Tunable bandgap energy of fluorinated nanocrystals for flash memory applications produced by low-damage plasma treatment

The fabrication of a Gd₂O₃-NC flash memory device is shown schematically in figure 1. After an n-type (100) silicon wafer received the standard RCA wafer cleaning process, a 3 nm thick tunnel oxide was thermally grown in a horizontal furnace. Next, a 10 nm thick amorphous Gd₂O₃ layer was deposited by sputtering with a Gd (99.9% pure) target in an oxygen (O₂) and argon (Ar) mixture ambient at room temperature. The flow rates of O₂ and Ar were 3 and 21 sccm, respectively. After the dielectric films formed, the samples were treated by rapid thermal annealing (RTA) at 900 °C for 30 s in nitrogen ambient to form the Gd₂O₃-NCs. The samples were then treated using a low-damage CF₄ plasma in a PECVD system with a complementary quartz filter near the substrate, as shown schematically in figure 2(a). As shown in figure 2(b), the quartz filter consists of upper and lower plates separated by a quartz spacer. Both plates contain many stripes and slits of the same width. The slits of the upper plate are aligned with the stripes of the lower plate. One side of the upper plate is coated with a thin metal film. The filter is placed on a supporter coated with an insulating material so that energetic ions can be confined to the plasma area between the upper electrode and the filter [28]. In this arrangement, the filter can efficiently shield the sample from plasma damage caused by energetic electrons, energetic ions, and UV photons. It permits only neutral and reactive radicals to diffuse through the filter and reach the heated substrate (300 °C), thus realizing a low-damage plasma treatment procedure. In pure CF₄ plasma, the radicals are primarily fluorine radicals with a small concentration of complex fluorocarbon species [29]. As a result, Gd₂O₃-NC films can be effectively fluorinated with minimal plasma damage. Radio frequency (RF) power (100 W, 13.56 MHz) was applied to the upper electrode through a matching box. The base pressure of the PECVD was less than 1 × 10⁵ Torr, and the process pressure was maintained at 500 mTorr. To confirm that the filter reduced the UV irradiation, optical emission spectroscopy (OES) with a quartz lens was used to measure the optical emission between the filter and the substrate through a quartz window on the side wall of the PECVD system, aimed into a fibre bundle. The emission at the same position was also measured when the filter was not inserted into the PECVD system. Figure 2(c) shows the emission spectra in the UV range with and without the filter. After the filter was inserted, the UV photon intensity decreased dramatically. By calculating the area under the spectra, we found that more than 95% of the UV photons can be eliminated. According to [28] and the OES results, we believe that plasma damage resulting from energetic ions and UV irradiation can be efficiently eliminated. After CF₄ plasma treatment, a SiO₂ film serving as a blocking oxide was deposited by PECVD on all of the samples using SiH₄ (5 sccm) and N₂O (200 sccm) mixed gases at 300 °C. Finally, a 300 nm aluminum film was deposited by a thermal coater on both sides of the samples, and a gate on the blocking oxide was defined by lithography and wet etching.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

X-ray spectroscopy (XPS) was used for material analysis of the fluorinated Gd₂O₃-NC films. UV absorption spectra of the films at 150–350 nm were measured with light from a synchrotron source. The light was dispersed from a high-flux beamline with a 6 m cylindrical grating monochromator coupled to the 1.5 GeV storage ring at the National Synchrotron Radiation Research Centre in Taiwan. In this work, incident light with a resolution of 0.5 nm and intensity of 1.2 × 10⁶ photons s⁻¹ was conducted through a gold mesh with about 90% transmission; subsequently, the photocurrent of the gold mesh was detected by an electrometer (Keithley 6512) for monitoring and normalization of the beam. The light transmitted through the sample impinged on a glass window coated with sodium salicylate, the fluorescence of which was detected by a photomultiplier tube (Hamamatsu R943) in photon-counting mode. For UV absorption measurements, a fluorinated Gd₂O₃-NC film was prepared on a fused quartz substrate. The absorption of the film was determined by the following formula [30]:

$$\begin{eqnarray*} &&\ln ({I}_{0}/I)=\alpha d, \end{eqnarray*}$$

where α is the absorption coefficient, d is the film thickness, and I₀ and I are the intensity of transmitted photons through the quartz substrate and the Gd₂O₃-NC film/quartz substrate, respectively.

Atomic force microscopy (AFM) and high-resolution transmission electron microscopy (HRTEM) observations were also made to observe the nanostructure of the Gd₂O₃-NC films. For AFM measurement, the root-mean-square R_rms was calculated from surface height data z_i obtained from a 1 μm × 1 μm scanning area using numerical integration and the following relationships:

$$\begin{eqnarray*} &&{R}_{\mathrm{rms}}=\left [\frac{1}{N}\sum _{i}^{N}\vert {z}_{i}-\bar {z}\vert ^{2}\right ]^{1/2} \end{eqnarray*}$$

where N is the number of surface height data (i.e. N = 512 × 512 = 262,144) and $\bar {z}$ is the mean-height distance [31]. For electrical analysis, the capacitance–voltage (C–V) hysteresis was measured using an HP4285 precision LCR meter to obtain the memory window. At least five samples were prepared for each measurement.

HRTEM was used to observe the structure of the Gd₂O₃-NCs after amorphous Gd₂O₃ layer deposition followed by RTA and low-damage CF₄ plasma treatment for 10 min. As shown in figure 3(a), the Gd₂O₃ NCs embedded in an amorphous Gd₂O₃ dielectric layer were clearly visualized by HRTEM. The Gd₂O₃ NCs were around 5 nm in size. The grain boundary material between Gd₂O₃-NC and amorphous Gd₂O₃ may exhibit a silicate phase, which was observed in our previous study [32]. Figure 3(b) displays tapping mode 3D AFM images and the estimated surface roughness of Gd₂O₃-NC films for various low-damage plasma treatment times. For treatment times of less than 10 min, the R_rms increased only slightly, which means that the low-damage plasma treatment has almost no effect on the surface roughness. For a treatment time of 30 min, R_rms increased significantly (more than 20%), indicating that some plasma damage occurred at the film surface after lengthy treatment.

Zoom In Zoom Out Reset image size

XPS was used to confirm the incorporation of fluorine atoms into the Gd₂O₃-NCs subjected to low-damage CF₄ plasma treatment. Figure 4 shows XPS spectra of the Gd₂O₃-NCs after treatment for different times in the F 1 s region after using the Shirley method to subtract the background. Peaks were clearly observed, and the peak intensity increased as the treatment time increased, which indicates that fluorine atoms can be introduced into the nanostructure and that the number of incorporated fluorine atoms can be controlled. In comparison with our previous optimized condition for memory characteristics [24], we found that many more fluorine atoms can be incorporated into the Gd₂O₃-NC films, yielding highly fluorinated films. Figure 4 also shows the peak deconvolution with two Gaussian profiles for all XPS spectra at F 1 s. The main peak at a binding energy of ∼685.5 eV corresponds to the formation of Gd-F indicating the fluorination of the Gd₂O₃-NC film. The other peak at a binding energy of ∼688.0 eV with small intensity corresponds to C–F bonding, possibly due to the residue of fluorocarbon during CF₄ plasma treatment. Such a kind of organic residue can be easily removed by high-temperature annealing.

Zoom In Zoom Out Reset image size

Some researchers have reported that fluorine atoms in dielectric films will expand the bandgap owing to their high electronegativity [33, 34]. Therefore, we also investigated the UV absorption to determine the optical E_g of the fluorinated Gd₂O₃-NC films. Figure 5(a) shows the absorption coefficients of the films, calculated from the vacuum UV absorption spectra, as a function of plasma treatment time. The absorption edge of the films clearly depends on the plasma treatment time and shows a large blueshift when the plasma treatment time increases. The result is attributed to the incorporation of more fluorine atoms, which expands E_g because of the high electronegativity of fluorine atoms. To determine the optical E_g of the fluorinated Gd₂O₃-NC films, we used the Tauc formula [35, 36],

$$\begin{eqnarray*} &&(\alpha h v)^{n}=A(h v-{E}_{\mathrm{ g}}), \end{eqnarray*}$$

where A is a constant, h is Planck's constant, v is the frequency, and n is 2 in the case of direct allowed and forbidden electronic transitions. We plotted (αhv)² versus the photon energy (hv), as shown in figure 5(b). These plots have a distinct linear region that indicates the onset of absorption. Therefore, extrapolating this linear region to the abscissa, which represents (hv), yields the E_g value of the fluorinated Gd₂O₃-NC films. Table 1 displays the fitting region of each plasma treatment time and its coefficient of determination (R²) and extracted E_g. We found that the E_g values of fluorinated Gd₂O₃-NC films increased significantly when the plasma treatment time increased. These E_g values of fluorinated Gd₂O₃-NC films treated by low-damage CF₄ plasma treatment are all greater than those obtained after optimized plasma treatment in our previous study because more fluorine atoms can be successfully incorporated into Gd₂O₃-NC films, as shown by the XPS results (see figure 4).

Zoom In Zoom Out Reset image size

Figure 6 shows representative C–V hysteresis curves of fluorinated Gd₂O₃-NC flash memory devices fabricated using low-damage CF₄ plasma treatment. The gate voltage was swept from −10 to 10 V and then back. The hystereses of the fluorinated devices are all significantly wider than that of the untreated device. To quantitatively compare the memory window, all the memory windows were extracted at 0.6C_ox from their respective C–V curves and displayed in table 2. An optimized memory window (∼2.3 V) was obtained for a plasma treatment time of 10 min. In comparison with the sample that did not undergo CF₄ plasma treatment, the memory window improved more than twofold. The results are significantly better than that using a filter-less plasma system. XPS depth profile analysis in our previous study [23, 24] showed that the fluorine intensity was greatest at the Gd₂O₃-NC film surface and decreased with increasing film depth. As mentioned in section 3.1, incorporation of fluorine atoms increases the E_g of Gd₂O₃-NC film. The more fluorine atoms incorporated into the Gd₂O₃-NC film, the larger the E_g obtained. On the basis of this finding, we proposed a gradual nanostructure energy band diagram, i.e., larger E_g near the blocking oxide and smaller E_g near the tunnelling oxide, of Gd₂O₃-NC memory. This is believed to generate a built-in electric field in the Gd₂O₃-NC films, preventing injected charges from the tunnelling oxide from flowing through the blocking oxide to the gate. Consequently, many charges can be stored in the Gd₂O₃-NCs, yielding a larger memory window. Because more fluorine atoms can be incorporated into the Gd₂O₃ layer using our proposed low-damage CF₄ plasma treatment, as shown in figure 4, a higher E_g and thus a larger built-in electric field can be obtained, as demonstrated in figure 5. As a result, it is more difficult for charges to flow through the blocking oxide to the gate, and the memory window is larger. For a treatment time of 30 min, the memory window decreased even though the fluorine intensity was the highest. The reason may be the surface roughness. A rough surface can reportedly induce interface states, lowering the interface barrier [37–40]. Consequently, the leakage current can be enhanced at a rough surface but not at a smooth one. Furthermore, the optimized memory window in our previous study was only 1.03 V (RF power: 50 W; treatment time: 1 min), as indicated by the C–V hysteresis curve shown in figure 6(a). It is noteworthy that the memory window of the sample with the optimized low-damage plasma condition recorded in the current study is further improved by more than twofold over the window obtained under an optimized plasma condition [24]. We also performed CF₄ plasma treatment for longer times and at higher RF powers (50 W for 5 min and 100 W for 1 and 3 min) without the filter to incorporate more fluorine atoms into the Gd₂O₃-NC film; however, the resulting memory devices did not function. This is attributed to plasma damage that caused defect states in the film, leading to leakage of stored charge from the film via trap-assisted tunnelling and eliminating the memory window. In addition, the samples were not treated by annealing after the low-damage CF₄ plasma treatment, which was required in our previous study.

Zoom In Zoom Out Reset image size

Figure 7 shows the room-temperature retention characteristics of the low-damage CF₄-treated Gd₂O₃-NC flash memory devices. For comparison, the characteristic of devices treated by CF₄ plasma without the filter in our previous study [24] are also included. In the case of CF₄ plasma treatment without the filter, the charge loss of the devices improves slightly after 1 min treatment (optimized plasma treatment condition), owing to E_g expansion, and the charge loss becomes significantly worse after 3 min treatment. It is noteworthy that the retention curves can be approximately divided into two regions that have different loss rates. The higher initial charge loss rate is observed from 0 to 1000 s, while a lower charge loss rate is observed from 1000 to 10 000 s. In our previous studies [24, 41], we have concluded that the former region of charge loss curve is associated with charge loss from shallow traps (ST) in the Gd₂O₃-NC film, while the latter one is associated with charge loss from deep traps (DT) via a direct tunnelling to the interface state at the SiO₂/Si interface. In addition, shallow-trap charge loss reportedly results from defect generation in the dielectric layer [42, 43]. The charge loss rates of ST and DT can be obtained by linear fitting in the respective regions. The fitted charge loss rates are shown in table 2. We can clearly see that the charge loss rate after 3 min treatment in the ST region is much higher than that after 1 min treatment, while both charge loss rates in the DT region are almost the same. Because the ST charge loss corresponds to defect generation, we considered that the longer plasma treatment time causes plasma damage leading to many defects in the Gd₂O₃-NC films, resulting in current leakage due to trap-assisted tunnelling [13, 14]. In the case of the low-damage CF₄-treated Gd₂O₃-NC flash memory devices, as the treatment time increased, the charge loss decreased because more fluorine atoms were incorporated into the Gd₂O₃-NC film with minimal defect generation, which not only produced a larger E_g to prevent charge loss through the blocking oxide but also minimized the trap-assisted tunnelling. However, after a 30 min treatment, the charge loss increased. We attributed this to the rough surface, which gives rise to interface states and leakage current, i.e., charge loss. In this study, the optimized treatment time was 10 min. Regardless of treatment time, the charge losses of devices treated with the filter clearly exhibited great improvement and were significantly superior to those of samples treated without the filter. Table 2 shows the charge losses at 10⁴ sec in devices that received plasma treatment with and without the filter. A comparison between both optimized results shows the maximum improvement of more than 60%. As shown in table 2, we can see that the charge loss rate of samples treated with the filter in the region of ST are much smaller than that of samples treated without the filter while the charge loss rates of both kinds of samples are almost the same. We can conclude that the improvement of charge loss is mainly due to the reduction of ST, because low-damage CF₄ plasma treatment causes only minimal plasma damage, i.e., minimal defect generation, in the Gd₂O₃-NC films, which can maximize the effect of fluorination of the films to obtain large E_g expansion and little charge loss. These results further prove that low-damage CF₄ plasma treatment during the fabrication of fluorinated Gd₂O₃-NC flash memory devices can be realized using our proposed method.

Zoom In Zoom Out Reset image size