Effects of Hydrogen on Endurance Characteristics in NAND Flash Memories

This study comprehensively investigates the effect of hydrogen content in forming gas annealing (FGA) on the endurance of NAND flash memories by statistically analyzing the transconductance ( Gm,max ) characteristics. The Gm,max degradation ( ΔGm,max ) worsened with higher cycling temperature, delayed time period from erasing to programming ( tEP ) operation, and higher word-line bias ( VA ) during tEP. Moreover, these effects become more pronounced as the hydrogen content in FGA increases. Using the measured Gm,max distributions and Monte -Carlo TCAD technology, the activation energy ( EA ) of oxide damage creation can be extracted. The extracted EA values were 50 meV and 160 meV for the diluted (4%) and pure (100%) hydrogen samples, respectively. This suggests that a higher hydrogen concentration results in more ionized hydrogen atoms remaining in the oxide layer. During tEP, these hydrogen ions can drift near the Si/SiO2 surface, where they may react to form trap states. EA is revealed to have a linear relationship with respect to VA, where the slope is independent of the hydrogen content in FGA.

NAND flash memory has received considerable attention due to the increasing demand for storage capacity. The scaling path of NAND flash memory technology has evolved from planar technology to three-dimensional (3D) memory arrays. [1][2][3] Despite the revolutionary developments potentiated by modern 3D NAND technology, traditional device reliability issues continue to persist and must be addressed. In particular, the creation of oxide traps and Si/SiO 2 interface damage due to electron flow through the oxide layer in Fowler-Nordheim (FN) tunneling during program/erase (P/ E) operations. [4][5][6] These defects may cause random telegraph noise, stress-induced leakage current, and enlargement of threshold voltage (V th ) distribution because of the trapping/detrapping process. [7][8][9][10] Previous studies have reported that hydrogen plays an important role in the properties of oxide film. [11][12][13][14] In the forming gas annealing process (FGA), hydrogen can passivate oxide defects and dangling bonds by forming Si-H bonds. [15][16][17] However, hydrogen-containing species may also generate carrier trapping sites in the oxide under either hot-carrier or ionizing radiation conditions. 18,19 Although several studies have been carried out on this subject, few studies have investigated whether hydrogen annealing has any side effects may degrade the reliability of NAND flash memories. Therefore, the purpose of this study is to optimize the ambient hydrogen content and understand the underlying degradation mechanisms.
In this study, we used a recently developed transconductance (G m max , ) technique, 20,21 which enabled the extraction of cyclinginduced gate oxide charge density (Q T ) to quantitatively analyze the effect of ambient hydrogen content on the endurance characteristics of NAND flash memories. The remainder of this paper is organized as follows: section 2 describes the structure of NAND flash memory devices and experiments. Section 3 presents the presents the Monte Carlo method for evaluating Q .
T Section 4 presents the mechanisms of Q T generation under different hydrogen contents of the annealing environment. A link between the activation energies (E A ) of Q T and the generation mechanisms was built. Finally, the conclusions are drawn.

Experimental
A planar floating-gate (FG) NAND flash array was used as a test vehicle. A NAND string comprises 32 cells, a source select line transistor, and a drain select line transistor in series, as shown in Fig. 1a. The cell dimensions were as follows: the effective channel length (L) and width (W) were both 42 nm, and the tunneling oxide thickness (T ox ) was 8 nm. All cells on the selected word line (WL) were separated into even and odd pages. The samples were divided into two groups: one group went through the conventional forming gas annealing (FGA), where 4% H 2 + 96% N 2 gas were flown (48 slm N 2 + 2 slm H 2 ) at 400°C at normal pressure; the other group was annealed in 100% hydrogen ambient (50 slm H 2 ) at 400°C at normal pressure. Notably, except the difference of hydrogencontaining annealing ambient, other conditions (pressure and temperature) are the same. Moreover, to make fair comparisons, all the other fabrication processes were kept the same. Thus, it is plausible to say that reliability difference should come from difference of hydrogen-containing annealing ambient. Figure 1b is a schematic representation of the experimental procedure, which allows the investigation of the effect of hydrogen content in annealing ambient on the endurance characteristics. During cycling, the program operations are carried out only on the odd page of WL15 while the other cells in the block were kept in the erased state. Thus, only the degradation of the cells on WL15 can be measured. The programing operation is accomplished using the incremental step pulse programing (ISPP) technique 22 with an initial program voltage of 14.6 V, voltage step of 0.2 V, pulse durations of 10 μs, and verify voltage of 2 V. The erase operation is performed on a selected block by adopting the incremental step pulse erasing (ISPE) technique with an initial erase voltage of 13.6 V, voltage step of 0.4 V, and pulse durations of 10 μs applied to the p-well. As the time delay from erase to program (t EP ) has been reported to have a more significant impact on oxide degradation than that from program to erase (t PE ), 23 the effect of t EP on endurance characteristics was precisely evaluated, and endurance tests were performed with different values of t EP (i.e., 0.1, 1, and 4 s), whereas t PE was set to 0.1 s. All functional samples were cycled at various temperatures (T cyc ) and the devices were cooled to room temperature (RT) at the end of the cycling phase to monitor the cell degradation. WL15 was then swept from 0 V to 5 V to collect the current voltage (I−V) curves. Two hundred z E-mail: yungyueh.chiu@gmail.com cells on the odd page of WL15 were randomly selected to collect the Extraction of oxide charge density.-To extract the cyclinginduced Q , T we performed 3-D technology computer-aided design (TCAD) simulations on the NAND cell structure, as shown in Fig. 2a. A NAND string can be modeled as a read cell with equivalent source and drain resistances (R s and R d ). Equivalent R s and R d can be accurately obtained by calibrating the I−V characteristics of the fresh cells. After defining the cell structure, the Poisson and drift-diffusion (DD) equations were solved to obtain the I−V and G m max , of the fresh cell. Since the cycling-induced trap generation may occur at the interface and inside the tunnel oxide, discrete charges corresponding to charge trapping by cycling were randomly placed in the tunneling oxide. Figure 2b is a schematic representation of a Monte Carlo loop. Notably, the generation of random discrete charges in each cell follows an established procedure. 20 The actual number of discrete charges in the tunneling oxide follows a Poisson statistic, whose average value is derived from the product of Q T and oxide volume (W × L T ox × ). The localized charge in the tunneling oxide was reproduced by placing the electronic charge in cubic volumes with a 1-nm side. I−V and G m max , were simulated again after a fixed number of discrete charges were randomly placed in the tunneling oxide. Two hundred Monte Carlo runs were performed to obtain the G m max distributions. Finally, comparing the simulated and measured results allowed us to extract the Q . th Under this condition, the occupied D it can be regarded as fixed oxide charges located at the oxide interface. Figures 3a and 3b  Δ became more severe as the hydrogen content increased. In Figs. 4a and 4b, the extracted Q T s, as a function of the number of P/E cycles (N) are plotted on a log-log scale. When N is less than 30k, Q T generation can be approximately described by the following power law:

Results
where α is the exponential factor. 0.84 α = and was independent of t EP and hydrogen content in the annealing ambient. As the extracted Q T does not obey the power law at a large N, Eq. 1 needs to be modified as follows 20, 21 where Q 0 is the maximum value of Q T and k is the reaction constant. k is expressed as where T cyc is the temperature during P/E cycling, E A is the activation energy of oxide damage creation, and k B is the Boltzmann's constant. To evaluate the values of Q , 0 N was further increased to 200k, as shown in Fig. 5. Figure 6 shows the extracted Q T as a function of N, where Eq. × − for the 4% and 100% hydrogen samples, respectively. Notably, Q 0 is supposed to be determined by the total number of weak Si-Si and Si-H bonds. Accordingly, Q 0 is kept constant even though hydrogen may help passivate dangling bonds by forming Si-H bonds.

Discussion
E A is a well-known empirical parameter that characterizes the dependence of the failure rate coefficients on temperature and provides information on the physical mechanism of failure in NAND flash memories. 24 To distinguish the G m max , Δ degradation mechanisms, P/E cycles were performed at different T cyc values from RT to 85°C. Note that, the value of t EP was set to 0.1 s with the aim of preventing damage recovery during t EP at high T .
increased as T cyc increased. Furthermore, the    distributions exhibit a stronger discrepancy between the 4% and 100% hydrogen samples as T cyc increases. The simulations (lines) and measurements (symbols) agree over the total temperature range. Using Eqs. 2 and 3, the T cyc -dependence of the Q T can be calculated 20 : are the Q T generated at T cyc and T , R respectively. As shown in Fig. 8, Q ln T is approximately linearly dependent on k T 1 .

B cyc
/ E A can be calculated from the slopes of the graph. For the 4% hydrogen sample, the extracted E A was approximately 50 meV, whose value is close to that of defect creation and charge trapping (60-90 meV) during the cycling operations. 9 On the other hand, the extracted E A was approximately 160 meV for the 100% hydrogen sample. Since the hydrogen ions (H + ) are introduced via annealing in a hydrogen ambient, 26 the 100% hydrogen sample supplies a large amount of H + in the oxide. Accordingly, E meV 160 A ≅ for 100% hydrogen sample is related to the H + -induced defects.
The possible physical mechanisms for the oxide degradation are schematically illustrated in Figs. 9a-9c. During P/E operations, energetic electrons injected from the cathode result in anode hot-hole injection (Fig. 9a). During t , EP the generated holes may drift to the Si/SiO 2 interface under the electric field generated by the positive charges stored in the FG. Subsequently, these holes may recombine with the channel electrons, thereby creating additional trap states (Fig. 9b). These degradation processes can occur in both 4% and 100% hydrogen samples. In the case of the 100% hydrogen sample, these H + can also drift to near the Si/SiO 2 interface (Fig. 9c) where D + is a dangling bond, which acts as trapping center and lead to the cell degradation.
To verify the validity of the model in more detail, an additional experiment was performed, as shown schematically in Fig. 10a. An additional bias (V A ) was applied to the selected WL during accelerate hole and hydrogen ion transport in the tunneling oxide. The median values of the G m max , Δ distribution with various V A and T cyc values of the 4% and 100% hydrogen samples are shown in Figs. 10b and 10c, respectively. The G m max , Δ became more severe as V A increased, which is consistent with the model prediction. Using the previously mentioned methods, the value of E A for different V A was then obtained. Figure 11 shows the extracted E A as a function of V , A which can be expressed as   where E A,0 is the activation energy under the condition of V Notably, the linear relation is similar to that obtained previously for oxide breakdown under FN stress. 28

Conclusions
In this study, we investigated the effect of the hydrogen content in FGA on the endurance characteristics of NAND flash memory devices by monitoring the G m max , Δ distribution. Based on Monte Carlo simulations, we can extract the Q T and characterize the Q T evolution during cycling. It was observed that the maximum charge density Q 0 and power factor α were independent of the hydrogen content in FGA, whereas E A was the only parameter that was affected by the hydrogen content. These results can be explained by a model in which a higher hydrogen content results in more H + remaining in the oxide. These positive ions are transported through the oxide, generating carrier trapping sites in the oxide. Therefore, a clear understanding of hydrogen-related phenomena is helpful for the development of emerging NAND flash memories.