5.1 Preface
Random telegraph noise arising from electron emission and capture at an interface trap site has been recognized as a new scaling concern in flash memories [1.37]-[1.40]. Vt fluctuations originated from a large-amplitude RTN tail will cause a read error and become a prominent issue in designing a multilevel-cell (MLC) flash memory in 45nm technology node and beyond [1.39]. A statistical model based on a three-dimensional Monte Carlo simulation has shown that single-trap RTN amplitudes and thus Vt fluctuations exhibit an exponential distribution, i.e., f(vt)=exp(−vt/)/
[4.7][5.1]. In a FG flash memory, the RTN tail is attributed to random dopant induced current-path percolation effects and is dependent on a substrate doping concentration. Unlike a FG flash, where program charges are stored in a conducting poly-silicon FG, program charges in a SONOS cell are stored randomly and discretely in a silicon nitride layer. (Fig. 5.1) Current percolation paths in a program-state of a MLC SONOS are determined by the placement of both substrate dopants and nitride trapped charges.
In this work, we characterized RTN in erase and program states in FG flash, planar SONOS and FinFET SONOS cells. +Vg/−Vg FN injection is employed for program and erase. The program Vt window is chosen to be 1V for MLC application.
5.2 Program Charge Effect on RTN Amplitude in Floating Gate and SONOS Flash Memory
5.2.1 Measurement of RTN Amplitude in MLC Flash
In order to identify the concept of program charge induced percolation effect, first of all we measured single-trap RTN relative amplitudes (Id/Id) versus drain current in both FG flash cell (Fig. 5.2) and SONOS flash cell (Fig. 5.3) with different program window for MLC application. The FG flash cell dimension is W/L=0.11m/0.09m.
RTN amplitudes in erase-state and in two different program-state are set at the same read current level of 500nA, drain voltage at 0.7V.
We find that program-state and erase-state RTN amplitude are identical, no matter program Vt is 1V or 2V. However, as we measured single-trap RTN relative amplitudes versus drain current in a SONOS flash cell (cell dimension:
W/L=0.09m/0.08m, a 2.8 nanometer tunnel oxide, a 6 nanometer silicon nitride and a 6 nanometer top oxide) with program window 0.8V and 1.5V, the curve of erase-state and program-state misalign.
The result can be explained by the idea we mentioned earlier in preface: (i) The program charges in the FG flash is continuous distribution and does not affect the percolation paths caused by substrate dopants due to a conducting charge storage layer (poly-Si). (ii) The current percolation paths are affected by the placement of both substrate dopants and random program charges in SONOS flash due to discrete charge storage. Furthermore, we can get that RTN amplitude decreases as drain current increases indicating that number fluctuation dominates at high current level and percolation effect plays a more important role at low current level.
5.2.2 Statistics Result of Program-state and Erase-state RTN
We also measured single-trap RTN relative amplitudes (Id/Id) in 40 FG flash cells and 60 SONOS flash cells, then we perform a bit-by-bit tracking plot of program-state RTN amplitude versus erase-state RTN amplitude. Devices with RTN amplitudes less than 3% are excluded to avoid possible measurement errors. In the case of FG flash, we find that almost all the dots lay on the straight line with slope=1, which indicates program-state and erase-state RTN have identical amplitudes in each FG cell (Fig. 5.4).
As a contrast, a distinctly different feature is obtained in planar SONOS cells.
The RTN amplitudes spread in a wide range after programming and are almost independent of erase-state RTN (Fig. 5.5). One erase-state RTN amplitude might have many possible program-state RTN amplitudes after programming. Therefore, we can deduce that program charge effect on RTN amplitude is insignificant in FG flash but severe in SONOS flash.
5.2.3 Correlation Factor for Program-state and Erase-state RTN
To quantify program charge effect on RTN amplitude, a correlation factor, f, for program-state and erase-state RTN is defined as [5.2]
(5.1)
where x and y denote RTN amplitudes in erase-state and in program-state, respectively, and x and y are average values.
A larger correlation factor suggests a smaller program charge induced percolation effect. Table 5.1 shows the measured correlation factor is 0.998 in FG flash,
suggesting no program charge effect on RTN while the correlation factor reduces to 0.286 in planar SONOS flash.
5.2.4 P/E Cycle Dependence of RTN
The RTN amplitude versus the drain current in the first three P/E cycles in FG flash is shown in Fig. 5.6. The result shows program-state and erase-state have the same RTN characteristics and implies that program charges in a FG do not alter current percolation paths caused by substrate dopants and no P/E cycle dependence.
The first three P/E cycles in SONOS flash is shown in Fig. 5.7. The program-state RTN amplitude varies from cycle to cycle, suggesting that random program charges play an important role in current percolation paths. The measured RTN waveforms and the Id −Vg for SONOS flash are shown in Fig. 5.8 and the waveforms of the first two program-states are shown in Fig. 5.9. Two-level current switching is observed in both erase and program-states, showing that RTN arises from a single interface trap and no additional traps are created during P/E cycles. As a result, we affirm that the variation of RTN amplitude from cycle to cycle is attributed to different program charge percolation paths, not additional trap creation.