Atomic Layer Deposition of High Quality HfO 2 Using In-Situ Formed Hydrophilic Oxide as an Interfacial Layer

High-quality HfO 2 cannot be grown directly on Si substrate using atomic layer deposition (ALD), and an interfacial oxide layer is needed. Traditionally, interfacial oxide layer is formed either in SC1 solution (2 NH 4 OH: 4 H 2 O 2 : 200 H 2 O) or by ozonated water spraying. A highly hydrophilic SiO 2 interfacial layer was in-situ formed in the ALD chamber using 1 cycle of ozone and water. The HfO 2 deposited on this interfacial layer showed great growth linearity. The gate leakage current is comparable to that formed using chemical oxide as the interfacial layer. The capacitance-voltage (C-V) curves have negligible frequency dispersion and hysteresis, which suggest high quality in both the interface and electrical properties. The in-situ formation of hydrophilic interfacial layer have advantages over the traditional interfacial layer. This might be useful for formation of interfacial layer on sophisticated 3-D MOS structures such as FinFETs and nanowire FETs. In addition, the chemical oxidation step can be eliminated from the integrated circuits manufacturing processes, which is economically beneﬁcial to the industry.

After a decade of investigation, atomic layer deposition (ALD) of high dielectric constant (high-k) material HfO 2 became the mainstream manufacturing processing in semiconductor industry.This is because the superior quality of HfO 2 films obtained by ALD.ALD provides precise control of film thickness and uniformity at the atomic scale, due to its self-limiting surface reaction.However, not all surfaces are suitable for ALD deposition.
Extensive work has been done to explore the proper interface conditions for ALD of metal oxides.][6][7][8][9][10][11] Plus, underlying SiO 2 would grow at the Si/metal oxide interface at moderate temperature, which is harmful to the scaling of MOS devices. 12,13This has been proved by infrared spectroscopy. 14There are also researches on nucleation of HfO 2 on Si (110) and Si (111), 15 and Ge surface, 16,17 which is a potential substrate material with high mobility.L. Nyns, etc. further explored the initial few cycles of the ALD process on H-terminated, OH-terminated Si surface and different Si orientations.
High-quality HfO 2 can only be grown using ALD on hydrophilic surface, i.e.OH terminated surface.][20][21][22] It was found that the density of -OH groups on Si surface has an effect on the atomic surface roughness and dielectric leakage current of HfO 2 .Green et al. demonstrated that chemical oxide grown by SC1 solution (2NH 4 OH:5H 2 O 2 :200H 2 O) is full of -OH groups and works effectively as an interfacial layer for ALD growth of HfO 2 . 23zone based wet chemical oxidation and ozonated water spraying are widely used in industry for ALD growth of HfO 2 . 24,25The growth of the interfacial layer was controlled by ozone concentration in water.However, in 3-D silicon MOS devices structures, such as FinFETs and nanowire FETs, it is difficult to allow uniform oxide formation on the sides of the fins and underneath the nanowire in wet chemical solution or by spraying.
In this work, we report an ALD-based in-situ formation of SiO 2 interfacial layer using one cycle of ozone and water.The interfacial layer formed is highly hydrophilic, and the ALD HfO 2 grown on this interfacial layer has comparable qualities to HfO 2 grown on SC1 chemical oxide.The interfacial layer was in-situ formed in the ALD chamber.This method might be used in sophisticated 3-D MOS structures because ALD allows molecules to be deposited on the surface that cannot be accessible using other methods.In addition, the chemical oxidation step can be eliminated from the integrated circuits manufacturing processes, which is economically beneficial to the industry.

Experimental
Boron doped p type Si(100) wafers with a resistivity of 1-20 cm were used as substrates.The substrates were cleaned using Radio Corporation of America (RCA) cleaning to remove organic and ionic contaminations.H-terminated Si surface was achieved by immersing Si substrates into BOE (buffered oxide etch) solution.BOE etching is preferred to diluted HF etching because Si surface etched by BOE is more smooth.After BOE etching, Si substrate was loaded into ALD chamber (Cambridge Nanotech Inc) immediately at 100 • C. Then the surface was exposed to 1 ALD cycle of ozone (O 3 ) and deionized water (DI H 2 O), which were introduced into ALD chamber sequentially with ozone coming first.O 3 was produced from O 2 by an ozone generator (A2Z Ozone Inc.).It was found that after this process, an ultrathin layer of SiO 2 was thermally grown on Si substrate, and the surface of the SiO 2 is highly hydrophilic.Usually Si could not be effectively oxidized by O 2 at temperature below 500 • C; O 3 is a strong oxidant, and can grow desired ultrathin oxide layer with only one pulse.The H 2 O pulse was supplied to provide -OH groups on the interfacial layer surface.The details of this process are listed as following: O 3 pulse time is 0.3 second, exposure time is 30 seconds, pump time is 23 seconds; H 2 O pulse time is 0.5 second, exposure time is 15 seconds, pump time is 20 seconds.The pulse and exposure times of ozone and DI water were carefully chosen to minimize the thickness of the SiO 2 layer, so that the equivalent oxide thickness (EOT) of the gate dielectric stack is minimal.Without taking the sample out of the chamber, HfO 2 was deposited on this interfacial layer using tetrakis(dimethylamino)hafnium (TDMAH) and DI water at 300 • C, and the pulse times were chosen to meet the saturation requirement of ALD (TDMAH and H 2 O pulse time: 0.5 second.Exposure time: 0 second.Pump time: 5 seconds).It took about 20 minutes to increase the chamber temperature from 100 • C to 300 • C in our ALD system.As a control sample, chemical oxide was also grown using SC1 solution (NH 4 OH:H 2 O 2 :H 2 O = 2:4:200) at 60 • C for 2 minutes, and HfO 2 was deposited using the same parameters as the ones used with the in-situ formed interfacial layer.This component ratio of SC1 solution was developed by M. L. Green, etc., 23 and was modified in our previous work to be an effective method to grow interfacial layer for ALD of high quality HfO 2 . 26Multi-angle spectroscopic ellipsometry (J. A. Woollam M3000V) was used to measure the physical thickness and characterize the optical properties of the thin films.
MOS capacitors were fabricated to electrically evaluate the effectiveness of the interfacial layer/HfO 2 stacks as gate dielectric materials.Titanium (Ti, 100 Å) and nickel (Ni, 1000 Å) were deposited by electron-beam evaporation as gate metals, because Ti enhances adhesion between the gate metal and the dielectric stacks.The metal gates were patterned by photolithography and wet etching.The gate patterns were circles and the diameter was 100 μm.The back side of the Si substrates was coated by aluminum (1000 Å).The samples were then annealed at 450 • C in forming gas (N 2 :H 2 = 10:1) for 30 minutes to create Ohmic contact between Al and bulk Si.The capacitance-voltage (C-V) curves of the MOS capacitors were measured using an Agilent 4284A LCR meter at multi-frequencies, and the current-voltage (I-V) curves were measured using an Agilent 4155B semiconductor parameter analyzer.Equivalent oxide thickness (EOT) and flatband voltage (V FB ) were extracted by fitting the 100 KHz C-V curves with the theoretical quasi-static C-V simulation (UC Berkeley's quantum-mechanical C-V simulator).

Results and Discussion
Characterization of the in-situ formed interfacial layer.-Thegrowth of chemical oxide in SC1 solution has been extensively investigated in our laboratory before. 26The physical thickness of the ultrathin chemical oxide grown at the condition described above was ∼4.5 Å.The chemical oxide thin film was full of -OH groups and highly hydrophilic on the surface.Our in-situ formed SiO 2 -based interfacial layer grown by 1 ALD cycle of O 3 and H 2 O had similar hydrophilic surface as chemical oxide based on observation.The physical thickness of the in-situ formed interfacial layer was ∼3.5 Å measured by spectroscopic ellipsometry, which is ∼1 Å thinner than that of the chemical oxide.This is beneficial to further scaling of the gate dielectric stack.Usually H-terminated Si surface would only be oxidized at elevated temperature (>600 • C), not at the temperature (100 • C) for our in-situ interfacial layer formation process.The Htermination of Si surface could remain for about 10 minutes in air at room temperature.The successful growth of our in-situ interfacial layer is because O 3 is a stronger oxidant than O 2 , and the oxygen atoms dissociated from O 3 molecules directly attack the back bonds of Si. 27 The thin oxide film grown by O 3 was also reported with a stable Si-O-Si network even formed at low temperature. 27The mean square error (MSE) of the spectroscopic ellipsometry measurements of the chemical oxide and the in-situ interfacial layer were comparable.This suggests that the theoretically calculated ideal SiO 2 model fits well with these SiO 2 -based interfacial layers.Therefore, the qualities of our newly developed in-situ interfacial layer is not degraded in its oxide properties comparing to the chemical oxide.
The interfacial layers prepared by different methods before depositing HfO 2 were characterized by the Thermo Scientific K-Alpha X-ray Photoelectron Spectrometer (XPS) using an Aluminum Kα micro-focused monochromatic X-ray source with a 400 μm spot size.The depth profile was obtained by an Ar Ion gun of 1000 eV energy with 5 seconds for each layer (the estimated etching speed is 0.5 nm/s for Ta 2 O 5 ).away from the ideal 1:2 of SiO 2 and the Si amount is much higher.This phenomenon could be attributed to following reasons: (1) the interfacial layers are ultrathin (<5 Å), X-ray could easily penetrate through, and the signal would be affected by the Si substrate.(2) Si substrate cannot be fully oxidized by either the chemical oxidation method or the O 3 + H 2 O method.(3) Estimation of atomic percentage by XPS is a rough method, the results could be significantly affected by contamination and equipment limitation.From layer 1 to layer 3, the Si:O ratio increases faster in O 3 and H 2 O formed interfacial layer, which is because the O 3 and H 2 O formed interfacial layer is thinner than the chemical oxide, so the X-ray arrives the Si substrate sooner.
Figure 2 displays the oxygen 1s spectrum of the surface of interfacial layers from different treatments before depositing HfO 2 .Figure 2a is from chemical oxide interface, Figure 2b is from O 3 + H 2 O interface, and Figure 2c is from the surface of Si after only one pulse of O 3 .Our O-1s spectrum show that the main chemical states of oxygen on the surface are hydroxyl (∼531.5 eV), 28 Si-O(∼532.5eV) 28 and H 2 O (∼534 eV), 29 but their amounts (the area under each peak) on the surfaces vary between different preparation methods.
Comparing chemical oxide (Figure 2a) and O 3 + H 2 O (Figure 2b) methods, the amount of -OH group on the surface interface from O 3 + H 2 O method (area under -OH peak : area under Si-O peak is 0.24) is smaller than that on the interface from chemical oxide method (area under -OH peak : area under Si-O peak is 0.57).Introducing H 2 O in the end of O 3 process is necessary, since the amount of -OH group on the interface from O 3 + H 2 O process is increased comparing with O 3 only method (area under -OH peak : area under Si-O peak is 0.17).
Growth linearity of ALD HfO 2 on in-situ formed interfacial layer.-Afterformation of the interface layer, the substrate temperature was increased to 300 • C.During the temperature ramping up, the samples remained in the ALD vacuum chamber, so there was no unwanted SiO 2 re-growth.Therefore, it is an in-situ deposition process.At 300 • C, HfO 2 was deposited on the samples by ALD for 15, 21, 24, 35, 50, 100, 150 and 200 cycles.The physical thicknesses were measured using spectroscopic ellipsometry as shown in Figure 3a with square marks.The measurement error is within 3% of the measured thickness.The growth of HfO 2 shows excellent linearity even during the starting cycls, suggesting a stable atomic-layer-by-atomic-layer deposition.The deposition rate of ∼1 Å/cycle was extracted by calculating the slope of the straight growth line.MOS capacitors were fabricated using the interfacial layer/HfO 2 stacks as the gate dielectric.The EOTs of the gate dielectric stacks were extracted from the C-V curves measured at high frequency (100 KHz), which were plotted in Figure 3a   the measured devices.The linearity of EOT vs. ALD cycle numbers will be discussed in details later.Figure 3b shows the physical thickness of HfO 2 deposited for 1, 2, 3, 4, 5, 10 and 30 cycles on chemical oxide interface and O 3 + H 2 O interface.During 1-5 cycles, the growth of HfO 2 is not linear.The HfO 2 growth tracks on these two interfacial layers are similar, which implies the O 3 + H 2 O interface is as effective as the chemical oxide interface.The measurement error in this thickness range is ∼0.5 Å, and the error bar would mix together with the data point mark, so it is not shown in Figure 3b.Electrical characterization of MOS capacitors.-Figures 4 and 5 are the high frequency (100 KHz) Capacitance density-Voltage (C-V) curves of the MOS capacitors using the in-situ formed interfacial layer/ALD HfO 2 as the gate dielectric stacks.HfO 2 was deposited for 21 cycles on the sample shown in Figure 4, and 24 cycles on the sample shown in Figure 5.In Figure 4, the C-V curve of the MOS capacitor using ALD HfO 2 deposited for 21 cycles on the chemical oxide is also plotted as a reference.The dots are the measurement data (circles are data from the chemical oxide sample and squares are data from the in-situ formed interfacial layer sample), and the solid curves are the theoretical quasi-static C-V simulation that takes quantum effects into consideration.In all the C-V results, the deviation of capacitance density is under 5 × 10 −8 F/cm 2 .This is too small to be drawn as an error bar, since the error bar would mix together with the data point.The fitting of the simulation to the measurement data is precise in the depletion region and in the initial part of the accumulation region.In the high-voltage part of the accumulation region, the measurement data have deformation and the capacitance value drops.This is because the interfacial layer plus the 21-cycle-ALD HfO 2 stack is ultrathin with its physical thickness of <3 nm, so that the direct tunneling current through the dielectric stack is high, which causes C-V deformation in the high gate voltage region. 30By fitting the measurement data with simulation, we extracted the flatband voltage (V FB ) and the EOT of the capacitors.The physical thicknesses of the HfO 2 thin films are also marked in the Figures.The physical thicknesses of the HfO 2 films on these two samples are 2.59 nm and 2.79 nm, separately, and the corresponding EOTs are 0.86 nm and 0.98 nm.The HfO 2 physical thickness of the in-situ formed interfacial layer sample is 2 Å thinner than that of the chemical oxide sample, which was caused by experiment variation.The 2 Å difference in physical thickness corresponds to 0.4 Å difference in EOT, but the EOT of the in-situ interfacial layer sample is 1.2 Å thinner than that of the chemical oxide sample.This extra 0.8 Å less EOT should be attributed to the thinner interfacial layer comparing to chemical oxide.
The C-V curve of the MOS capacitors using the in-situ formed interfacial layer plus the 24-cycle-ALD HfO 2 gate stack is plotted in Figure 5.The physical thickness of HfO 2 on this sample is almost the same as that of the chemical oxide sample, and the EOT is slightly thinner (0.4 Å) than that of the chemical oxide sample.The thickness of interfacial layers is slightly less than 5 Å.The EOTs of the samples in both Figures 4 and 5 suggest that the quality of HfO 2 deposited on the in-situ formed interfacial layer is comparable to that deposited on the chemical oxide.
MOS capacitors were fabricated using samples with 35, 50 and 100 ALD cycles of HfO 2 .The samples were analyzed in the same way as described above, and the EOT vs. cycle number is plotted in Figure 3a with the triangle mark.The C-V curves of the capacitors using HfO 2 with cycle numbers of <21 have large deformation in the accumulation region due to large leakage current.The linearity of EOT vs. ALD cycle numbers is excellent, suggesting that it is a true ALD process.Figure 6 shows the Current density-Voltage (J-V) curves of the capacitors discussed above.The J-V curve of the chemical oxide sample is also plotted as a reference.In all the J-V results, the deviation of the current density is within half order.The uniformity is pretty good among all the measured devices.The gate leakage current is evaluated at V G = V FB + 1V.At this gate voltage, the leakage current of the chemical oxide sample is 2.65 × 10 −2 A/cm 2 , which matches with the previously reported results. 31,32The leakage current of the in-situ interfacial layer sample with 21-cycle-ALD HfO 2 is 0.2 A/cm 2 , which is 1-order-of-magnitude higher than that of the chemical oxide sample.This is because the EOT of the in-situ interfacial layer sample is 1.2 Å thinner than that of the chemical oxide sample.The physical thickness and EOT of the 24 cycles ALD HfO 2 sample are almost the same as those of the chemical oxide sample, and its gate leakage current is 5.32 × 10 −2 A/cm 2 , which is also comparable to that of the chemical oxide sample.The gate leakage current of HfO 2 grown by ALD could be affected by the interfacial layer because the initial nucleation is dependent on the interfacial layer.If the interfacial layer is lack of -OH groups, or is not hydrophilic, there are not enough sites on the sample surface for molecules of TDMAH to attach to, so that nucleation is disturbed.In this case, the molecules might be stacked on each other randomly, resulting in HfO 2 films with non-stoichiometry and defects.The gate leakage current of the dielectric stacks using the in-situ formed interfacial layers comparable to that using chemical oxide interfacial layer suggests that this method is useful.
Frequency dispersion and hysteresis of MOS capacitors.-Inorder to further understand the trap charges in the interface and in the HfO 2 thin film, we carried out measurement of frequency dispersion and hysteresis behaviors of the C-V curves, the results are shown in Figure 7.The chemical oxide/21-cycles-ALD HfO 2 sample was used as a reference, and the in-situ interfacial layer/24-cycle-ALD HfO 2 sample was studied, because the two samples have the same EOT.Figure 7a shows three C-V curves of MOS capacitor on the chemical oxide sample, measured at 10 KHz, 100 KHz and 1 MHz, separately.The three curves overlap with each other in depletion region with the same V FB .The C-V curve in accumulation region at 1 MHz does not match with the curves at 10 kHz and 100 kHz.4][35] The samples were kept in air for months before we measured the C-V frequency dispersion, and a thin layer of Ni x O y would have already formed on the Ni gate surface, which increases the gate series resistance.The C-V curves at 10 kHz and 100 kHz overlapping with each other suggests that there is no frequency dispersion and the interfacial quality is very good.Figure 7b shows the C-V hysteresis behavior of the same capacitor at 100 KHz.The solid curve in the figure was measured from depletion to accumulation, and the dash curve was measured from accumulation to depletion.The two curves overlap with each other and the V FB shift is negligible, suggesting the mobile charge in the HfO 2 thin film is small.Figures 7c and 7d are the C-V frequency dispersion and hysteresis behavior of the insitu interfacial layer/ALD HfO 2 sample.Figures 7c and 7d show C-V curves similar to those in Figures 7a and 7b: the C-V curves measured at different frequencies are all smooth; the C-V curves measured at different frequencies have the same V FB ; the hysteresis behavior is negligible.Therefore, we conclude that the ALD in-situ formed interfacial layer provides comparable interfacial quality and dielectric stack quality to that of the chemical oxide interface.

Conclusions
In summary, a highly hydrophilic SiO 2 -based interfacial layer was in-situ formed in the ALD chamber using 1 cycle of O 3 and DI H 2 O. HfO 2 was deposited on this interfacial layer using a conventional ALD process.Studies of interfacial layer characterization, HfO 2 growth linearity, EOT of the stack, gate leakage current, and frequency dispersion and hysteresis behavior of the C-V curves suggest that the in-situ formed interfacial layer can result in the same high-quality HfO 2 dielectric layers as the chemical oxide interfacial layer.Using this approach, the wet chemical oxidation step can be eliminated from the processes of deposition of high-k gate stacks.

Figure 1 Figure 1 .
Figure 1.The main atomic percentage change of main components (Si and O) of chemical oxide interface and O 3 + H 2 O interface from XPS depth profiling (layer1: the bare surface, layer3: the most in-depth layer).

Figure 4 .
Figure 4. Experimental (dots) and simulated (lines) C-V curves of Ni/Ti/HfO 2 /interfacial layer/p-Si MOS capacitors.The experimental C-V curves were measured at 100 KHz.The interfacial layers were grown by SC1 chemical oxidation and 1 ALD cycle of O 3 and DI H 2 O at 100 • C, separately.The HfO 2 films were deposited for 21 cycles.

Figure 5 .
Figure 5. Experimental (dots) and simulated (lines) C-V curves of Ni/Ti/HfO 2 /interfacial layer/p-Si MOS capacitors.The interfacial layers were grown by 1 ALD cycle of O 3 and H 2 O at 100 • C. The HfO 2 were deposited for 24 cycles.

Figure 6 .
Figure 6.Leakage current density curves of the same MOS capacitors as shown in Figures 2 and 3.

Figure 7 .
Figure 7. (a) Frequency dispersion (10 KHz, 100 KHz, 1 MHz) of C-V curve of MOS capacitor using chemical oxide/21-cycle-ALD HfO 2 as gate dielectric.(b) Hysteresis behavior of 100 KHz C-V curve of the MOS capacitor in Figure 5a.(c) Frequency dispersion (10 KHz, 100 KHz, 1 MHz) of C-V curve of MOS capacitor using ALD O 3 + H 2 O interface/24-cycle-ALD HfO 2 as gate dielectric.(d) Hysteresis behavior of 100 KHz C-V curve of the MOS capacitor in Figure 5c.
with triangle marks.Several devices were measured for each number of ALD cycles, and the EOTs have variation <2 Å among all