Fig. 5 | Nature Communications

Fig. 5

From: Signal and noise extraction from analog memory elements for neuromorphic computing

Fig. 5

Separation of device-to-device variability and inherent randomness for PCM. a Device G as a function of pulse number for two representative PCM devices (devices 1 and 2), fabricated with the identical process. The predicted noise-free signals from the GPR fitting are shown in red and blue solid lines. The difference of two fitted lines corresponds to device-to-device variability, whereas the fluctuations of the experimental data around the fitted lines are attributed to inherent randomness from individual devices. Histograms of ΔG values extracted from experimental data (blue) and fitted curves (red) (b) after the second pulse and (c) after the sixth pulse. d The mean and standard deviation of ΔG obtained from the experimental data (shown in black circles and error bars) were compared with the root mean square of inherent randomness (r) obtained from the GPR-based methodology (shown in red error bars) as a function of pulse number. The inherent randomness accounts for 67% of the total variability

Back to article page