Formation of Light-Emitting Defects in Ag-based Memristors

TL;DR

The paper addresses how light-emitting defects form during the activation of Ag-based memristors by correlating electrical stimulation with in-situ photoluminescence and electroluminescence measurements. It demonstrates that PL fluctuations track the diffusion and aggregation of Ag within the dielectric gap, preceding measurable current and leading to filament formation that enables resistive switching. EL bursts occur during current instabilities and are localized to the gap, linking optical emission to dynamic filament evolution. This work offers a framework for engineering hybrid optoelectronic memristors where electrical control and light emission co-inside within a single nanoscale platform, with implications for neuromorphic photonic-electronic systems.

Abstract

Optical memristors are innovative devices that enable the integration of electro-optical functionalities - such as light modulation, multilevel optical memory, and nonvolatile reprogramming - into neuromorphic networks. Recently, their capabilities have expanded with the development of light-emitting memristors, which operate through various emission mechanisms. One notable process involves the electroluminescence of defects generated within the switching matrix during device activation. In this study, we explore the early-stage formation and evolution of the species responsible for light emission in Ag-based in-plane memristors. Our approach combines electrical stimulation with correlated optical electroluminescence and photoluminescence measurements. The findings provide valuable insights into controlling emission processes in memristors, paving the way for their integration as essential components in neuromorphic circuits.

Figures (6)

• Figure 1: a Bright-field optical image of the sample. It contains 16 connected memristors (labeled A to P), and a series of isolated structures at the center for characterization purpose. The entire coverslip is covered by a 160 nm-thick Poly(methyl methacrylate) layer. b Scanning electron micrograph of a typical memristive gap. It consists of the two Ag tapered electrodes separated by a gap of 300 nm. c Experimental setup. The activation of the memristor and repetitive resistive switching are induced by applying voltage pulses from an arbitrary function generator. The current flowing is measured using a trans-impedance amplifier acting as a current-to-voltage converter. A protection resistance $R$ limits the current. The photoluminescence (PL) is excited with a 515 nm continuous laser tightly focused to a diffraction limited spot on the device. The same objective collects the PL emission (shown in orange) and the laser reflection (shown in green). The sample is raster scanned and these signals are detected by avalanche photodiodes at each positions to reconstruct spatial maps. PL spectra are obtained by a spectrometer. Finally, electroluminescence (shown in red) is detected by a third avalanche photodiode and a charge-coupled device.
• Figure 2: a Time plot of the voltage and the current at the end of the activation phase. $V=3\ \mathrm{V}$, $T=500\ \mathrm{ms}$, and $D=0.2$. Inset: SEM image revealing a filamentary structure between the two Ag electrodes formed during the activation of the device. b Zoomed-in showing the rise of the current (i.e. the potentiation) during 30 consecutive pulses.
• Figure 3: a Time plot of the voltage/current during a sequence of voltage pulses following the activation phase. $V=2$ V, $T=500$ ms and $D=0.2$,. b Corresponding electroluminescent trace. c Composite false-color CCD image of the memristor during operation. The overlay consists of an image acquired during the pulse sequence combined with an image of the sample observed under a weak illumination and in absence of an applied voltage. EL is restricted to the gap region. The electrodes are recognized as the darker grey areas.
• Figure 4: a Reconstructed pulse-to-pulse evolution of the PL spectrum (i) and the current (ii) during the activation sequence $n=8$. The laser power is $P=50$$\mu$W and the integration time is equal to the duration of the voltage pulse (100 ms). All spectra are corrected for the background noise of the CCD and the quantum efficiency of the detection path (transmission of the objective, CCD and grating efficiencies). The spectrum displayed in (i) is the average spectrum over the entire sequence. (iii) and (iv) are confocal scans of the PL and the laser reflection obtained after the pulse sequence. An image-wide profile across the gap is shown in white in (iv). b, c and d are similar measurements taken for sequences $n=13$, $n=16$, and $n=17$, respectively. Horizontal arrows in b(i) are pointing at various types of the spectral fluctuations such as intensity flare, spectral shift, and intermittency. The vertical arrows in b(ii) and c(ii) mark the onset of current flow during the sequence.
• Figure 5: a Pulse-to-pulse evolution of the PL (i) and current (ii) measured during a regime of unstable current level ($N<980$) and while current is at compliance ($N>980)$. b,c Probability densities of the spectral PL fluctuations measured under unstable current conditions, and at current compliance, respectively.