Sodium impurities in amorphous and crystalline Ta2O5

Tantalum pentoxide (Ta₂O₅) thin films are considered highly promising for ion-sensitive field-effect transistors, dynamic random-access memory, and various actuator technologies [1‐5]. Achieving reliable device performance, however, requires well-controlled deposition processes and strict management of electrically active defects formed during growth or subsequent thermal treatments. These defects may arise from intrinsic point defects such as vacancies and interstitials, or from extrinsic impurities unintentionally introduced during sputtering or annealing. Identifying and quantifying such defects in Ta₂O₅ remain challenging for several reasons.

First, as-deposited Ta₂O₅ is typically amorphous, while crystallization begins at ~ 750–800 °C and can lead to multiple polymorphs depending on the annealing conditions. Because defect formation energies and mobilities differ strongly between amorphous and crystalline phases, the defect landscape varies considerably. Second, defect densities in high-k oxides are often below the detection limits of conventional techniques. Optical methods commonly used in semiconductor defect studies generally lack the sensitivity required for Ta₂O₅, even though such low defect levels can strongly influence electrical behaviour. Third, defect analysis methods developed for SiO₂ are not directly transferable to high-k materials due to their different bonding characteristics, dielectric response, and limited reference data. Finally, variations in target or precursor purity can introduce uncontrolled impurity levels, complicating comparisons between studies and masking the intrinsic defect behaviour of the oxide.

In this work, we investigate the nature and electrical behaviour of mobile positively charged ions, specifically Na⁺, in (i) amorphous Ta₂O₅ deposited on thermally grown SiO₂ and (ii) crystalline β-Ta₂O₅ obtained after annealing at 950 °C for 30 min [6‐8]. Our earlier studies on SiO₂/crystallized high-k oxide stacks indicated the presence of mobile Na⁺-related defects [6], but analysis in such stacks is ambiguous because Na mobility differs significantly in Ta₂O₅ and SiO₂, and interfacial trapping complicates interpretation. To avoid these issues, in the present study, we analyse crystalline Ta₂O₅ deposited directly on Si and compare it with amorphous Ta₂O₅ deposited on SiO₂, thereby ensuring well-defined interfaces and reducing uncertainties related to Na transport through underlying layers.

Using capacitance–voltage (C–V) and triangular voltage sweep (TVS) measurements [9, 10], we quantify mobile positive charges and extract their electrical activation parameters. Vapour phase decomposition ion-coupled plasma mass spectrometry (VPD ICP-MS) performed on Ta₂O₅ films deposited by magnetron sputtering was carried out to identify and compare the impurity concentrations present in these films. The objective of this study is to conclusively identify, quantify, and electrically characterize mobile positively charged ions in both amorphous and crystalline Ta₂O₅ and to demonstrate—through the combination of electrical measurements and chemical analysis—that these ions originate from Na⁺ contamination introduced during deposition. Our findings show that Na⁺ is present in both amorphous and crystalline Ta₂O₅ and is highly mobile already at temperatures as low as 150–200 °C, which raises significant reliability concerns for microelectronic applications.

For our experiments, we utilized Czochralski-grown n-type Si wafers with a resistivity of about 2 Ωcm. Before the deposition of Ta₂O₅, the wafers were dipped into a dilute (1%) HF solution for 30 s to remove the native oxide. This step ensures a well-defined Ta₂O₅/Si or SiO₂/Si interface prior to deposition and prevents additional fixed charges or trapping effects originating from residual SiO₂. On some wafers, a 75 nm SiO₂ layer was thermally grown prior to the deposition of Ta₂O₅, whereas other wafers were coated with Ta₂O₅ immediately after an HF etch. As reference samples, Si wafers with only the thermally grown 75 nm SiO₂ layer (after an HF clean) were used to verify that the HF-etching process does not introduce alkali contamination. Ta₂O₅ layers with thicknesses of 90 nm and 120 nm were deposited by RF magnetron sputtering (RFMS) using an Ardenne Anlagentechnik system, with a target–substrate distance of 10.5 cm. The sputtering gas pressure was constant at 0.957 Pa, and the RF power was set at 2600 W. The substrate was maintained close to room temperature (RT), with a constant O₂/Ar flow ratio of 1:12 and a deposition rate of approximately 1 nm/s.

To identify unintentional impurities introduced during Ta₂O₅ deposition, the purity of the target was analysed using commercially available VPD ICP-MS. For this purpose, a thin amorphous Ta₂O₅ film was sputtered onto a Si wafer by briefly opening the corresponding shutter for 1–3 s. Extending the shutter-open time produces an intense Ta signal, which suppresses the visibility of minor impurities and makes them difficult to reliably detect in the ICP-MS spectra. The resulting VPD solutions were analysed using a standard quadrupole ICP-MS equipped with pneumatic nebulization and an argon plasma source. Spectra were acquired in no-gas, He-collision, and H₂-reaction modes to suppress polyatomic interferences. Instrument parameters (plasma conditions, gas flows, dwell times) were optimized using a multi-element tuning solution. Only signals exceeding three times the baseline noise were considered detectable.

Several wafers underwent annealing using a rapid thermal annealing tool at 950 °C in O₂ and Ar atmospheres for 20 and 10 min, respectively. The purpose of the annealing process was to induce crystallization of amorphous layers into β-Ta₂O₅ [7, 8]. Subsequent to the heat treatment, the wafers were cooled at a rate of 50 K/s. The thickness of Ta₂O₅ was determined through ellipsometry measurements. The structural analysis of both as-deposited and annealed Ta₂O₅ layers was confirmed using X-ray diffraction (XRD) measurements. The XRD spectra closely resembled those previously reported in Refs. [7, 8], affirming the amorphous structure of as-deposited Ta₂O₅ and an orthorhombic structure of Ta₂O₅ after annealing the amorphous layers at 950 °C for 30 min. We note, however, that the intensity of the dominant peaks varies among samples due to differences in the stress that develops after Ta₂O₅ deposition.

Metal oxide semiconductor structures, with sizes of 0.25 mm² and 1 mm², were created by sputtering 1000 nm Al onto the top of Ta₂O₅, followed by lithography processes. Capacitance–voltage (C–V) characteristics were recorded at room temperature using an HP4284A measurement unit, both before and after electrical stressing at different voltages for 30 min at 200–250 °C. In addition, quasi-static triangular voltage sweep (TVS) measurements were performed using a K595 system to characterize mobile defects in both amorphous and crystalline Ta₂O₅ over a range of voltage sweep rates and temperatures.

Figure 1 presents representative C–V characteristics for wafers containing amorphous Ta₂O₅ on 75 nm SiO₂ (a) and crystalline (b) β-Ta₂O₅, measured before and after annealing under different bias conditions at 200 °C for 30 min. All measurements were conducted at room temperature. After annealing, the samples were cooled while maintaining the applied bias at elevated temperatures. For structures with amorphous Ta₂O₅, the C–V curves were obtained in the ± 5 V range. In contrast, measurements for crystalline β-Ta₂O₅ were limited to the + 2 V to − 2 V range, since significant conduction was already observed at voltages around 3 V. Previous reports [11‐14] have attributed the higher leakage currents in crystalline Ta₂O₅ to grain boundaries, which provide additional pathways for electron transport. The dielectric constant of Ta₂O₅ was extracted from the maximum capacitance at + 5 V for the amorphous films and at + 2 V for the crystalline films, yielding values of 25 for a- Ta₂O₅ and 41 for β- Ta₂O₅. These values are consistent with the typical ranges reported in the literature for both amorphous and crystalline Ta₂O₅ [7, 8, 14].

Full size image

It is noteworthy that the C–V curves of as-deposited structures with amorphous Ta₂O₅ are shifted towards more negative voltages compared to those measured for crystalline Ta₂O₅ (see Figs. 1a and 1b). In general, a larger negative shift in the C–V characteristics is associated with a higher density of positively charged defects in the corresponding samples. To provide a qualitative estimate of these defects, the flat-band voltage (V_FB) was determined following the procedure described in Ref. [15]. The density of positively charged defects was then calculated from the deviation of the measured V_FB from its theoretical value [15]. The resulting defect densities are approximately 1.5 × 10¹¹ cm⁻² for amorphous Ta₂O₅ and − 1 × 10¹² cm⁻² for crystalline Ta₂O₅, respectively (see Table 1). For these calculations, the effective work function difference between Al and n-type Si was taken as − 0.25 V for amorphous Ta₂O₅ and − 0.3 V for crystalline Ta₂O₅.

In general, mobile defects in MOS structures can be detected through shifts in the flat-band voltage when the structures are stressed under an applied bias at elevated temperatures. Consistent with this, we observed a flat-band voltage shift in structures containing amorphous Ta₂O₅ and β-Ta₂O₅ after annealing with an applied bias at 200 °C. However, both the magnitude and direction of the shift depend strongly on the polarity and magnitude of the applied bias. The flat-band voltage after annealing with a negative bias becomes more negative in structures with a-Ta₂O₅ and more positive in structures with β-Ta₂O₅ compared to that observed in unannealed structures. It is important to note that the values of the shifts are relatively small (below 1 V). In contrast, a significant negative shift of V_FB was observed by annealing amorphous Ta₂O₅ with + 5 V or β-Ta₂O₅ with + 2 V at 200 °C. Usually, a negative shift of V_FB in the annealed samples indicates the presence of mobile positively charged defects [15]. According to [15], the density of such mobile defects in an oxide can be calculated by comparing V_FB before and after annealing at a definite temperature:where C_ox is the capacitance of the oxide per unit area and ΔV_FB are the changes in the flat-band voltage before and after the annealing.

The densities of mobile positively charged defects determined from Eq. (1) after annealing under positive bias at 200 °C are 1.5 × 10¹¹ cm⁻² for amorphous Ta₂O₅ and 1.7 × 10¹¹ cm⁻² for crystalline Ta₂O₅, and these values are also summarized in Table 1.

Figure 2 shows G_p/ω versus voltage, where G_p is the equivalent parallel conductance measured at 100 kHz in structures with amorphous Ta₂O₅, recorded before and after annealing under applied bias (+5 V (blue curve) or − 5 V (red curve)) at 200 °C for 30 min. Before annealing, G_p/ω values are approximately constant (around 1.5 × 10^–9 Ss) at positive biases, and they are a factor of 30 lower at negative biases. Similar G_p/ω values were also observed in wafers annealed with − 5 V. In contrast, the drop in conduction values shifts towards more negative voltage (at around − 10 V) in wafers annealed with + 5 V. Notably, we did not observe any characteristic peaks of G_p/ω that could be correlated with interface states before and after annealing under applied bias, regardless of the polarity of the applied voltage at 200 °C. This shows that the density of interface states is below the detection limit, which is around 5 × 10¹⁰ cm⁻² in these wafers.

Full size image

Figure 3 shows TVS curves recorded in structures with amorphous Ta₂O₅ (a) and crystalline (b) β-Ta₂O₅ with different sweep rates α at 160 °C. By varying α from 0.001 V/s to 0.1 V/s, a significant TVS peak was observed in structures with a-Ta₂O₅ and β-Ta₂O₅ at this temperature. For α = 0.1, the peak’s maximum was around 0 V in a-Ta₂O₅, whereas it was around 0.75 V in β-Ta₂O₅. Decreasing α caused the peak maximum to shift towards a more positive bias in both amorphous and crystalline Ta₂O₅. It’s worth noting that the intensity of TVS peaks decreases significantly with decreasing α, and the changes in intensity as a function of α are significantly larger in structures with β-Ta₂O₅ than those in a-Ta₂O₅. The decrease in TVS peak intensity with decreasing α can be explained by the fact that mobile ions have more time to redistribute under the applied electric field. At slower sweep rates, ions move gradually rather than accumulating abruptly, which reduces the transient current observed during the sweep. This behaviour is a characteristic of mobile ionic species in high-k dielectrics and enables them to be distinguished from fixed charges. Consistent with Refs. [9, 10], we attribute the TVS peaks shown in Fig. 3 to mobile positively charged ions drifting towards the crystalline Ta₂O₅/Si or amorphous Ta₂O₅/SiO₂ interface under the influence of the linearly increasing gate voltage. When the ions reach the interface, the TVS current reaches a maximum. To determine the mobility of the ions observed as TVS peaks in Fig. 3, we applied a simple model successfully used for the quantitative analysis of mobile alkali ions in SiO₂ [9, 10].

Full size image

According to this model, the mobility of ions could be determined from the slope of the shift of a TVS peak as a function of α from the following equation [9, 10]:where d_ox is the thickness of the oxide and ΔV is the shift of the maximum of the TVS peak.

Then a plot of ΔV as a function of α^1/2 gives a slope which is proportional to μ^−1/2.

According to Eq. (2), which is used to determine the ion mobility, we performed TVS measurements at temperatures ranging from 130 to 190 °C, varying the voltage sweep rate α at each temperature. Figure 4 illustrates the shift in the position of the TVS peak maximum for measurements performed with different sweep rates α at various temperatures in structures with a-Ta₂O₅ (a) and β-Ta₂O₅ (b). Linear fits of the experimental points are also presented in this figure. It’s important to note that a significant deviation from the linear dependencies was observed for α > 0.1 V/s, and the corresponding points were not considered for the linear regression analysis. This deviation from the linear dependence could be attributed to the fact that mobile ions may not be able to follow the rapid changes in bias applied to the structures at the temperatures of measurements.

Full size image

From the slope of the fits in Fig. 4, we calculated the mobility of ions at different temperatures using Eq. (2). Subsequently, the activation energy of drift can be determined from this equation by plotting μ as a function of temperature:where μ₀ is the mobility at 0 K, E_d is the activation energy of drift, T is the temperature, and k is the Boltzmann constant.

We also note that no such TVS peaks were observed in the reference wafers containing only 75 nm of thermally grown SiO₂ on Si, even when subjected to the same HF-etching procedure used prior to Ta₂O₅ deposition.

Figure 5 shows the Arrhenius plots recorded for mobilities of positively charged ions in both amorphous and β-Ta₂O₅. In both cases, the experimental points could be well fitted by linear fits (see solid lines in Fig. 5). The activation energy of ion drift calculated from the linear fits gives 360 meV and 380 meV for structures with a-Ta₂O₅ and β-Ta₂O₅, respectively, whereas μ₀ was determined as around 5 × 10^–6 cm²V⁻¹ s⁻¹ in a-Ta₂O₅ and 10^–5 cm²V⁻¹ s⁻¹ in β-Ta₂O₅.

Full size image

Figure 6 shows the ICP-MS spectra of amorphous Ta₂O₅ films recorded with three different tuning modes: no-gas, He-collision, and H₂-reaction modes. The ICP-MS intensities are shown on a nonlinear (compressed/logarithmic) scale to allow visualization of both low- and high-count signals in the same spectrum. Although the instrument software labels the x-axis as “m”, this corresponds to the mass-to-charge ratio (m/z), as nearly all analyte ions in ICP-MS are singly charged. In all cases, the spectra are dominated by the strong Ta signal at mass ~ 181 amu, as expected for Ta₂O₅. When no collision or reaction gas is applied, several additional peaks appear at low masses, some of which originate from polyatomic interferences. Operating the ICP-MS in He-collision mode significantly suppresses many of these interference-related signals, leading to a cleaner background and improved separation of impurity-related peaks. The H₂-reaction mode further reduces residual interferences and improves the detectability of light and alkali elements. Across all three modes, a distinct peak corresponding to Na is consistently observed at mass 23 amu, confirming the presence of sodium as an extrinsic impurity in the deposited Ta₂O₅. Additionally, a weak but measurable Fe signal at mass 56 amu appears in He-collision and H₂-reaction modes, indicating the presence of trace iron contamination; however, its intensity is more than an order of magnitude lower than that of Na and therefore does not influence the electrical behaviour discussed in this work. The comparison of tuning modes demonstrates how the use of collision and reaction gases enhances impurity identification by minimizing polyatomic contributions.

Full size image

Mobile positively charged defects with concentrations of approximately 5 × 10¹¹–10¹² cm⁻³ were observed by TVS measurements in structures containing both amorphous Ta₂O₅ and crystalline β-Ta₂O₅. The HF-etching process itself does not introduce alkali contamination, as no TVS peaks were detected in reference MOS structures with thermally grown SiO₂ on Si after the native oxide was removed by an HF dip. This confirms that HF etching does not alter the concentration of alkali ions in Ta₂O₅.

Analysis of G_p/ω versus voltage (see Fig. 2) revealed no peaks attributable to interface states, with their density remaining below the detection limit of our measurement system (~ 5 × 10¹⁰ cm⁻² eV⁻¹) both before and after annealing under applied bias at 200 °C in structures with amorphous Ta₂O₅. Therefore, interface states can be ruled out as the cause of the flat-band voltage shift observed before and after annealing.

It is highly unlikely that the observed positively charged defects originate from native defects formed during deposition. Intrinsic oxygen-vacancy-related defects generally require much higher temperatures for thermal activation due to their deep-level nature within the Ta₂O₅ bandgap. Moreover, TVS peaks arising from the ionization of such defects should not shift when α is varied at any temperature. If the defects were intrinsic oxygen-vacancy defects, the TVS peaks would appear at fixed positions (since the energy levels are fixed) and would not change with variations in α. In contrast, the observed TVS peaks are more consistent with extrinsic impurities, which can diffuse and interact with the oxide lattice at relatively low temperatures. Given that Na is expected to be positively charged in high-k oxides, it is natural to attribute the observed positively charged defects to interstitial Na⁺ introduced during deposition, likely due to target contamination. This assignment is further corroborated by VPD ICP-MS measurements, which reveal several impurities (Na and Fe) in amorphous Ta₂O₅ layers, with the Na signal significantly higher than that of Fe. High-temperature annealing at 950 °C does not remove Na from the films: Alkali ions are highly stable in high-k oxides and remain trapped in the bulk unless a diffusion pathway to the surface or an external getter is available. In our experiments, annealing was performed in a closed stack without such sinks and, consequently, the Na incorporated during sputter deposition remains in the crystalline layer after annealing and continues to act as the dominant mobile impurity. Here, it is important to note that although the thermal diffusion coefficient of Na⁺ in bulk Ta₂O₅ is very low, the movement observed in TVS measurements at 140–190 °C does not require long-range bulk diffusion. The applied electric field during TVS can drive field-assisted migration (ionic drift) of Na⁺, enabling redistribution even at relatively low temperatures. Moreover, TVS is highly sensitive to local changes in defect charge, so small-scale rearrangements of Na⁺ near the electrodes can produce observable shifts in the peaks without the ions needing to traverse the entire film.

The activation energies of Na⁺ drift, obtained from Arrhenius plots, were 0.36 eV and 0.38 eV in structures with amorphous Ta₂O₅ and β-Ta₂O₅, respectively. The close agreement of activation energies in amorphous and crystalline films suggests that Na occupies interstitial sites without forming bonds with other atoms in the lattice. These values are also comparable to previously reported activation energies for Na⁺ in SiO₂ (~ 0.3–0.44 eV, depending on the oxide quality) [9]. Notably, the activation energy of Na⁺ in crystalline Ta₂O₅/SiO₂ stack structures has been reported to be slightly lower (~ 320 meV [6]), which may be related to differences in stress between Ta₂O₅ deposited on SiO₂ versus Si, although further investigation is needed to confirm this.

As discussed above, the presence of Na leads to significant shifts in the flat-band voltage extracted from C–V curves after annealing with a positive bias at elevated temperatures (see Fig. 1). In contrast, the smaller flat-band voltage changes observed under negative bias may be associated with the annealing or redistribution of native defects such as vacancies and interstitials, which can assume different charge states in amorphous versus crystalline Ta₂O₅ depending on the Fermi level position. However, further studies are required to clarify the origin of these defects.

In this work, we identified and quantified mobile positively charged defects in structures with amorphous and crystalline Ta₂O₅ films deposited by RF magnetron sputtering and subsequently annealed at 950 °C. Capacitance–voltage and triangular voltage sweep (TVS) measurements consistently revealed the presence of mobile ionic species whose behaviour could not be explained by intrinsic point defects. Complementary VPD ICP-MS analysis demonstrated that sodium (Na) is the dominant extrinsic impurity in Ta₂O₅ and that its concentration exceeds that of any other detectable element by more than an order of magnitude. Based on these results, we attribute the observed mobile positive charge to interstitial Na⁺ ions incorporated during deposition.

By analysing TVS spectra recorded at different voltage sweep rates and temperatures, we extracted the drift activation energies and pre-exponential mobilities for Na⁺ transport in both structural phases. The activation energy was found to be ~ 360 meV in amorphous Ta₂O₅ and ~ 380 meV in crystalline β-Ta₂O₅, with corresponding µ₀ values of 5 × 10^–6 cm² V⁻¹ s⁻¹ and 1 × 10^–5 cm² V⁻¹ s⁻¹, respectively. These values confirm that Na⁺ ions are highly mobile at temperatures as low as 150–200 °C, where they can significantly influence the electrical response of Ta₂O₅ layers.

Overall, our findings demonstrate that Na⁺ contamination is the primary source of mobile positive charge in both amorphous and crystalline Ta₂O₅, highlighting the critical importance of impurity control and, in particular, the purity of sputtering targets for ensuring the reliability of Ta₂O₅-based microelectronic devices.