Figure 1: Memristive device in a plasmonic geometry. a, ReRAM in a nanoparticle-on-mirror (NPoM) geometry, with plasmonic field tightly confined in the spacer material STO between a Au nanoparticle (NP) and TiN film. b, Optical spectroscopy is performed through a transparent but electrically conductive cantilever used to contact each individual AuNP isolated by the surrounding insulating parylene layer. c, Dark field top view of AuNPs on STO layer seen through the cantilever. Figure 2: Switching cycles of NPoM memristive cell. a, Scattering intensity collected simultaneously with b, the electrical signal (voltage and current) from a single AuNP on D8. Dashed lines show times highlighted in c, giving spectra for (I) pristine sample, (II) device pre-switching, (III) device switched ON and (IV) reset of device into OFF state. A clear peak appears in the infrared (λ_2∼780-800 nm) without disappearing upon RESET. d-e, Scattering intensity collected simultaneously with f-g, the electrical signal (voltage and current) from a single AuNP on D4. Weaker scatter is observed with a strong λ_1∼ 630 nm peak and a weak λ_2∼ 750nm peak (highlighted as inset in d). h, Detailed ON (red) and OFF (black) spectra with multi-peak fits and i, extracted λ_1∼630 nm (blue) and λ_2∼750 nm (purple) intensity oscillations with switching (ON, shaded).
Figure 3: Numerical simulation of optical gap modes. a, Geometry of NP above oxygen vacancy accumulation region, field cuts taken over mid-gap blue dashed plane parallel to substrate. b, First order bright and dark modes l_10 and l_11 at 687 nm and 760 nm and second order mode l_20 at 550 nm, respectively c, NPoM with oxygen vacancy-rich bridge linking AuNP (facet w = 30 nm) to surface and d, with O2 bubble formation. Corresponding finite-difference time domain simulation modes for the two STO thicknesses varying e,f, bridge width b = 0, 10, 20, 30nm and g,h, oxygen bubble height h = 0, 1, 2, 3, 4nm after bridge of b = 30 nm is formed. Figure 4: Numerical simulations of optical modes through first switching cycle. a, Scattering intensities of structures shown in b, (I) oxygen vacancies distributed through pristine sample, (II) device pre-switching with h= 2 nm, (III) device switched ON with h= 2 nm, b= 30 nm, and (IV) resetting of device into OFF state with h= 4 nm.

Figure S1: a, Scattering intensity collected simultaneously with b, the electrical signal from a single AuNP on a 8 nm thick STO film where stages c, Scattering spectra for I, II, III and IV. Similarly d-f for a different device. Figure S2: a, Scattering intensity collected simultaneously with b, the electrical signal from a single Au NP on an annealed Sample (300 Torr O2) where the oxygen vacancies are removed. Figure S3: a, XRD θ-2θ detailed scan around (002) reflection b, RHEED diffraction patterns of MgO substrate, TiN and STO layers. c, Ex-situ atomic force microscopy (AFM) images of TiN and STO layers. Figure S4: High-resolution TEM images of a, 3.5 nm and b, 8 nm thick STO films. c, Local selected-area electron diffraction patterns and d, EDX elemental mapping show the presence of the TiO2 layer between STO and TiN (yellow arrows), originates from the oxidization of the top TiN layer when the STO films are deposited under an oxygen atmosphere. Figure S5: DFT calculation of a, real and b, imaginary dielectric function components of bulk STO and defective STO with different oxygen vacancies content (1 to 5 vacancies in a cell). Figure S6: Impact of parylene layer on near field spectra for 80 nm AuNP with 30 nm facet, dSTO = 4 nm. Figure S7: Impact of parylene layer on far field spectra, i.e. a, scattering and b, absorption, for 80 nm AuNP with 30 nm facet, dSTO = 4 nm. Figure S8: Impact of dispersive STO on far field spectra, i.e. a, scattering and b, absorption, for 80 nm AuNP with 30 nm facet, dSTO = 4 nm and parylene with n = 1.6. Figure S9: Impact of parylene thickness on far field spectra, i.e. a, scattering and b, absorption, for 80 nm AuNP with 30 nm facet, dSTO = 4nm and parylene with n = 1.6. Figure S10: Impact of TiO2 layer on the far field spectra, i.e. a, scattering and b, absorption, for 80 nm AuNP with 30nm facet, dSTO = 4 nm and TiO2 with n = 1.6. c, Far field intensity for a 80nm Au NP30 nm facet, dSTO = 8 nm, dTiO2 = 8 nm. Figure S11: a, real and b, imaginary dielectric function of TiN for different losses γ. Experimental optical data for TiN, taken from Pflüger et al. Figure S12: The optical field is mostly confined within the lateral width of FWHM Δx≃√wd~ 11nm and 15nm for D4 and D8 (with w~30 nm typical facet size of 80nm NPs). Figure S13: Comparison of a, experimental (D8) and b, simulated % change of 800 nm peak intensity gives c, the estimation of O2 bubble size. Figure S14: a, Simulated variation of the scattering spectra with the oxygen vacancies bridge formation. Comparison of b, simulated and c, experimental (D4) % change of 600 nm peak intensity with its current response. Figure S15: a-c, IV curves of sample D4 in their a, first, b, last, and c, all switching cycles. Colour scale indicates the time evolution over the cycles, i.e. in a single cycle plot (as in a,b) earlier times correspond to red forward IV curves ‘1’ and later times to green backward IV curves ‘2’, while in a multiple cycles plot (as in c) earlier times correspond to full red loops and later times correspond to full green loops. d, Endurance plot showing the HRS and LRS over a number of cycles (read at 9V). Figure S16: IV curves of TiO2/TiN on MgO substrate (read at 2V, 5V, 8V, 10V, 12V and 15V) showing typical metallic behaviour. No memory effect at any measured voltage is observed.