Supporting Information A 50 mv Cu/SiO /W Memristor with Half-Integer Quantum Conductance States S. R. Nandakumar, Marie Minvielle, Saurabh Nagar, Catherine Dubourdieu, and Bipin Rajendran, Department of Electrical Engineering., Indian Institute of Technology Bombay, Mumbai, India, and Institut des Nanotechnologies de Lyon, CNRS, École Centrale de Lyon, Université de Lyon, 69134 Ecully, France E-mail: bipin@njit.edu Methods Device fabrication Our devices consist of a SiO dielectric layer sandwiched between an active Cu top electrode (TE) and an inert Ti/TiN/W bottom electrode (BE), over a p-type Si(100) wafer with a SiO electrical isolation layer. The Ti/TiN/W stack used as a bottom electrode was sputter-deposited and patterned with a photolithography and lift-off process. The 10 nm thick SiO film and the Cu top electrode were then sputter-deposited over a patterned double layer photoresist (LOR3B and S1813) in a single vacuum and room temperature process. The 100 100 µm cross-point devices (Figure 1a To whom correspondence should be addressed Currently at New Jersey Institute of Technology, USA Institut des Nanotechnologies de Lyon, CNRS, École Centrale de Lyon, Université de Lyon, 69134 Ecully, France 1
in the main article) were finally obtained by a common lift-off process. The devices were subjected to 400 C, 5 min annealing in Ar environment after top electrode patterning. The effect of annealing was to reduce the switching threshold and HRS resistance as shown in Figure S1. 1 0-3 1 0-5 C u rre n t (A ) 1 0-7 1 0-9 A n n e a le d N o t a n n e a le d 1 0-1 1 0.0 0. 0.4 0.6 0.8 V o lta g e (V ) Figure S1. Effect of annealing. Each line in the plot corresponds to the first voltage sweep in a different device. The set of devices which are annealed have a lower switching threshold and lower HRS resistance compared to those which are not annealed. The electrical characterization result of the control experiment where the TE and BE are deposited as described above without the intervening SiO dielectric layer is shown in Figure S. C u rre n t (m A ) 1 0 0-1 0 C u W -0.5 0.0 0.5 V o lta g e ( V ) Figure S. Linear I-V characteristics seen in Cu/W material stack, used as a control experiment to verify if SiO layer is indeed needed for the switching and if any other metal oxides are present at the interface.
Electrical Characterization We used an Agilent B1500 system for device characterization. The approximate measurement setup is as shown in Figure 1f in the main article. To ensure the continuity and the low resistivity of the electrodes and to make sure of proper probe contact, before each device characterization we performed an electrode characterization. We probed the TE and BE through their end contact pads for a linear low resistance response using four separate W probes and then continued with the device characterization without disturbing the contact. For device characterization, voltages and current waveforms were applied to the top Cu electrode with the bottom W electrode grounded. We used voltage sweeps to determine the initial memristive response, current sweeps (00 na/10 ms) to find the filament growth, and periodic low voltage read pulse (50 mv, 5 ms) sampling to determine the filament decay process. The compliance current used during voltage sweep measurement is 100 µa. G ( e /h ) 6 5 4 3 1 0-0. -0.1 0.0 0.1 0. 0.3 V o lta g e (V ) Figure S3. Conductance vs. voltage graph for the voltage sweep measurement shown in Figure 1d in the main article. XPS Characterization XPS measurements were carried out with a focused monochromatized Al K α X-ray source (hν = 1486.6 ev). The pressure in the analysis chamber was ca. 1 10 9 Torr. The spectra were recorded at normal emission and the beam spot area is around 10 mm. Spectral analyses were 3
performed with CasaXPS software, using a Shirley background subtraction and deconvolution into Voigt components (70 % Gaussian, 30 % Lorentzian). The binding energy scale was referenced to the carbon contamination C 1s peak at 85 ev. For quantification, we used Scofield s photoionization cross sections and the instrumental influence is taken into account by a transmission correction factor. The Figure 1c in the main article shows the Si p spectrum of a 10 nm sputter deposited SiO film on thermal oxide on a (100) p Si type wafer. Because of spin-orbit coupling, each Si p signal consists of a p 3/ p 1/ doublet with a 0.61 ev splitting and an intensity ratio close to /1. This spectrum reveals a single chemical environment which corresponds to the +4 oxidation state (Si p 3/ at 103.6 ev). 1 3 This is consistent with our quantitative analysis that leads to an O/Si ratio of. Diffusion Simulation COMSOL with Matlab was used for the Cu diffusion simulation. We modelled the structure as shown in Figure S4, where we used Matlab to create a random initial Cu filament profile inside SiO dielectric. The profile boundary was determined from a set of a Gaussian distributed random numbers. We used COMSOL to solve the Fick s law for this D system, and determined the time at which the Cu concentration along the filament falls below a particular threshold (10 %) which is recorded as the filament break time. R a n d o m C u fila m e n t a t t = 0 s S io > 1 0 (D t) Figure S4. The D structure used in COMSOL for diffusion simulation. The lateral dimension is chosen such that it is greater than 10 Dt where D is the diffusion coefficient and t is the simulation time. We used no flux boundary condition at top and bottom and open boundary condition at the lateral edges. 4
Diffusion coefficient of Cu in SiO The experimentally determined diffusion coefficients of Cu in SiO in literature are generally those measured at high temperatures. We used the Arrhenius relation, D = D 0 exp( E a /k B T ) to determine the corresponding room temperature value. Here D 0 is the pre-exponential factor, E a is the activation energy for diffusion, k B is the Boltzmann constant, and T is the temperature. The list of room temperature values of diffusion coefficients determined by this method is given in Table S1. These values are compared with the diffusion coefficient (5 10 16 cm /s) that fits the filament decay time scale from our experiment. Table S1. Cu diffusion coefficients from literature and the corresponding room temperature values determined using Arrhenius relation. No. 1 3 Parameters D = 1. 10 11 cm /s T = 450 C E a = 1.8 ev D 0 =.5 10 8 cm /s E a = 0.93 ev D = 1.9 10 16 cm /s T = 350 C E a = 0.91 ev Room temperature Diffusion coefficient Method 9.68 10 30 cm /s Experimental 4 4.60 10 4 cm /s Experimental 5 1.1 10 4 cm /s Experimental 6 4 D = 10 13 cm /s 10 13 cm /s Experimental 7 5 5 10 16 cm /s Our model 6 D 0 = 4.5 10 5 cm /s E a = 0.18 ev, q = 0 4.06 10 8 cm /s Simulation 8 References (1) Howard, J. M.; Craciun, V.; Essary, C.; Singh, R. K. Appl. Phys. Lett. 00, 81 (18), 3431. 5
() Wang, S. J.; Lim, P. C.; Huan, a. C. H.; Liu, C. L.; Chai, J. W.; Chow, S. Y.; Pan, J. S.; Li, Q.; Ong, C. K. Appl. Phys. Lett. 003, 8 (13), 047. (3) Cho, E.-C.; Park, S.; Hao, X.; Song, D.; Conibeer, G.; Park, S.-C.; Green, M. a. Nanotechnology 008, 19 (4), 4501. (4) McBrayer, J. D. Diffusion of Metals in Silicon Dioxide. J. Electrochem. Soc. 133, 14 (1986). (5) Shacham-Diamand, Y. Copper Transport in Thermal SiO[sub ]. J. Electrochem. Soc. 140, 47 (1993). (6) Bo, C., Yan-Hui, J., Gong-Ping, L. & Xi-Meng, C. Atomic diffusion in annealed CU/SiO /Si (100) system prepared by magnetron sputtering. Chinese Phys. B 19, 06601 (010). (7) Tappertzhofen, S.; Menzel, S.; Valov, I.; Waser, R. Appl. Phys. Lett. 011, 99 (0), 03103. (8) Zelený, M. et al. Ab initio study of Cu diffusion in α-cristobalite. New J. Phys. 14, 11309 (01). 6