Physical and Electrical Analysis of Poly-Si Channel Effect on SONOS Flash Memory

In this study, polycrystalline silicon (poly-Si) is applied to silicon-oxide-nitride-oxide-silicon (SONOS) flash memory as a channel material and the physical and electrical characteristics are analyzed. The results show that the surface roughness of silicon nitride as charge trapping layer (CTL) is enlarged with the number of interface traps and the data retention properties are deteriorated in the device with underlying poly-Si channel which can be serious problem in gate-last 3D NAND flash memory architecture. To improve the memory performance, high pressure deuterium (D2) annealing is suggested as a low-temperature process and the program window and threshold voltage shift in data retention mode is compared before and after the D2 annealing. The suggested curing is found to be effective in improving the device reliability.


Introduction
As the nonvolatile memory market grows rapidly, a lot of research has been reported to improve the device performance and reliability. Especially, 3D silicon-oxide-nitride-oxidesilicon (SONOS) flash memory structure has been suggested to overcome the physical limitation in scaling down the feature size of the existing 2D structure [1][2][3][4][5]. Representative ones are the Stacked Memory Array Transistor (SMArT) by SK Hynix, the Pipe Bit Cost Scalable (P-BiCs) by Kioxia, and Terabit Cell Array Transistor (TCAT) by Samsung. One of the distinct changes in these 3D structures is that a crystalline silicon (c-Si) channel is replaced by a polycrystalline silicon (poly-Si). Poly-Si film is composed of crystalline grains with different crystallographic orientations and grain boundaries with highly defective interfaces [6]. The random mixed structure of grains and grain boundaries is known to cause rough surface compared to a c-Si, which can deteriorate the device performances. In addition, as the area and thickness of the cell decrease due to high integration, the polysilicon channel effect may become more severe. Among the 3D SONOS structures, the devices similar to TCAT structure are based on a gate-last process. That is, a poly-Si channel is first formed and then silicon nitride (Si 3 N 4 ) as a charge trapping layer (CTL) is deposited. In this case, the characteristics of the CTL will be affected with the underlayer's topology because thickness of the CTL is only a few nm. The memory window, date retention as well as program/erase speed of SONOS devices are most affected by trap properties of CTL [7][8][9]. Therefore, when the underlaying poly-Si has large surface roughness as discussed above, the mismatch between materials can be intensified causing larger interface traps and a problem in the device reliability.
In this study, the physical and electrical properties of SONOS device with poly-Si channel are analyzed. Atomic force microscope (AFM) to measure the surface roughness and x-ray photoelectron spectroscopy (XPS) to find out the bonding energy of CTL films with poly-Si underlayer were used. For the electrical analysis, threshold voltage (V TH ) shift 2 of 9 was extracted through the data retention measurements. Moreover, to improve the memory properties, high pressure deuterium (D 2 ) annealing is suggested. D 2 annealing has recently emerged to improve the reliability of MOSFET device by curing shallow traps [10,11]. The experimental results show that by D 2 annealing the reliability of the SONOS device with poly-Si channel can be improved at low temperature.

Experiments
A SONOS structured capacitors were fabricated with c-Si and poly-Si as channels. Figure 1 shows the cross-sectional view and process flow of the device. Prime grade p-type c-Si was used as the substrate, and the thickness of tunneling oxide (TO, SiO 2 ), CTL (Si 3 N 4 ), and blocking oxide (BO, SiO 2 ) was 7 nm, 15 nm and 15 nm, respectively. In the case of a poly-Si channel device, 200 nm of thermal SiO 2 was grown to isolate c-Si and poly-Si, and 50 nm of poly-Si was deposited by low pressure chemical vapor deposition (LPCVD). For TO, c-Si channel device was grown using thermal oxidation and poly-Si channel device was deposited using LPCVD. Then, CTL and BO were deposited using LPCVD. For gate electrode, a 100 nm thick titanium (Ti) film was deposited by RF sputter. Table 1 shows the process conditions of SiO 2 and Si 3 N 4 deposited by LPCVD. In this study, high pressure D 2 annealing is suggested as a passivation method of Si 3 N 4 . High pressure annealing has the advantage of reducing both processing temperature and time. After the gate formation, the annealing was performed at 450 • C, 10 atm, 1 h. The fabricated devices have a gate width by length of 100 µm/100 µm. with poly-Si underlayer were used. For the electrical analysis, threshold voltage (VTH) shift was extracted through the data retention measurements. Moreover, to improve the memory properties, high pressure deuterium (D2) annealing is suggested. D2 annealing has recently emerged to improve the reliability of MOSFET device by curing shallow traps [10,11]. The experimental results show that by D2 annealing the reliability of the SONOS device with poly-Si channel can be improved at low temperature.

Experiments
A SONOS structured capacitors were fabricated with c-Si and poly-Si as channels. Figure 1 shows the cross-sectional view and process flow of the device. Prime grade ptype c-Si was used as the substrate, and the thickness of tunneling oxide (TO, SiO2), CTL (Si3N4), and blocking oxide (BO, SiO2) was 7 nm, 15 nm and 15 nm, respectively. In the case of a poly-Si channel device, 200 nm of thermal SiO2 was grown to isolate c-Si and poly-Si, and 50 nm of poly-Si was deposited by low pressure chemical vapor deposition (LPCVD). For TO, c-Si channel device was grown using thermal oxidation and poly-Si channel device was deposited using LPCVD. Then, CTL and BO were deposited using LPCVD. For gate electrode, a 100 nm thick titanium (Ti) film was deposited by RF sputter. Table 1 shows the process conditions of SiO2 and Si3N4 deposited by LPCVD. In this study, high pressure D2 annealing is suggested as a passivation method of Si3N4. High pressure annealing has the advantage of reducing both processing temperature and time. After the gate formation, the annealing was performed at 450 °C, 10 atm, 1 h. The fabricated devices have a gate width by length of 100 μm/100 μm.

Physical Characteristic Analysis
AFM was used to compare the surface roughness of nitride-oxide (NO) stacked film on c-Si and poly-Si. Figure 2a,b shows the AFM images of NO on c-Si, poly-Si. As a reference, the surface roughness of poly-Si on Si substrate is presented in Figure 2c. From Figure 2a,b, it can be seen that Si 3 N 4 has very rough surface on poly-Si. In Table 2, the extracted roughness values are summarized. From the results, it can be seen that the surface roughness of a film is affected by the underlayer.

Physical Characteristic Analysis
AFM was used to compare the surface roughness of nitride-oxide (NO) stacked film on c-Si and poly-Si. Figure 2a,b shows the AFM images of NO on c-Si, poly-Si. As a reference, the surface roughness of poly-Si on Si substrate is presented in Figure 2c. From Figure 2a,b, it can be seen that Si3N4 has very rough surface on poly-Si. In Table 2, the extracted roughness values are summarized. From the results, it can be seen that the surface roughness of a film is affected by the underlayer.  Table 2. AFM analysis results on surface roughness of NO stacked film on c-Si and poly-Si. In addition, poly-Si roughness is also measured for a reference which is expressed as "Only poly-Si" sample.

Condition
Peak From the AFM results, the cross section of the poly-Si device can be drawn as Figure  3, where the roughness of poly-Si causes poor coverage of TO and CTL resulting in the thickness and electric filed fluctuation in each layer. In this case, the electrical characteristics of device are deteriorated by the local increase of electric field in the thin area [12,13]. In addition, the interface roughness influences the interface defect formation by intensifying the lattice constant mismatch between the layers [14,15]. It is well known that the shallow interface traps between TO and CTL affects SONOS flash memory performance [16,17].  Table 2. AFM analysis results on surface roughness of NO stacked film on c-Si and poly-Si. In addition, poly-Si roughness is also measured for a reference which is expressed as "Only poly-Si" sample. From the AFM results, the cross section of the poly-Si device can be drawn as Figure 3, where the roughness of poly-Si causes poor coverage of TO and CTL resulting in the thickness and electric filed fluctuation in each layer. In this case, the electrical characteristics of device are deteriorated by the local increase of electric field in the thin area [12,13]. In addition, the interface roughness influences the interface defect formation by intensifying the lattice constant mismatch between the layers [14,15]. It is well known that the shallow interface traps between TO and CTL affects SONOS flash memory performance [16,17].

XPS Analysis
The depth profile analysis of XPS was performed to investigate the bonding structure of the TO and CTL interface according to the channel material change. Figure 4 shows the XPS multi-peak fitting results of Si 2p peak, which are corrected to 285.5 eV of C 1s. The Si 2p spectra can be fitted by 4 peaks using a Gaussian function. Si-Si peak is 99.9 ± 0.15 eV, Si-SixNy peak is 101.3 ± 0.15 eV, Si3N4 peak is 102.1 ± 0.1 eV, and SiO2 peak is 103.4 ±

XPS Analysis
The depth profile analysis of XPS was performed to investigate the bonding structure of the TO and CTL interface according to the channel material change. Figure 4 shows the XPS multi-peak fitting results of Si 2p peak, which are corrected to 285.5 eV of C 1s. The Si 2p spectra can be fitted by 4 peaks using a Gaussian function. Si-Si peak is 99.9 ± 0.15 eV, Si-Si x N y peak is 101.3 ± 0.15 eV, Si 3 N 4 peak is 102.1 ± 0.1 eV, and SiO 2 peak is 103.4 ± 0.1 eV [18,19]. Si-Si x N y bonding represents a combination of Si 3 N 4 that does not match the composition ratio. Table 3 shows the peak positions and ratios in each device. Ratio is the percentage of the area of each peak to the total area of the peak. The Si-Si and SiO 2 ratios show similar percentages, but the SixNy and Si 3 N 4 ratios show opposite results. The reason why the Si-SixNy bonding ratio is higher in poly-Si devices can be explained by the interface degradation as in AFM analysis. Si-Si x N y bonding acts as trap in the CTL and affects the reliability of the memory.

XPS Analysis
The depth profile analysis of XPS was performed to investigate the bonding structure of the TO and CTL interface according to the channel material change. Figure 4 shows the XPS multi-peak fitting results of Si 2p peak, which are corrected to 285.5 eV of C 1s. The Si 2p spectra can be fitted by 4 peaks using a Gaussian function. Si-Si peak is 99.9 ± 0.15 eV, Si-SixNy peak is 101.3 ± 0.15 eV, Si3N4 peak is 102.1 ± 0.1 eV, and SiO2 peak is 103.4 ± 0.1 eV [18,19]. Si-SixNy bonding represents a combination of Si3N4 that does not match the composition ratio. Table 3 shows the peak positions and ratios in each device. Ratio is the percentage of the area of each peak to the total area of the peak. The Si-Si and SiO2 ratios show similar percentages, but the SixNy and Si3N4 ratios show opposite results. The reason why the Si-SixNy bonding ratio is higher in poly-Si devices can be explained by the interface degradation as in AFM analysis. Si-SixNy bonding acts as trap in the CTL and affects the reliability of the memory.      Figure 5 shows the basic structure of Si 3 N 4 and the trap model. In Si 3 N 4 , silicon vacancy (V Si ) and nitrogen vacancy (V N ) can be made, and their properties can be changed by atoms entering the vacancy. In general, interface traps have relatively shallower energy traps than bulk traps. As a characteristic of the SONOS structure, since the CTL is adjacent to the oxide layer, oxygen related defects may be formed by oxygen diffusion in the TO or BO. In particular, the bond by the O atom substituted with V N has a small energy level [20]. From the physical analysis, it was confirmed that the roughness deterioration of the underlayer formed more V Si and V N between the TO/BO and CTL, and the increase of the interface trap had a significant effect on the reliability of the memory [20][21][22]. by atoms entering the vacancy. In general, interface traps have relatively shallower energy traps than bulk traps. As a characteristic of the SONOS structure, since the CTL is adjacent to the oxide layer, oxygen related defects may be formed by oxygen diffusion in the TO or BO. In particular, the bond by the O atom substituted with VN has a small energy level [20]. From the physical analysis, it was confirmed that the roughness deterioration of the underlayer formed more VSi and VN between the TO/BO and CTL, and the increase of the interface trap had a significant effect on the reliability of the memory [20][21][22].

Data Retention Measurement
The program (PRG) and data retention behavior of the fabricated devices were measured as shown in Figure 6 and the charge loss in data retention mode was calculated. In data retention mode, C-V were measured after baking at 75~125 °C (25 °C step) for 1 h after programming. Poly-Si channel shows large program window (higher VTH shift) at the same program voltage than c-Si channel. However, larger VTH shift (ΔVTH) in the data retention meaning inferior reliability. Figure 7 shows ΔVTH in the data retention mode according to temperature. VTH was extracted as a gate voltage at 80% of the maximum capacitance. ΔVTH is much larger in all temperature conditions and the data retention characteristics are deteriorated in poly-Si devices. Considering shallow traps improves program windows with the traditional tradeoff in data retention properties, the experimental results show that more traps exist in the poly-Si channel device, which is same with the previous physical analysis.

Data Retention Measurement
The program (PRG) and data retention behavior of the fabricated devices were measured as shown in Figure 6 and the charge loss in data retention mode was calculated. In data retention mode, C-V were measured after baking at 75~125 • C (25 • C step) for 1 h after programming. Poly-Si channel shows large program window (higher V TH shift) at the same program voltage than c-Si channel. However, larger V TH shift (∆V TH ) in the data retention meaning inferior reliability. Figure 7 shows ∆V TH in the data retention mode according to temperature. V TH was extracted as a gate voltage at 80% of the maximum capacitance. ∆V TH is much larger in all temperature conditions and the data retention characteristics are deteriorated in poly-Si devices. Considering shallow traps improves program windows with the traditional tradeoff in data retention properties, the experimental results show that more traps exist in the poly-Si channel device, which is same with the previous physical analysis.

High Pressure D2 Annealing Effect
In this study, high pressure D2 annealing is suggested to make the device stable as a low temperature process method. Figure 8 shows trap models in Si3N4. Figure 8a shows 4N-H defect where hydrogen enters into VSi. This defect can be ignored because its energy level is not in the bandgap [23]. In Figure 8b, hydrogen enters into nitrogen vacancy (VN) and forms a 1Si-H defect and some Si-Si bonds [24,25]. Figure 8c shows O-related traps due to oxygen diffusion into VN [26], which is common near oxide film similar to TO/CTL or CTL/BO interface. In this experiment, to suppress O-related defects near TO/CTL or CTL/BO interface, passivation of VN is focused and high pressure D2 annealing is applied as stable curing method at low temperature.

High Pressure D 2 Annealing Effect
In this study, high pressure D 2 annealing is suggested to make the device stable as a low temperature process method. Figure 8 shows trap models in Si 3 N 4 . Figure 8a shows 4N-H defect where hydrogen enters into V Si . This defect can be ignored because its energy level is not in the bandgap [23]. In Figure 8b, hydrogen enters into nitrogen vacancy (V N ) and forms a 1Si-H defect and some Si-Si bonds [24,25]. Figure 8c shows O-related traps due to oxygen diffusion into V N [26], which is common near oxide film similar to TO/CTL or CTL/BO interface. In this experiment, to suppress O-related defects near TO/CTL or CTL/BO interface, passivation of V N is focused and high pressure D 2 annealing is applied as stable curing method at low temperature.  Figure 9 shows the C-V data retention measurement results according to high pressure D2 annealing of a poly-Si channel device. Table 4 shows the VTH extracted according to the measurement temperature and high pressure D2 annealing. After D2 passivation through high pressure annealing, the memory window decreased, but the device reliability was greatly improved. These results are consistent with the previously predicted effect of shallow trap curing of Si3N4 by D2 annealing [27]. This result implies that even though the hydrogen can be dissociated during the post annealing period, some could form stable bondage with Si and N atoms. Furthermore, previous research had conducted D2 high pressure annealing even at 900 °C temperature [28].
For the physical analysis on the reduced shallow trap density, it is needed to detect the change in atomic bonding in Si3N4 by the deuterium bonding. FT-IR analysis can be employed for detecting and determining bond densities of light atoms such as H2 or D2 [29]. In this experiment, the results are not presented but Thermo-Nicolet 5700 FT-IR spectrometer is analyzed on the c-Si channel device with and without D2 treatment where the  Figure 9 shows the C-V data retention measurement results according to high pressure D 2 annealing of a poly-Si channel device. Table 4 shows the V TH extracted according to the measurement temperature and high pressure D 2 annealing. After D 2 passivation through high pressure annealing, the memory window decreased, but the device reliability was greatly improved. These results are consistent with the previously predicted effect of shallow trap curing of Si 3 N 4 by D 2 annealing [27]. This result implies that even though the hydrogen can be dissociated during the post annealing period, some could form stable bondage with Si and N atoms. Furthermore, previous research had conducted D 2 high pressure annealing even at 900 • C temperature [28].
For the physical analysis on the reduced shallow trap density, it is needed to detect the change in atomic bonding in Si 3 N 4 by the deuterium bonding. FT-IR analysis can be employed for detecting and determining bond densities of light atoms such as H 2 or D 2 [29]. In this experiment, the results are not presented but Thermo-Nicolet 5700 FT-IR spectrometer is analyzed on the c-Si channel device with and without D 2 treatment where the sample with D 2 HPA with 600 • C annealing shows slightly increased absorbance in the rage of 2375 cm −1 . It is difficult to detect the light atoms such as hydron quantitatively, more precise physical method should be studied.
sure D2 annealing of a poly-Si channel device. Table 4 shows the VTH extracted according to the measurement temperature and high pressure D2 annealing. After D2 passivation through high pressure annealing, the memory window decreased, but the device reliability was greatly improved. These results are consistent with the previously predicted effect of shallow trap curing of Si3N4 by D2 annealing [27]. This result implies that even though the hydrogen can be dissociated during the post annealing period, some could form stable bondage with Si and N atoms. Furthermore, previous research had conducted D2 high pressure annealing even at 900 °C temperature [28].
For the physical analysis on the reduced shallow trap density, it is needed to detect the change in atomic bonding in Si3N4 by the deuterium bonding. FT-IR analysis can be employed for detecting and determining bond densities of light atoms such as H2 or D2 [29]. In this experiment, the results are not presented but Thermo-Nicolet 5700 FT-IR spectrometer is analyzed on the c-Si channel device with and without D2 treatment where the sample with D2 HPA with 600 °C annealing shows slightly increased absorbance in the rage of 2375 cm −1 . It is difficult to detect the light atoms such as hydron quantitatively, more precise physical method should be studied.

Conclusions
In this study, the physical and electrical characteristics of the SONOS flash memory device with poly-Si channel were analyzed, and high pressure D 2 annealing was suggested to improve the device performance. For physical analysis, surface roughness through AFM and trap of TO/CTL interface through XPS were analyzed. From the physical analysis, it was confirmed that underlaying poly-Si deteriorates the surface roughness of CTL and enlarges physical defects. For electrical analysis, ∆V TH was measured in data retention mode and larger in poly-Si devices as confirmed in the physical analysis. However, by high-pressure D 2 annealing, the deteriorated memory characteristics can be improved. From the data retention measurement before and after D 2 annealing, it was confirmed that the memory window slightly decreased due to the curing of the interface trap, but the ∆V TH significantly decreased. The results show a problem that appears when poly-Si channel is used in SONOS devices and indicate that high pressure D 2 annealing is effective method to control the trap sites in interface of CTL.