Random telegraph noise (RTN) in scaled RRAM devices

The random telegraph noise (RTN) related read instability in resistive random access memory (RRAM) is evaluated by employing the RTN peak-to-peak (P-p) amplitude as a figure of merit (FoM). Variation of the FoM value over multiple set/reset cycles is found to follow the log-normal distribution. P-p decreases with the reduction of the read current, which allows scaling of the RRAM operating current. The RTN effect is attributed to the mechanism of activation/deactivation of the electron traps in (in HRS) or near (in LRS) the filament that affects the current through the RRAM device.

I. Introduction

Resistive random access memory (RRAM) devices promise unique scalability, utilizing a filamentary conduction mechanism [1]–[3]. However, RTN-like read current instability observed in RRAM may effectively reduce a memory window, limiting further scaling of operational currents if the noise amplitude does not scale along with it. While previous reports were primarily focused on analyzing individual RTN signal characteristics [4]–[9], in this work we developed a quantitative approach for the description of RTN induced memory window instability. The approach allows predicting the read instability amplitude in a statistically significant number of cycles of individual RRAM cells, and in arrays of identically formed cells. Using the proposed methodology, we demonstrate that the RTN-caused read amplitude fluctuations reduce along with the operation current that removes the possible obstacle for further scaling. We propose physical mechanisms for the LRS and HRS instability explaining this experimentally observed reduction of current noise amplitudes with scaling of the operating currents