Low frequency noise variability in ultra scaled FD-SOI n-MOSFETs: Dependence on gate bias, frequency and temperature

Noise and Fluctuations in Fully Depleted Silicon-on-Insulator MOSFETs

“…This trend is different than the one shown in Figure 34(a) for V G = 0.4 V, thus the way that the noise variability depends on the frequency changes with the device bias. Following the same data analysis methodology, a variability dependence on gate voltage and temperature was also found [68] (see Figure 35). To further probe the gate bias dependence of σ(log(S Vg )), a more intuitive plot is shown in Figure 36 (after [68]), where σ(log(S Vg )) is plotted versus frequency and V G at the same time.…”

“…In Figure 25, a dispersion of up to 4 decades of magnitude can be observed at high frequencies, while almost 2 decades around f = 1 Hz, revealing a possible frequency dependence of the standard deviation. Indeed, as shown in Figure 34(a) (after [68]), where the measured spectra of 85 devices at V G = 0.4 V are plotted along with the noise standard deviation, a clear dependence of LFN variability with frequency is visible. In fact, in the frequency regions where σ(log(S Vg )) is increased, there are Lorentzian-like spectra with significantly higher density levels than the remaining spectra.…”

“…Thus, the CNF/CMF flicker noise model, needed for the extraction of N t , is very difficult to be applied in nano-scale MOSFETs, due to the lack of uniformity in the oxide trap distribution, which results in non-flicker noise spectra. However, if the statistical sample is large enough, the average noise spectrum will be 1/f-like (see Figure 25) [29], [68], making it suitable for extracting the average trap density N t [72]. Indeed, if we plot the extracted density values at 10 Hz from the spectra of Figure 26 versus the drain current (see Figure 29), the average LFN level follows the CNF model [72].…”

“…Thus, a detailed study of the advanced FD-SOI LFN variability is crucial in order to better understand its origin and by turn its prediction and/or reduction. An example of noise variability experimental results is shown in Figure 25 (after [68]), where S Vg is plotted versus frequency for all measured devices across a 300 mm wafer biased under the same gate voltage bias V G = 0.4 V. Notice that few spectra are 1/f-like, but there is also a significant number of Lorentzian-like spectra, commonly observed in such small area (W.L = 0.0012 μm 2 ) devices, due to Random Telegraph Noise (RTN), or due to generation-recombination (GR) centers located in the Si film [65], [69]. From Figure 25 (after [68]), it is clear that the noise variability is strongly enhanced when Lorentzian noise is present.…”

“…or, even better, in order to eliminate the influence of the subthreshold swing or threshold voltage variability in the LFN variability estimation, one can use [68]:…”