0 V with average t_g of 0.03 s. The major difference at these voltages is the distribution of nucleation times (highlighted in the insets of Figure 5c): the wide t_n distributions at 0.7 and 1.4 V lead to low data density between i_n and i_f, whereas the narrower t_n distribution at 3.0 V increases the density of data as indicated by red.

In contrast, nanofilaments formed at 1.8 V show a more gradual growth (average t_g = 0.62 s), as shown in the t_g histogram in Figure 5b and as indicated in Figure 5c by the red data with a noticeable positive slope. When the applied bias enters the inverted region at 2.0 and 2.1 V, abrupt growth is again observed (average t_g = 0.03 and 0.02 s, respectively), but here it is accompanied by extremely scattered t_n. It is notable that the t_n distributions at biases outside the inverted region exhibit a normal distribution, while those in the inverted region do not.

Because the current at 1.8 V changes gradually with time after the current reaches i_n, it suggests the possibility of achieving multiple resistance states – a requirement for neuromorphic computing^{[9, 23, 32]}–during a single nanofilament formation under constant formation bias. To test whether or not the PEGDA/IL system can accommodate such a requirement, we set the system compliance to 100 nAand selected eight nanofilament “programming currents” ranging from 15 to 50 nA with a step of 5 nA. The formation bias was set to 1.8 V, and after each program current was reached and maintained for 20 ms, a read bias of 0.8 V was applied for 2 s while monitoring the conductance of the nanofilament. The data are plotted in Figure 6. The conductance initially decreases on the timescale of ms, consistent with the discharge of a portion of the EDL formed at 1.8 V, followed by a more stable current. While we cannot rule out a read disturbance at 0.8 V, the time constant for forming a nanofilament at 0.8 V is more than 50 times longer than the 2 s read time, and when the nanofilament does eventually form, the formation is abrupt and not gradual (Figure S4). In addition, a linear correspondence (R²= 0.95) between nanofilament resistance states and program pulses is observed – as indicated by the dashed line in Figure 6 – which is desirable for artificial neural networks.^{[23, 32]} In most cases, repeated programming pulses are used to controllably adjust resistance,^{[23, 32]} but Figure 6 strongly suggests that multiple resistance states may be achieved via the competition between the EDL formation and Ag redox reaction – thereby introducing a new mechanism to tune resistance.

Figure 6.

Conductance measured for 2 s at a 0.8 V read bias after nanofilament reached 8 programming currents (15, 20, 25, 30, 35, 40, 45 and 50 nA) at 1.8 V for 20 ms (65/35 PEGDA/IL wt%). Shaded bars indicate the locations where a 1.8 V formation bias was applied.

The nanofilament growth dynamics captured in Figure 5 are likely to result in different nanofilament morphologies; therefore, SEM was used to image the nanofilaments directly, producing both top-down and cross-section views (Figure 7). Four, 5 × 5 arrays of nanofilaments were created using 0.7, 1.4, 2.1 and 3.0 V nanofilament formation biases, respectively (4.8 × 4.8 μm square with a pitch of 1.2 μm). Surprisingly, the cross-sectional images in Figure 7 combined with C-AFM conductivity mapping (Figure S5) show that the majority of the nanofilament structure is confined below the surface of the polymer film. The plan view image makes it clear that nanofilaments formed at 2.1 V have the largest features among all arrays, followed by 0.7 V and 1.4 V, while only subtle features in the nanofilament array can be resolved at 3.0 V. The cross-section imaging shows nanofilaments formed at 0.7, 1.4 and 2.1 V all have obvious lateral buildup, especially those formed at 2.1 V. In contrast, no nanofilament structure formed at 3.0 V can be resolved through cross-section imaging.

Figure 7.

Plan view (left) and cross-section (right) SEM images of arrays of silver nanofilaments formed at (a) 0.7, (b) 1.4, (c) 2.1 and (d) 3.0 V. The locations of the cross-sections are indicated by red dashed line on the surface image. All scale bars are 1 μm.

Combining the time-dependent current data in Figure 5 with the SEM images of Figure 7 leads us to hypothesize a model of bias-dependent nanofilament growth shown schematically in Figure 8. At biases ≤ 1.4 V where abrupt nanofilament connections are made and for which nanofilaments are visible by SEM, the relatively weak vertical electric field results in more lateral build-up (i.e., thicker nanofilaments) before the final connection is made. The switching behavior shown in Figure S6 provides additional support for this interpretation in that it becomes increasingly difficult to dissolve nanofilaments with increasing switching cycles – consistent with thickening of the nanofilament during switching. While similar abrupt current transitions are also observed at 3.0 V, the vertical field is stronger, leading to faster and more directed (i.e., less lateral) growth. The hundreds of nanofilaments formed at this higher field remain switchable for at least 20 cycles (Figure S6).

Figure 8.

Schematic illustration of nanofilament formation dynamics at different applied biases. Gradual growth type (shaded red) is only observed at 1.8 V for the 65/35 PEGDA/IL wt% system. Silver atoms are represented by grey spheres. The width of the nanofilaments relative to the depicted atom sizes are purely for illustrative purposes and do not reflect the actual ratios.

In the inverted region, we measure a wider distribution of nucleation times, extreme lateral growth, and both gradual and abrupt formation events consistent with a competition between redox reactions and EDL formation that weakens the vertical electric field. Clearly, 1.8 V represents the threshold at which EDL formation impacts the growth kinetics in our PEGDA/IL system – changing them from abrupt to gradual by interfering with the electric field. Here, the nanofilament morphology also changes from thin (high resistance) to thick (low resistance) over longer timescales, and this gradual and controllable thickening could lead to enhanced switching characteristics. In fact, of the three measured voltages, the nanofilaments formed at 1.8 V retain the best switching characteristics among those tested (Figure S6). In addition, when the nanofilament does reform at 1.8 V during the switching measurements, the switching is also gradual (Figure S7), another characteristic which is required for neuromorphic applications.

3. Conclusion

C-AFM-based direct-writing of nanofilaments in a PEGDA/IL electrolyte reveals an unexpected relationship between formation times and applied bias. Specifically, the nanofilament formation time does not decrease monotonically with increasing driving force, but instead exhibits a pronounced maximum near 2.0 V, described as the “inverted region”. We interpret this behavior as resulting from a competition between IL EDL screening and electrochemical nanofilament growth. Time-dependent analysis of formation current over hundreds of formation events reveals two types of growth dynamics – abrupt and gradual. Gradual dynamics are observed at ^~ 1.8 V representing the threshold where EDL formation begins to interfere with nanofilament growth, changing it from abrupt to gradual. Here, multiple resistance states can be accessed within a single nanofilament, due to the gradual growth. The voltage-dependent dynamics are also correlated with filament structure: nanofilaments formed in the inverted region have more lateral and diffuse structures than those formed outside the voltage window. The results suggest that by tuning the competition between the IL EDL and electrochemistry, the growth dynamics and nanofilament morphology can be tuned. This control mechanism could potentially be exploited in applications such as artificial neural networks, where multiple, distinguishable resistance states are required. Further optimization of this electrolyte is expected to extend the accessible range of resistance states to smaller write/read biases, making the PEGDA/IL electrolyte a highly customizable system and potential candidate for organic electronics requiring low-power neuromorphic computing operation.

4. Experimental Section

Sample preparation for C-AFM measurements.

Poly(ethylene glycol) diacrylate (PEGDA, Mn=2,000), 1-butyl-3-methylimadazolium hexafluorophosphate ([BMIM]PF₆, ≥98.5%), 2-Hydroxy-2-methylpropiophenone (HMPP, 97%) and anhydrous acetonitrile (ACN, 99.8%) were purchased from Sigma-Aldrich. All reagents were used as received without further purification. Silicon wafers (University Wafer, P/Boron, 500 ± 15 μm) were cleaned by sonication in acetone followed by a 2-Propanol rinse and N₂ blow dry. 5 nm/100 nm of Ti/Ag were deposited by electron-beam evaporation at 5 × 10⁻⁷mbar base pressure (Plassys, MEB 550 S). The following steps were completed inside an argon-filled glovebox (Mbraun, MB-200B) where O₂ and H₂O < 1 ppm. [BMIM]PF₆ (IL) and PEGDA were dissolved in ACN and combined to prepare a total of 3 samples with PEGDA/BMIMPF₆ compositions of 60/40, 65/35 and 70/30 wt%. In all 3 samples, the PEGDA concentration was 1 wt% together with 0.02 wt% of HMPP (photoinitiator). The polymer electrolytes were spin-coated on the Ag-coated silicon substrate at 4000 rpm for 30 s and annealed at 80 °C for 2 mins to drive-off ACN. Samples were photo-crosslinked with a UV lamp (UVP Compact UV Lamp, λ = 365 nm, P = 1.3 mW/cm²at 3 inches) at a working distance of 0.5 inch for 1 hour.

C-AFM characterization.

A Bruker Dimension Icon AFM was operated in contact mode using a conductive AFM tip (SCM-PIT-V2, Pt-Ir coating, k = 4.0 N/m). A custom script was created to modulate the location and potential applied to the tip to: (1) move the tip from point-to-point in a raster scan pattern at a preset spacing; (2) apply the desired voltage; (3) measure the current between the conductive AFM tip and the chuck. The AFM tip was grounded, and the voltage was applied to the chuck. Electrical contact was made between the AFM stage and the Ag layer of the substrate using copper tape (McMaster-Carr). At each point, the chuck voltage was set to a positive formation voltage until either (1) the current increased above a set formation current threshold for 2 consecutive data points (2 milliseconds) to ensure nanofilament formation while avoiding overgrowth, or (2) the maximum time window was reached, indicating no formation. In either case, the voltage was then switched to a dissolution voltage of −0.5 V until the current (now of the opposite sign) either (1) decreased in magnitude below a set threshold, or (2) reached the maximum dissolution time window meaning that the nanofilament did not dissolve. Formation and dissolution current thresholds were chosen as +4.0 and −0.5 nA, respectively and both the maximum formation and dissolution time windows were set at 50 s. The compliance current of the instrument is ca. 5 nA at a current sensitivity of 1 nA/V, which is selected by the user.

In this work we focused on electrolytes with IL concentrations of 30, 35 and 40 wt%. This IL concentration range was chosen because they have faster formation kinetics compared to 50 wt% IL, and a more homogeneous structure (i.e., less obvious crystal features) compared to a 10 wt% IL sample.^[[17]]The thickness of the 35 wt% IL was ^~40 nm by FIB-SEM (Figure 7), and we expect the 30 and 40 wt% IL samples to be of similar thickness as our previous report indicated no thickness variations in the range of 10 to 50 wt% IL.^[[17]]Silver nanofilament formation times were measured over a bias range from 0.7 to 3.0 V, with ^~100 nanofilament formation events captured at ^~5 distinct regions on each sample for a total of 500 measurements. Within each region, nanofilaments were formed and dissolved at different locations forming a square array with a pitch of 1.2 μm.

Characterization of silver nanofilament structures.

Arrays containing 25 nanofilaments were formed over 4.8 × 4.8 μm squares on the 65/35 PEGDA/IL wt% sample using 4 different biases (0.7, 1.4, 2.1 and 3.0 V). A focused ion beam (FIB, FEI Scios DualBeam) was used to create the cross section of PEGDA/IL film at locations that contained nanofilaments, and scanning electron microscopy (SEM, FEI Scios DualBeam) was used to image the cross-section.