a In te n s ity [a.u.] c In te n s ity [a.u.] 6 4 2 4 3 2 1 3 2.5 2 1.5 1 p O 2 3.5 1,5 3, 4,5 R a m a n s h ift [c m -1 ] p ris tin e 1 o C 2 o C 3 o C 4 o C 5 o C b d In te n s ity [a.u.] In te n s ity [a.u.] 6 4 2 1, 1,2 1,4 1,6 1,8 2, R a m a n s h ift [c m -1 ] 5 4 3 2 p O 2 3.5 3 2.5 2 1.5 1 p ris tin e 1 o C 2 o C 3 o C 4 o C 5 o C 1, 2, 3, 4, R a m a n s h ift [c m -1 ] 1 1, 1,2 1,4 1,6 1,8 2, R a m a n s h ift [c m -1 ] Supplementary Figure 1. Visible (λ = 633 nm) Raman spectra of a-co x layers. (a) Raman spectra of a-co x deposited at increasing p O2. (b) Raman spectra of a-co x layers (p O2 = 1.5 µbar) after 5 min of rapid thermal annealing in vacuum (1 5 mbar) at temperatures increasing from 1 C to 5 C. 1
a 1 4 b 1 4 Set time [ns] 1 3 1 2 1 1 Reset time [ns] 1 3 1 2 1 1 1 1 2 3 4 5 6 7 Set voltage [V] 1 1 2 3 4 5 Reset voltage [V] Supplementary Figure 2. Dependence of SET and RESET times on the voltage amplitude. Time width plotted as a function of the voltage amplitude of pulses able to SET (a) and RESET (b) a-co x memory cells. Note that during SET, voltage pulses with positive polarity are applied to the bottom electrode, whereas during RESET, voltage with negative polarity are applied to the bottom electrode. Both SET and RESET require longer pulses for lower pulse voltage amplitudes. This result is also suggestive of the role played by the electric field in the switching process. When the voltage amplitude and, consequently, the electric field is lower, ionic migration is lower and hence it takes more time to SET. During RESET, in addition to ionic migration, Joule heating plays a more significant role, and this could explain the higher speed of the RESET with respect to the SET. 2
a 35 3 25 b -1 Current [µa] 2 15 1 Current [µa] -2-3 5-4 -5 2 4 6 8 1 Time [ns] c 1 2-5 2 4 6 8 1 Time [ns] 1 Current [µa] 1-2 1-4 -1 -.5.5 1 Voltage [V] Supplementary Figure 3. Multiple switching cycles. Nine consecutive (a) SET and (b) RESET pulses are shown to illustrate the variability. The current rising peak during RESET is commonly observed. This is most likely due to the low resistance of the SET state. Once RESET occurs, the current reduces to a lower value that corresponds to the current that will flow through the RESET state when such a high voltage is applied. (c) Also shown are the low-field I-V curves that correspond to the SET and RESET states. The pulses exhibit a certain amount of variability despite of their qualitative similarity. The variation in the I-V is indicative of the differences between the various RESET states. This type of variability is typically observed in almost all resistive memory devices. 3
1 7 @ t = @ t = 6 m o n th s R e s is tiv ity [Ω m ] 1 5 1 3 1 1 1 2 3 [µb a r] p O 2 Supplementary Figure 4. Time stability of the electrical characteristics of a-co x. The electrical resistivity (ρ) was first measured for different values of p O2, and then once more after 6 months for p O2 = 1 µbar and 1.5 µbar. The resistivity is estimated at.4 V from DC measurements carried out on 6 memory cells with diameters of 1 µm, 2 µm and 4 µm. The measurements show that the resistivity values did not change substantially. Error bars represent the standard deviation. 4
Supplementary Figure 5. Cross-sectional TEM images of a typical memory device. The a-co x -based cell shows the spatial distribution of the different layers. A high-resolution TEM image of the a-co x cell is depicted in the inset confirming that the cell is fully amorphous in the pristine state with no significant interfacial WO x. Scale bar, 5 nm. Scale bar inset, 1 nm. 5
1 2 C u rre n t [µa ] 1 1-2 1-4 1-6 2 4 V o lta g e [V ] Supplementary Figure 6. Forming process in a-c:h devices. Representative forming curves in a-c:h devices. After forming, these devices go into a SET state. This low-resistance state is attributed to Jouleheating-induced clustering of sp 2 hybridized carbon atoms 1 3. Temperature-dependent transport measurements on these low-resistance states show remarkable similarity with the SET state associated with a-co x devices. 6
1 4 R e s e t c u rre n t [µa ] 1 3 1 2 P t W T i E le c tro n a ffin ity Supplementary Figure 7. RESET current versus top electrode material. RESET current corresponding to devices with Pt, W and Ti top electrodes (W serves as the bottom electrode). The RESET current reduces with increasing electron affinity of the top electrode material. Each data point represents the mean of ten measurements and the error bars represent the standard deviation. 7
Resistance [Ω] 1 6 1 5 1 4 1 3 1 2 W/aCO x /Pt W/aCO x /W W/aCO x /Ti 1 1 Pt W W W 1 2 3 4 Ti W W Supplementary Figure 8. Dependence of the SET state on the electrodes. The SET states obtained after the forming step were studied for W/a-CO x /Pt, W/a-CO x /W, and W/a-CO x /Ti devices. For each electrode combination, some devices were formed by applying the positive bias to the bottom electrode and some were formed by applying it to the top electrode. The subsequent resistance of the SET states is shown. Red bars denote the devices formed towards the top electrode and blue bars denote those formed towards the bottom electrode. There is a noticeable difference in the resistance of the SET states depending on which electrode which is positively biased during the application of a pulse. This is strongly indicative of the extent of oxygen trapping by the various electrodes. It can be seen that the SET states formed by applying a positive bias to the Pt electrode have the lowest resistance. The SET states formed by applying a positive bias to the W electrodes have intermediate resistance values and those formed by applying a positive bias to the Ti electrode have the highest resistance. There is a clear correlation with the electron affinity of those metal electrodes. The higher the electron affinity, the less favorable the metal is for oxidation, and the SET state resistance is correspondingly smaller. Each data point represents a mean of five measurements and the error bars represent the standard deviation. 8
Supplementary Table 1. Comparison of carbon-based memory concepts. Graphitic Diamond-like Graphene Graphene Oxygenated carbon 4 carbon 5 oxide 6 oxide 7 carbon (this work) Deposition Chemical Filtered Hummer s Hummer s Physical technique vapor deposition cathodic method and method and vapor (9 C) vacuum arc spin casting spin casting deposition (FCVA) (RT) (ex-situ, RT) (ex-situ, RT) (PVD) (RT) Switching Thermal Thermally Redox of inter- Oxygen Electrochemical mechanism breakdown/ induced cluster- facial oxide migration in redox reaction electrostatic ing & rupture of graphene oxide of carbon rejoining of graphitic layer graphitic bonds bonds Switching >1 µs <5 ns Not available Not available <1 ns speed Cycling 1, 1, 1 1 > 1, endurance 9
SUPPLEMENTARY NOTE 1: EVALUATION OF XPS DATA O:C ratio survey spectra from to 14 ev binding energy (BE) are acquired on each thick a-co x sample to exclude the presence of elements other than C and O (or at least not above the detection limit of the system). C 1s and O 1s core level spectra are then acquired in fixed analyzer transmission mode with 1 ev pass energy and a.5 ev energy step. To quantify the amounts of O and C, the peak areas of C 1s and O 1s are determined by integrating the peaks over their Shirley background lines. The integrated areas of the specific C 1s and O 1s peaks after background subtraction can be expressed as I(C 1s ) = Jσ C1s L C1s n C1s λ C1s cos θt (E C1s ) (1) I(O 1s ) = Jσ O1s L O1s n O1s λ O1s cos θt (E O1s ), (2) where J is the x-ray line flux intensity, σ the photoelectron cross section, L the angular asymmetry of the photoemission intensity for each atom, n the atomic density of the specific element in the matrix, λ the inelastic mean free path, and T (E) the transmission function of the analyzer. The T (E) is experimentally determined on clean gold and silver reference samples for the same pass energy and lens settings as used to analyze the a-co x samples. For photoelectric cross section values we assign Scofield s tabulated values of 1.36 1 4 X 2 and 3.985 1 4 X 2 to σ C1s and σ O1s, respectively. Alternatively, we observe that using the Yeh and Lindau tabulated values leads to the same σ C1s : σ O1s ratio. L C1s = L O1s = 2 for the C 1s and O 1s orbitals. The last parameter to be defined is λ, which is more difficult to quantify as it depends on specific parameters of the material (density, band gap and valence electrons), and they are unknown. Therefore we use different values of λ that represent the lowest and the highest limit of the range in which the real value can be. λ 1 is derived from the simulation software Sessa, which defines the density and the number of valence electrons from the molecular formula provided as input and uses these numbers to calculate λ with the TPP-2M predictive formula 8. λ 2 is calculated using an empirical formula suggested by Seah and Dench 9 : λ = A KE 2 + B KE.5, with A = 1.7 and B =.96 for inorganic compounds. Finally, λ 3 is the inelastic mean free path for photo-emitted electrons in amorphous carbon as measured and reported in ref 1. 1
Fittings of C1s peaks and C-O groups The C1s peaks are fitted by using the minimum amount of components. The difference in binding energy between the oxidized C species and the C-C sp 3 peak ( BE) is used to determine the type of CO groups in the layers. The table below compares the values of BE measured in this work and the values reported in ref. 11 and ref. 12. Chemical bond CO group BE in 11 BE in 12 BE (this work) O-C-O Epoxy 1.5 1.2 1.5 2.1 C=O Carbonyl 2.7 2.5 O-C=O Carboxyl 4.3 3.9 3.8 4.4 O-(C=O)-O Carbonate 5.5 6.2 6.5 Clearly, O-C-O, O-C=O and O-(C=O)-O groups are detected in a-co x layers. Although no peak corresponding to the C=O groups can be detected, its presence cannot be excluded: the peak at BE = 1.5 ev to 2.1 ev is quite broad and may contain a unresolved contribution from C=O groups. The C-C sp 2 peak has an asymmetric line shape resulting from the screening due to electron-hole pair excitations at the Fermi energy. The shape of the C-C sp 2 component is experimentally determined on a highly oriented, pyrolitic graphite (HOPG) reference sample that was cleaned and exfoliated with a tape prior to being loaded into the XPS chamber. The HOPG C-C sp 2 peak was fitted using a hybrid Doniach Sunjic/Gaussian Lorentzian (product) lineshape, and the parameters defining its asymmetric peak shape were then fixed in subsequent fittings of the a- CO x layers. A Gaussian Lorentzian peak lineshape was used for all other components. The BE between Csp 2 and Csp 3 was allowed to vary between.8 and 1. ev. Faint shake-up features were observed at about 292 ev in the spectra of pyrolytic graphite. However, the feature at 29 ev to 295 ev in the a-co x samples has a different origin. As the intensity ratio of this peak to Csp 3 is about the same in all different samples, in contrast to all other CO peak-area ratios, we associate this peak to some superficial moisture absorption that is the same on all samples. SUPPLEMENTARY REFERENCES 1 Robertson, J. Diamond-like amorphous carbon. Materials Science and Engineering: R: Reports 37, 129 281 (22). 11
2 Sebastian, A. et al. Resistance switching at the nanometre scale in amorphous carbon. New Journal of Physics 13, 132 (211). 3 Dellmann, L. et al. Nonvolatile resistive memory devices based on hydrogenated amorphous carbon. In Proceedings of the European Solid-State Device Research Conference (ESSDERC), 268 271 (IEEE, 213). 4 Li, Y., Sinitskii, A. & Tour, J. M. Electronic two-terminal bistable graphitic memories. Nature Materials 7, 966 971 (28). 5 Fu, D. et al. Unipolar resistive switching properties of diamondlike carbon-based RRAM devices. IEEE Electron Device Letters 32, 83 85 (211). 6 Jeong, H. Y. et al. Graphene oxide thin films for flexible nonvolatile memory applications. Nano Letters 1, 4381 4386 (21). 7 Liu, J. et al. Fabrication of flexible, all-reduced graphene oxide non-volatile memory devices. Advanced Materials 25, 233 238 (213). 8 Tanuma, S., Powell, C. J. & Penn, D. R. Calculation of electron inelastic mean free paths (IMFPs) VII. Reliability of the TPP-2M IMFP predictive equation. Surface and Interface Analysis 35, 268 275 (23). 9 Seah, M. P. & Dench, W. A. Quantitative electron spectroscopy of surfaces: A standard data base for electron inelastic mean free paths in solids. Surface and Interface Analysis 1, 2 11 (1979). 1 Prieto, P., Quiros, C., Elizalde, E. & Sanz, J. Electron inelastic mean free path and dielectric properties of a-boron, a-carbon, and their nitrides as determined by quantitative analysis of reflection electron energy loss spectroscopy. Journal of Vacuum Science & Technology A 24, 396 47 (26). 11 Moulder, J. F., Stickle, W. F., Sobol, P. E. & Bomben, K. D. Handbook of X-ray photoelectron spectroscopy, vol. 4 (Perkin Elmer Eden Prairie, MN, 1992). 12 Haubner, K. et al. The route to functional graphene oxide. ChemPhysChem 11, 2131 2139 (21). 12