Nanocrystal memory employing distributed nanodots as storage elements instead of the continuous polysilicon floating gate in conventional flash memory structures has been a promising candidate for replacing conventional nonvolatile memory (NVM). Due to discrete storage nodes, continuous charge leakage caused by localized oxide defects will not happen. This allows the more aggressive scaling of the tunnel oxide to achieve lower operating voltages and faster write/erase speeds for a high performance NVM. Compared with the semiconductor counterparts, metal NCs exhibit several advantages, such as a wide range of available work functions, higher density of states around the Fermi level, stronger coupling with the conduction channel, and smaller energy perturbation due to carrier confinement.^[1,2] Especially, the noble metals do not oxidize nor react with the surrounding dielectric layers.^[3] For these reasons, Au NCs receive a lot of attention. In addition, the high dielectric constant (high-k, k > 3.9) materials instead of SiO₂ (k ∼ 3.9) as the dielectric layers could also be utilized to enhance the memory performance. The use of the high-k materials as a blocking layer makes the voltage drop across the blocking layer lower, and more electric field will fall on the tunneling oxide,^[3] thus reducing the write/erase voltage of the device. Furthermore, high-k blocking layers offer a smaller equivalent oxide thickness (EOT) with relatively thick film, which is important for leakage suppression and device scaling. Among the high-k candidates, HfO₂ has been spotlighted due to its high dielectric constant (∼ 25), large band gap (∼ 5.6 eV), and thermodynamic stability.^[4,5] However, it was reported that oxygen vacancy (V_O) is dominant over an intrinsic defect, and is also considered to present a high concentration in HfO₂.^[4,6,7] The V_O levels in HfO₂ are located at about 1.2 eV below the bottom of the conduction bands, which has a crucial relation with electron leakage current^[8,9] and will degrade the performance of the device.

Recently, nitrogen incorporation has been investigated in ultrathin HfO₂ dielectric film to passivate the oxygen vacancy states. Several approaches, such as high temperature annealing in NH₃ ambient,^[10] in situ nitrogen incorporation using ALD by adding nitrogen plasma^[11] and the reactive sputtering method followed by a reoxidation anneal^[12] have been reported. These studies showed that the nitrogen incorporation provides some improvement in device characteristics, but there still remain some drawbacks. High temperature annealing may cause HfO₂ to crystallize for its crystallization temperature is rather low (∼400 °C). Grain boundaries in crystallized gate dielectric films may act as oxygen or dopant diffusion paths, leading to a significant increase in leakage current and making EOT scaling problematic. The NH₃ is toxic gas, and it can also introduce hydrogen-related traps (–H, –OH, and N–H), which will cause a degradation in the device stability. Moreover, post-deposition annealing may also affect the NCs in the memory.

In this paper, a simple method of incorporating nitrogen into HfO₂ by post-N₂ plasma without any post annealing is investigated. N₂-plasma treatment proves to be a feasible and simple fabrication process with low thermal budget to improve the electrical properties of the HfO₂ dielectric. The N₂-plasma treatment is also introduced into Au nanocrystal memory with HfO₂ blocking layers resulting in larger memory windows and better retention characteristics than that without N₂-plasma treatment. It is indicated that the simple method of N₂-plasma treatment promises to be applied to future nonvolatile memory.

A p-type Si(100) wafer with a resistivity of 1 Ω·cm–10 Ω·cm was used and cleaned with the standard RCA process, followed by a dry oxidation process at 900 °C to form a ∼ 5-nm-thick SiO₂ layer as a tunnel oxide. Subsequently, a thin layer of Au was deposited on the tunnel oxide using radio frequency (RF) magnetron sputtering at a relatively low power of 40 W. The Au NCs were formed after rapid thermal annealing at 600 °C for 60 s in N₂ ambient. Afterwards, a 21-nm-thick HfO₂ layer was deposited onto the tunnel oxide by electron-beam evaporation as the blocking oxide layer. Then, the N₂-plasma treatment was performed on the as-deposited HfO₂ under an optimized plasma processing condition (90 W for 10 min with 40-sccm N₂) by using reactive ion etching. Finally, the sputter-deposited 50-nm TaN and 300-nm Al gate electrodes were both patterned on the HfO₂ film by a shadow mask, while the 300-nm Al was sputtered as the back electrode. For comparison, the Au nanocrystal memory capacitor without N₂-plasma treatment was also fabricated. Moreover, the Al–TaN/HfO₂/SiO₂/p–Si MOS capacitors without Au NCs were also fabricated by using the same processing steps to investigate the properties of the HfO₂ films before and after N₂-plasma treatment.

The nitrogen incorporation into HfO₂ film was confirmed by x-ray photoelectron spectroscopy (XPS). The HfO₂ morphology was determined by image processing of atomic force microscopy (AFM). High-resolution cross-sectional transmission electron microscopy (HRTEM) analyses were performed to investigate the structure of the Au nanocrystal memory capacitor with N₂-plasma treatment. Capacitance–voltage (C–V) and current–voltage (I–V) characteristics were determined by using a Keithley 4200 semiconductor parameter analyzer with an HP4284 LCR meter. The C–V curves were measured at an applied frequency of 1 MHz.

Prior to the XPS analyses, about 2-nm-thick surfaces of the samples are removed by 4-keV Ar ion bombardment, thereby removing contaminants from the surfaces. The XPS depth profiles of the sample of ∼ 21-nm-thick HfO₂ film on the SiO₂/p-Si substrate with and without N₂-plasma treatment are shown in Fig. 1. From Fig. 1(a), we can see that no N 1s peak can be detected from the sample without N₂-plasma treatment. However, an obvious N 1s peak near 396.3 eV is detected for the sample after N₂-plasma treatment in an optimized condition, showing the formation of N–Hf bonds.^[12] The intensity of the N 1s peak is strongest in the middle area, and decreases with the depth increasing to the top area or decreasing to the bottom area. Figure 1(b) shows the XPS of the Hf 4f peak from HfO₂ film with and without N₂-plasma treatment in the middle area. The well-resolved transition metal core level doublet spectrum (Hf 4f_7/2 and Hf 4f_5/2) of the sample corresponds to as-deposited HfO₂ while sample after N₂-plasma treatment has a relatively poorly resolved spectrum. The spectrum acquired from the as-deposited HfO₂ shows an Hf 4f_7/2 peak at a binding energy of 17.1 eV, which corresponds to the Hf–O bond in bulk HfO₂^[13] as shown in Fig. 1(b). The main peak of the sample with N₂-plasma treatment shifts to a lower binding energy of about 16.9 eV due to the presence of nitrogen in the film.^[14] The modeling associated with achieving a very good fit with the measured spectrum for the sample with N₂-plasma in Fig. 1(b) indicates the contribution of Hf–N bond-related peaks at 15.8 eV, 17.8 eV in Hf 4f to the resultant spectrum.^[15] The results clearly indicate that a trace quantity of N element is introduced into the HfO₂ thin film and an oxygen vacancy might be occupied by an N atom after N₂-plasma treatment.^[9,12,14]

Fig. 1.

	Figure Option ViewDownloadNew Window
	Fig. 1. (color online)XPS spectra of (a) N 1s and (b) Hf 4f from HfO2 film deposited by e-beam in the cases with and without N2-plasma treatment.

Figure 2(a) shows the C–V measurement results of the Al–TaN/HfO₂/SiO₂/p-Si MOS capacitors with and without N₂-plasma treatment on HfO₂ film, respectively. The gate voltage is swept from accumulation to inversion and back to check the magnitude of hysteresis. The capacitor without N₂-plasma treatment exhibits a large hysteresis of about 1.3 V at ± 6 V, suggesting a higher density of charge trapping centers. In contrast, hysteresis is significantly reduced to 0.2 V in the plasma-nitrided HfO₂ MOS capacitor in the same sweeping range. It indicates that the N₂-plasma treatment on HfO₂ film can effectively reduce the density of charge trapping centers such as oxygen vacancies, which results in hysteresis. In addition, the larger accumulation capacitance of the capacitor after N₂-plasma treatment suggests the improved quality of HfO₂ and better electrical properties of the MOS capacitor. The leakage current characteristics of the MOS capacitors with and without N₂-plasma treatment are shown in Fig. 2(b). The leakage current density of gate injection for the capacitor with N₂-plasma treatment is about 4.1 × 10⁻⁷ A/cm² at −1-V applied voltage, which is one order of magnitude lower than that without N₂-plasma (about 3.1 × 10⁻⁶ A/cm²). The leakage current characteristic is improved by using the N₂-plasma process, which is believed to be due to the fact that the incorporated nitrogen atoms can occupy the oxygen vacancies, reduce the number of traps, and hinder the conduction of leakage electrons,^[8,9] which is in accordance with the result of the C–V measurement. After N₂-plasma treatment, the surface quality of HfO₂ can also be improved, which is confirmed by analyzing the surface morphology of the HfO₂ film on SiO₂/p-Si by AFM as shown in Fig. 3. As we can see, the root-mean-square (RMS) surface roughness of the as-deposited sample is about 0.45 nm, but it greatly reduces to 0.14 nm after N₂-plasma treatment. It was reported that the surface roughness can cause an additional potential across the insulating film, which increases the average electric field inside the film, resulting in the increase of the leakage current density.^[16] As a consequence, the smooth surface of HfO₂ after N₂-plasma treatment may also contribute to the reduction of leakage current density.

Fig. 2.

	Figure Option ViewDownloadNew Window
	Fig. 2. (color online)(a) Capacitance-voltage and (b) current-voltage characteristics of MOS capacitors with and without N2-plasma treatment on HfO2 film.

Fig. 3.

	Figure Option ViewDownloadNew Window
	Fig. 3. (color online) Tapping mode AFM 3D pictures of the HfO2 film on SiO2/p-Si with and without N2-plasma treatment. The surface roughness is also estimated from the RMS value.

Figure 4 shows the cross-sectional high-resolution transmission electron microscopy (HRTEM) micrograph of Au nanocrystal memory capacitor with N₂-plasma treatment. The Au NCs formed on the 5-nm thermal SiO₂ are covered with a 21-nm HfO₂ layer. The NC density is approximately 8.4 × 10¹¹ cm⁻², wherein the mean size is about 5 nm. The HRTEM of the memory capacitor without N₂-plasma treatment shows a similar NC distribution and size (not shown here). It indicates that the post-N₂-plasma treatment rarely affects the formation of Au NC.

Fig. 4.

	Figure Option ViewDownloadNew Window
	Fig. 4. (color online) Cross-sectional HRTEM micrograph of Au NCs memory capacitor with N2-plasma treatment.

Figures 5(a) and 5(b) show the C–V hysteresis after bidirectional sweeps in various voltage ranges, which compares with the memory characteristics of the two kinds of Au nanocrystal memory capacitors with and without N₂-plasma treatment on the HfO₂ blocking layer. The flat-band voltage shift and memory window are extracted and shown in Figs. 5(c) and 5(d), respectively. As we can see, the memory windows of the devices all increase almost linearly with applied voltage due to the observed flat-band voltage shifts at both bias polarities, which confirms successive charging by electrons (positive bias) and holes (negative bias).^[17,18] Nevertheless, the flat-band voltage shifts at either bias polarity of the memory capacitor with N₂-plasma treatment are higher throughout, leading to a much higher memory window and demonstrating the larger storage capacity. A 10.2-V memory window is obtained for the memory capacitor with N₂-plasma treatment at ±12-V sweep. Compared with other reported results,^[3,19,20] the memory window here is not the best result. However, it is twice as large as that without N₂-plasma treatment in the same applied voltage range, which is a big improvement. According to Fig. 2(a), negligible hysteresis windows are measured for the MOS capacitor with N₂-plasma treatment, indicating that almost all the charges are stored in the Au NCs.^[20] On the contrary, for the memory capacitor without N₂-plasma treatment, except Au NCs, part of charge can also be trapped in the HfO₂ blocking layer, which can contribute to the oxygen vacancy existing in HfO₂. The charge storage density can be calculated from the relation N = ΔV_FBC/q,^[3,18] where C is the capacitance density and ΔV_FB is the flat-band voltage shift. On the basis of this relation, the negative and positive charge densities trapped in the memory capacitor with N₂-plasma treatment at ±12 V are estimated at 1.7 × 10¹³/cm² and 1.8 × 10¹³/cm² respectively, while for the memory capacitor without N₂-plasma treatment are 2.1 × 10¹²/cm² and 1.1 × 10¹³/cm² respectively. The charge densities of the latter are much lower, although part of charges can also be trapped in the HfO₂ film. It confirms that the significant leakage passages exist in the HfO₂ film without N₂-plasma treatment. Charges stored in Au NCs may easily leak out via the trapping centers.^[21] Moreover, after N₂-plasma treatment, the oxide quality surrounding the Au NCs can be improved and de-trapping via traps is suppressed, which is consistent with the results as shown in Fig. 2. Saturation of the flat-band voltage shifts and memory window set in when the applied voltage range reaches to ±12 V for the N₂-plasma treated memory capacitor, which could be attributed to the Coulomb blockade effect originating from the field induced by the stored charge.^[22] It should be noted that the memory window compared with other reported results may not be the largest in this study. However, the goal of this work is to clarify the significant improvement of the NVM performance by the simple method of N₂-plasma treatment on the HfO₂ blocking layer. In addition, further optimization, such as the mean size and number density of Au NCs, can be carried out to improve the performances of these memory devices.

Fig. 5.

	Figure Option ViewDownloadNew Window
	Fig. 5. (color online) Normalized C–V characteristics for different voltage sweeps of Au NCs memory (a) with and (b) without N2-plasma treatment on HfO2 blocking layer. (c) Evolution of flat-band voltage shift and (d) total memory window as a function of absolute gate bias.

The retention characteristics of the memory capacitors with and without N₂-plasma at room temperature, extrapolated to 10 years, are shown in Fig. 6. The capacitors are programmed first with a drive voltage of +7 V for 1 s. Next, the C–V is measured with an applied voltage of +7 V to −0.5 V and the flat band voltage is extracted. The memory capacitor is then erased under −7 V for 1 s and the C–V characteristics are measured at different time intervals in a voltage range from −7 V to 0.5 V. The C–V measurements are repeated at different time intervals up to 10⁵ s for both cases. Extrapolation to 10 years yields charge losses of about 19% and 32% for the memory capacitors with and without N₂-plasma, respectively. The retention properties of the sample with N₂ plasma treatment are superior. In general, if more charges are stored in the traps rather than in the nanocrystals, the more serious degradation of memory window can be expected. As we can see from Fig. 2, the smaller hysteresis and the lower leakage current of the MOS capacitor fulfilled by using the N₂-plasma process suggest much less traps existing in HfO₂. Therefore, the improved retention characteristic is believed to be due to the fact that the N₂ plasma treatment can repair defects and improve the surrounding HfO₂ quality of Au nanocrystals.

Fig. 6.

	Figure Option ViewDownloadNew Window
	Fig. 6. (color online) Retention characteristics of memory capacitors with and without N2-plasma at room temperature.

In this paper, we have demonstrated the N₂-plasma treatment on the HfO₂ blocking layer of Au nanocrystal nonvolatile memory. After N₂-plasma treatment, the nitrogen atoms are incorporated into HfO₂ film and might passivate the oxygen vacancy states, leading to the reduction of traps. The surface roughness of HfO₂ film can also be reduced after N₂-plasma treatment. Those improvements of HfO₂ film lead to a smaller hysteresis and lower leakage current density of the MOS capacitor. The N₂-plasma is introduced into the Au nanocrystal nonvolatile memory to treat the HfO₂ blocking layer. For the N₂-plasma treated memory, it shows a larger memory window and better retention characteristics. Therefore, Au nanocrystal memory with suitable nitrogen incorporation promise to be used in future nonvolatile memory devices.