Resistive switching in diamondoid thin films

The electrical transport properties of a thin film of the diamondoid adamantane, deposited on an Au/W substrate, were investigated experimentally. The current I, in applied potential V, from the admantane-thiol/metal heterstructure to a wire lead on its surface exhibited non-symmetric (diode-like) characteristics and a signature of resistive switching (RS), an effect that is valuable to non-volatile memory applications. I(V) follows a hysteresis curve that passes twice through \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I(0)=0$$\end{document}I(0)=0 linearly, indicating RS between two states with significantly different conductances, possibly due to an exotic mechanism.

www.nature.com/scientificreports/ This paper reports the RS we see in the current through a thin diamondoid film on a metallic substrate. The LRS and HRS for the film are typically Ohmic at room temperature, with resistances differing by seventy percent. While the similarity of the LRS and HRS is not ideal for NVM, it appears that the mechanism for RS could be unusual. We present our finding because it may interest those studying the fundamental physics of RS, and could be a first step towards NVM devices based on the observed effect.

Experimental
The I-V curves were studied using the cat whisker setup. The current through a cat whisker, which was the name of an early design of the semiconductor diode, passes from a semiconducting material to a thin wire contacting its surface. For a surface of diamondoid (adamantanethiol) films on Au/W substrate, the current I versus voltage V showed non-symmetric (diode-like) behavior and the signature of resistive switching.
There is a well-established procedure for fabricating diamondoid monolayers via self-assembly on the surface of gold and other noble metals (e.g. Ref. 38 ). Following this procedure, diamondoid films were prepared starting from a precursor of adamantane powder with a purity of 99.0% , via the self-assembled monolayer (SAMs) technique. A 5 × 5 mm 2 tungsten plate in ethanal was cleaned with an ultrasonic bath for 5 min at room temperature. Gold was then coated on the plate using plasma sputtering in an argon atmosphere at 8 Pa for 5 min. All SAMs were fabricated on Au/W substrates immersed in 10mM 1-adamantanethiol solution for 24 h. 10 mM 1-adamantanethiol solution was prepared by inserting 1-adamantanethiol 0.1683 g in 9:1 toluene/ethanol (volume:volume) with final volume of 100 ml. After 24 h, we took a sample out from the solution and washed it with toluene to remove the unbounded molecules. We rinsed it with ethanol and dried it with warm air. A crude, preliminary characterization of the adamantanethiol films was done by rinsing deionized water on the sample to confirm surface hydrophobic properties. Images of the sample surfaces, from atomic force microscopy (AFM) and scanning electron microscopy (STM), are included in the appendix below.
The cat-whisker setup (see Fig. 1) employed a source measure unit (SMU) connected to a 200 µ m diameter copper wire mounted on an adjustable microstage. The wire was moved towards the sample surface, by 10 µ m increments, until a stable current from the sample through the wire was observed under an applied bias potential V. (The voltage V is that between the sample surface and cat whisker, which dominates the effective resistance of the circuit). With electrical contact established, we swept V from −1.0 to 1.0 V and back repeatedly. The voltage increments V = 0.01 V were made with each time step t = 0.02 s, with a complete cycle taking 9.2 s (including the 3 ms used to ramp up by V each step). We measured the current I throughout this cycle to trace the I(V) curve.

Results and discussion
Figure 2 (top) shows resistive-switching in the I(V) curve for the self-assembled diamondoid film on the Au/W substrate. At low V, the current I(V) is linear in V but can have two different conductances depending on its recent history. When V is increasing, the low-resistance state (LRS) is observed with I = G 2 V and G 2 ≈ 0.25 −1 . When V is decreasing, the high-resistance state (HRS) with I = G 1 V and G 1 ≈ 0.15 � −1 < G 2 is seen.
Similar hysteresis curves were obtained in measurements on several samples. Note that the I-V curves shown do not include transient effects seen at the beginning of the measurement, which vary somewhat from sample to sample, but only the reproducible hysteresis loop that establishes itself after ramping the voltage up and down. (The first complete cycle was omitted for these plots, so the first measurement made was at V = −0.8 V and they continued for five cycles, with the result shown being the average of these five.) The effect can definitively be attributed to the diamondoid film because we repeated the experiment with a Au/W substrate without the adamantane. The result (see Fig. 5) was a current I = G s V linear in V over the entire range of Fig. 2 with a G s ≈ 0.4 −1 .
The data of Fig. 2 is a typical observation, and the conductance values are representative. The transition from HRS to LRS begins when V = V − ∼ −0.6 V and the transition from LRS to HRS when V = V + ∼ 0.8 V. The Figure 1. Schematic of the experiment used to measure I-V curves. A current I is driven through the sample, a diamondoid (adamantanethiol) film on an Au/W substrate, when the bias voltage V = V (t) is applied. The bias was ramped up and down in a sawtooth pattern, sweeping from −1.0 to 1.0 V and back in 9.2 s. To test the memory effect of the resistive switch, the applied bias was abruptly switched off, and V maintained at zero for time t s , before it was switched back to its previous value and the sawtooth variation V(t) resumed. www.nature.com/scientificreports/ hysteresis curve is robust in the sense that each individual loop lies close to the average shown. Also, RS requires both voltage polarities. The transition between resistive states occurs alternatively at high positive and negative V. If one polarity is used, for example if V is cycled between 0 and 0.8V while remaining positive, then no RS is seen. These features are illustrated in Fig. 6. According to the conventional picture of NVM devices based on RS in oxides, the survival of the LRS, which depends on intact conducting filaments, determines memory retention. We estimated the time scale over which the LRS remains intact by interrupting a sweep of V(t) at time t i when the system was in the LRS: disconnecting the potential then waiting for time t s before resuming the cycle at the point of interruption. In the results show in the bottom panel of Fig. 2 Fig. 3, we cycled V(t) until V(t) was V (t i ) = −0.5 V and rising then abruptly dropped it to zero. After a delay of t s seconds we resumed the cycle at V (t i ) . For the data shown in the lower panel of Fig. 2 we used t s = 5 s and repeated the sequence five times to obtain the I-V curve shown. We also measured the difference |�I| = |I(t i + t s ) − I(t i )| before and after the interruption, and plotted it versus t s in Fig. 3.

and in
As seen in the Fig. 2, the I-V curve resumed its previous course along the hysteresis loop after the voltage was turned back on. So conducting states remain distinct after the 5 s interruption. According to Fig. 3, there is no apparent change in the current (and thus no change in the LRS conductance) for t s < 150 s but I increases with larger t s , approaching 10% for t s ≈ 400 s. In the Appendix, Fig. 9, we show that this same time scale, of the order of a few minutes, is that over which the qualitative hysteresis loop starts to break down.
If the diamondoid films are indeed monolayers, then the conventional picture of extended conducting filaments is not applicable, and a tunneling picture is more appropriate. Since the field perpendicular to the film displaces charged defects, it can change trap energy levels relative to carrier bands and strongly affect the tunneling amplitude. We should assume parallel channels: direct tunneling from the substrate to the electrode alongside indirect tunneling via the diamondoid layer. If the latter is turned on and off by the applied field, this would explain the RS behavior and imply that the system could be modified to greatly enhance the G 2 /G 1 ratio. Below we follow this line of reasoning in writing down a model that can provide a semi-quantitative description of the observed hysteresis loop.
). An electric field can induce changes between the states. So, under an applied potential V at time t, a fraction n(V, t) of the sites are in state 1 while the remaining fraction 1 − n(V , t) are in state 2. This picture is reminiscent of a system with two molecular dipole states, where a strong electric field displaces the charge distribution within the molecule. The flakes of the diamondoid film are unevenly distributed on the substrate (see Fig. 8), covering an average fraction χ of its surface. If the cat whisker probe is larger than the typical flake, it will receive a weighted average of two current contributions: one coming through the diamondoid film and the other directly from the bare substrate. The latter can be associated with a conductance G s = 0.4 −1 . The total current is written as where the conductance of the diamondoid film may depend on time and V according to According to the data, the current satisfies Ohm's law at small V, with two different possible values of the conductance: I = G 1 V in the HRS and I = G 2 V in the LRS. Eq. (1) will give this behavior if one of the dipole states has an extremely small conductance and the other is constant over the voltage range of interest. So, we take G 1 (V ) ≈ 0 and G 2 constant. The simplest situation is one in which the fraction n(V, t) undergoes a complete transfer from 0 to 1 over the course of a full voltage sweep. In the LRS n(V , t) = 0 and With an assumed fractional coverage of χ ≈ 0.625 , we obtain the measured value G 1 = 0.15 −1 . The conductance of the active dipole state is set to G 2 = 0.16 −1 so that Eq. (1) gives the observed value of G 2 as well.
Both the shape of the hysteresis curve and its time dependence are determined by n(V, t). Note that one could modify the interpretation of the model by restricting current flow to a subset of molecular sites, perhaps defect locations, but this would not affect results below. Also, it is clear that this model always predicts zero current when V = 0 , so it cannot account for hysteresis loops that do not pass through the origin, reported elsewhere 45 .
We follow Kramers theory 47 , assuming an effective potential energy U(X) that depends on a reaction coordinate X, related to the dipole moment. Consider independent harmonic potentials for state 1, and state 2, where E 0 is a uniform electric field and q a charge. Assuming the effective dipole length ℓ is of the same order as the thickness of the interface, the potential difference between opposite sides of the film is V = −E 0 ℓ . The level shift U is constant, ω 1 and ω 2 are harmonic frequencies (with unit mass). We define their ratio as η = ω 2 /ω 1 , which is presumably of order unity. The basic energy scale ǫ 1 = ω 2 1 ℓ 2 /2 , the energy of a vibration with amplitude equal to the molecular length ℓ , is expected to be of order 1eV. If we measure ǫ 1 in eV, take ℓ = 1nm, and write for the effective molecular mass m = 2Zm p where m p is a proton mass, then we have ω 1 ≈ √ ǫ 1 /Z10 13 s −1 , a typical phonon frequency.
(1) www.nature.com/scientificreports/ Weakly coupling U 1 (X) and U 2 (X) gives a mixed ground state with potential shown at the top of Fig. 4. The minima are located at X = X 1 = −qV /(2ǫ 1 ) where the energy is U(X 1 ) = U 1 (X 1 ) , and at X = X 2 = ℓ − qV /(2η 2 ǫ 1 ) where the energy is U(X 2 ) = U 2 (X 2 ) . A local maximum, in the form of a cusp, presents an energy barrier at X = X b between the two valleys. The barrier location is found from . The relative energies of the two valleys, and the barrier between them, strongly affect the transition rates between states. The Kramers transition rate from state 1 to 2 is where β = 1/(kT) at temperature T. The reverse rate 2→1 is given by the same expressions with 2 ↔ 1 . The fraction n(t) = n eq + δn(t) of sites in state 1 changes in time according to: where = 1→2 + 2→1 . Thus n(t) relaxes towards its equilibrium value n eq = n eq (V , T) = � 2→1 /�. To simulate the conditions of the experiment, we change the voltage stepwise with time. (We take these stepchanges to be instantaneous, so we can still use Eq. (7) between steps.) If the voltage is changed to a value V j at time t = t j , then the population n = n(V , t) is given, according to Eq. (7), by ). www.nature.com/scientificreports/ where δn(t j ) = n(t j ) − n eq (V j , T) . This dependence continues until the next time, say t j+1 , that the voltage is changed to V j+1 , after which we have where δn(t j+1 ) = δn(t j ) + n eq (V j , T) − n eq (V j+1 , T) . The current can be obtained using Eq. (1) and calculated for any time t. At the beginning of the simulation we have to assign an initial population n = n(t = 0, V , T) . The natural choice is n(t = 0, V , T) = n eq (V = 0, T) , which is the state of absolute equilibrium that would be achieved by waiting for an infinite time at zero voltage and finite temperature. But the initial condition has no effect on the hysteresis loop shown, which is established after the first cycle for any initial population.
In the simulated I-V curves shown at the bottom of Fig. 4, the voltage was swept stepwise from −1.0 to 1.0 V and back). We also took T = 300 K, ǫ 1 = 2.7eV, η = 1.2 , U = 0 , q = 0.5e , ω 1 = 10 12 Hz. With these parameters, there is a complete population transfer at each end of the hysteresis loop, from n = 0 to n = 1 at positive V and from n = 1 to n = 0 at negative V. So, if we were to continue increasing the voltage beyond positive 1.0V (beyond negative −1.0 V), the curve would continue along the slope of the HRS (of the LRS).
It should be evident that this model has far too many parameters, most of which we only know to within a factor of unity, to take a quantitative fit between model prediction and experiment seriously. But it may be useful in providing an intuitive picture of the RS. For this purpose, it is helpful to consider the symmetric case η = 1 , U = 0 , ignore terms quadratic in qV /ǫ 1 , and take kT = 0 . The barrier is then at X b = ℓ/2 with height U(X b ) = ǫ 1 /2 + qV /2 and valley energies are U(X 1 ) = 0 and U(X 2 ) = qV . When qV > 0 , state 1 has a lower energy than state 2 but a bias qV > ǫ 1 is needed before the system can overcome the barrier and switch from state 2 to 1. Similarly, a reverse bias qV < −ǫ 1 is needed for transitions from 1 to 2. RS occurs because the system is trapped in alternating energy valleys as the voltage sweeps through V = 0.
To test non-volatile memory for such a model, we interrupted the procession along the hysteresis loop at a given time, t = t i and voltage V s = V (t i ) when the current is I(t i ) = G 2 V s by dropping V to 0. Then we continued to update the time-dependent current, using Eq. (1), for time t s before restarting V(t). But the transition rates 1→2 and 2→1 are exponentially small at zero voltage, being determined by a factor of order exp(−ǫ 1 /kT) . So, the current is unaffected over any reasonable t s and memories effectively last forever.
The t s -dependence of Fig. 3, which would limit memory retention, likely arises from effects far beyond this model. For example, alternative relaxation mechanisms between state 1 and 2 may dominate the direct transition given by Eq. (6) at V = 0 , whereas the latter is dominant at finite V and thus responsible for the hysteresis loop. In the simplistic picture offered by Eq. (2), the effective conductance of the diamondoid film must always be intermediate to the constant values of the pure LRS and HRS. The results of Fig. 9, on the other hand, indicate that the conductances of both states may be decreasing with time. We have not done an extensive study 46 of memory-retention properties or examined in detail the time-dependent degradation of the RS. The main conclusion we draw from experiment is that two-state conduction for this system, with conductances G 1 and G 2 in a robust hysteresis loop, persists for several minutes but afterwards becomes unreliable.

Conclusion
The electrical transport properties of diamondoid thin films, deposited on an Au/W substrate, were investigated experimentally. The I-V curves exhibit a hysteresis loop, with high-resistance and low-resistance states, characteristic of resistive switching. The observation of resistive switching in a putative monolayer, which has appealing material properties, is promising for applications to non-volatile computer memory. Though the microscopic mechanism for the observed RS is unknown, we provided a phenomenological model that captures the behavior of the measured hysteresis loop. www.nature.com/scientificreports/ the HRS could be associated with a non-Ohmic conducting channel. However, for typical averaged I(V) curves, the simplified picture given above is sufficient. Figure 7 presents atomic force microscopy (AFM) images of the top view of the substrate, with and without the diamondoid film present. The surface roughness for the diamondoid-coated samples, from Fig. 7d is a few nanometers. The adamantane molecule has a length of order 1 nm. Figure 8 presents scanning electron microscopy (SEM) images of the top view of the substrate with the diamondoid present. The dark regions, indicating flakes of the diamondoid layer adsorbed on the gold surface, have a typical length scale of a few micrometers along the surface.
The AFM and SEM data shown here do not establish that the thin diamondoid film is a monolayer. They merely give some indication of the deposition pattern on the substrate surface. www.nature.com/scientificreports/ Figure 9 provides a more detailed picture of the persistence of the NVM associated with the observed RS. As discussed previously, a hysteresis loop is observed when the voltage V is ramped up and down with time in a sawtooth pattern. To simulate powering down the device, we interrupted this time variation, dropped the voltage abruptly to zero, then waited for a time t s . After this, we restored V to its previous value and resumed the sawtooth variation.
In Fig. 3, we showed the difference in the current value, measured just before and just after the interruption, as a function of t s . The current change began to increase significantly for t s > 200 s or so. In the three panels of 9, we show the hysteresis loops observed after interruptions of various time t s . Here it can be seen that the qualitative shape of the I-V loop obtained after an interruption of t s = 40, 240 s appears unchanged from that observed before interruption (i.e. from those loops shown above). When t s = 360 s the hysteresis loop, though remaining well-defined, is evidently distorted from its previous shape. These results again point to a retention time scale of order t s 200 s. For the first two panels, the hysteresis curve is essentially unchanged from that observed before the interruption. However, for t s = 360 s, the distortion of the I-V curve is evident.