† Corresponding author. E-mail:
We investigate the impact of random telegraph noise (RTN) on the threshold voltage of multi-level NOR flash memory. It is found that the threshold voltage variation (
Random telegraph noise (RTN) has become a dominant noise source as the flash memory cell size scales down. It induces the cell threshold voltage (
In this paper, we studied the impact of random telegraph noise (RTN) on the threshold voltage of multi-level NOR flash memory. The RTN amplitude and distribution under different program levels (
The flash cells used in the study were fabricated in 90-nm NOR flash technology. They were programmed by channel hot electron (CHE) injection. The bias condition for gate/drain/source/bulk (
RTN was evaluated by threshold voltage difference
According to the RTN measurement method mentioned above, we obtain the time traces of RTN amplitude (
![]() | Fig. 2. (color online) Time trace of RTN amplitude (![]() ![]() ![]() |
To gather RTN variability, several samples were measured and the complementary cumulative distribution function (1–CDF) of
![]() | (1) |
![]() | Fig. 3. (color online) RTN amplitude (![]() |
The obtained fitting parameters of η are about 22 mV, 29 mV, and 36 mV at the programming
To correctly explain the abnormal large RTN fluctuation in the NOR flash memory devices, a non-uniform channel surface potential distribution model is proposed.[9, 11] According to the quasi-two-dimensional Poisson equation, there is a surface non-uniformity potential distribution in the channel induced by the Si substrate random dopant distribution (RDD). However, the oxide charges trapping could also induce non-uniformity of potential profile in the channel, which will cause a large single RTN generation. In our study, the RTN average value is significantly increased with the program levels, therefore, we think that the oxide traps could be another reason for the non-uniform potential profile in the channel. Affected by the more fixed charges trapping in the oxide layer, the non-uniform potential distribution in the channel becomes much more obvious so that the channel currents could be divided into many narrow percolation paths. The RTN amplitude fluctuations at different program levels can be explained by the varying distance between the RTN trap position and the critical spot of the channel current path. When the RTN trap position is far from the critical spot of the channel percolation path, a smaller influence on the RTN characteristic is expected. However, when the RTN trap is just located above the critical spot of the percolation path, an abnormally large
The suggestion is further proved by the different
As the position of the RTN trap is always fixed in the tunneling oxide, we consider this is attributed to the shift of the percolation paths under different program conditions. One possible explanation is the additional oxide traps near the Si/SiO2 interface. According to the oxide trap model, more traps could be generated to capture charges under the programming condition with higher gate voltage.[21, 22] More charges trapping in the tunnel oxide could induce the larger non-uniform potential distribution in the channel, which could result in more narrow percolation paths. Thus, the RTN trap locating near the critical spot of the percolation path may induce a large RTN. To further study the impact of additional charges in oxide on the RTN characteristic, we extracted average capture time (
To verify the proposed non-uniform channel surface potential distribution model, an accurate reproduction of the influence of generated trap charges on RTN amplitude was simulated using a channel surface potential model in combination with a percolation path by 3D Silvaco Atlas software. The channel length and width of the simulated flash cell are 130 nm and 80 nm, which are the same as those of the tested flash cell. The channel doping is calibrated by comparing the simulated and tested
Figure
The simulated RTN amplitude is obtained by calculating
![]() | Fig. 7. (color online) Simulated ![]() |
The generated fixed charges trapping in the tunnel oxide at high program levels not only induce the non-uniformity potential profile in the channel, but also have an impact on the local oxide electrical field. The inset of Fig.
Based on the experimental and TCAD simulation results, we can deduce that the more fixed charges captured in the tunneling oxide traps at the high program level could lead to the larger non-uniformity of potential profile in the channel, ultimately inducing the RTN aggravation of multi-level flash memory cells.
The impacts of different
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] | |
[17] | |
[18] | |
[19] | |
[20] | |
[21] | |
[22] | |
[23] | |
[24] | |
[25] |