Fig. 4
From: Signal and noise extraction from analog memory elements for neuromorphic computing
Tradeoff between switching symmetry and SNR for ReRAM. The GPR-based methodology was applied to our ReRAM data with different metal oxide thicknesses (device A: 5 nm, device B: 4 nm) and different set (1.6–1.7 V) and reset (−1.8 to −1.9 V) voltages to investigate the relationships between the key metrics. a SNR and SF as functions of G. The original data for two representative device and pulse conditions are shown in the insets to contrast the tradeoff. b Cross-sectional 2D plot of SNR versus SF taken at G ~20 μs from a. c Plot of cumulative distribution function (CDF) of absolute values of SF for data in different SNR ranges. d Benchmark plot of SF. The optimized ReRAM data (device A, set: 1.6 V, reset: −1.8 V) were compared with refs. 28,29,30,31,32,33,34,35. The benchmark data points are color-coded (see the legend) and shown as symbols for same amplitude pulses and as crosses for varying amplitude pulses
