Nano-positive up negative down in binary oxide ferroelectrics Open Access

Doped hafnia (HfO₂) is considered a frontrunner material for emerging devices such as ferroelectric (FE) memory and negative-capacitance field-effect transistors.^1–3 A variety of physical mechanisms, including surface and size effects, doping, and strain, have been introduced to explain the nature of the polar non-centrosymmetric FE phase in HfO₂. Therefore, the ferroelectricity is generally attributed to the presence of the orthorhombic phase (space group Pbc21) in polycrystalline HfO₂.^4,5 A common experimental method for obtaining FE-HfO₂ is to deposit hafnia with dopant atoms and then anneal the film. This process results in polycrystalline thin films of FE-HfO₂, in which the tetragonal, cubic, and orthorhombic phases are more stable than the monoclinic phase. When in the form of thin films, the crystalline phases, grain size, and distribution play a major role in the overall device performance. This creates a challenge for the physical and electrical characterization of FE-HfO₂ devices. Despite the widespread use of doped-HfO₂, its integration in the form of ferroelectric capacitors results in the acquisition of electrode-averaged information. This means that many questions about the origins of ferroelectricity in HfO₂ remain unanswered. In addition, direct access to localized FE response by local observation at the nanoscale is also challenging, even for state-of-the-art materials characterization methods.⁶ Scanning probe microscopy (SPM) techniques are ideally suited for the local probing of piezo- and ferro-effects, and they have been widely explored in recent years.

However, the most commonly available methods, including piezoresponse force microscopy (PFM) and its various resonant implementations, are known to be limited by electrostatic artifacts that limit their application in ultra-thin dielectric layers.^7,8 On the contrary, at the macroscopic scale, a straightforward method for the analysis of ferroelectric switching is represented by the positive up negative down (PUND) method. This is based on the subtraction of current contributions from leakage and polarization reversal when a specific waveform is applied to the FE capacitor, resulting in a quantitative measurement of the polarization switching current. Recent works have shown this method to be applicable at the nanoscale using SPM hardware, i.e., nano-PUND.⁹ In this paper, we explore the possibility to use nano-PUND to study nanoscale ferroelectric switching on the bare surface of FE-HfO₂. After describing the setup implementation, we demonstrate for the first time the feasibility of this method in Si-doped HfO₂. Our results indicate the presence of polar and non-polar phases detectable with our approach, yielding different results in the nano-PUND response. Interestingly, we do not observe a clear correlation between the dielectric strength and ferroelectric response, as shown by the weak correlation between tip-sensed electronic leakage and nano-PUND response. Finally, we attempt to extract the polarization charge based on a direct estimate of the tip–sample contact area. In addition, we report on the unusual electrochemical activity on the film surface, which may indicate multiple sources of artifacts in the interpretation of conventional PFM reports.

Comprehensive reports on the oxide phases that are obtained by substitutional cation doping with Si, Al, Zr, La, and Y can be found elsewhere.¹⁰ Here, we focus the study on Si-doped HfO₂. A blanket layer is fabricated on a p⁺-Si substrate (10¹⁹ at/cm³), and the 8 nm-oxide is grown by atomic layer deposition (ALD) at 300 °C. Silicon is used as a dopant during the oxide growth; we select a concentration in the range of 3.6 mol. %. The ALD film is formed by depositing HfO₂ and Si sub-layers in an alternate manner. After the oxide deposition, a 50 nm thick Si capping layer is deposited by physical vapor deposition, and the stack is annealed at 1000 °C for 30 s. This thermal treatment induces oxide crystallization, leading to a polycrystalline film containing a mixture of crystallites with different phases such as monoclinic, tetragonal, cubic, and orthorhombic (m-, t-, c-, and o-phase) that are probed by means of x-ray diffraction. Our sample preparation method, involving the alternate deposition of HfO₂ and Si sub-layers followed by a high temperature annealing, is designed to induce crystallization and stabilize the ferroelectric orthorhombic phase. Afterward, the Si capping layer is completely removed by reactive ion etching. The stoichiometry is confirmed by Rutherford backscattering (RBS) with elastic recoil detection (ERD), which is not shown. Scanning electron microscopy and cross-sectional high-resolution transmission electron microscope (TEM) of the sample are shown in Fig. 1(a), confirming the layer thickness, the sharp interfaces, and the polycrystalline nature of the film. For comparison, standalone capacitors (100 × 100 µm²) are fabricated with the same flow. Polarization–voltage (P–V) hysteresis and current–voltage (I–V) loops are acquired on capacitors at a measuring frequency of 1 kHz [Fig. 1(b)]. During the electrical measurements, the top-electrode is biased, and the bottom-electrode is grounded. Note that the exposed surface of FE-HfO₂ in the vicinity of the electrodes is accessible to the probe of atomic force microscopy (AFM) and is used in this study [Fig. 1(c)]. Figure 1(d) shows the leakage current sensed on the surface of the Si-doped HfO₂ using conductive atomic force microscopy (C-AFM). In this study, while C-AFM primarily measures the local resistance, reflecting the flow of current through the probed region, it also indirectly informs us about the local permittivity. Permittivity influences the capacitive nature of the current observed; regions with higher permittivity can concentrate electric fields more effectively, thereby affecting the current measured by the C-AFM tip. This capacitive effect is subtly present in the C-AFM current data, particularly noticeable at the initial application of voltage, and is an indirect indicator of the material’s ability to store electric charge in response to an applied field. Here, for the sake of an improved signal-to-noise ratio, we use a custom-developed transimpedance amplifier (TIA) with a noise level estimated from the current spectral density as 1.28 $fAHz$ and capable of probing the small charge exchange that occurs between tip and sample. We performed our measurements at room temperature in a dry nitrogen environment, applying a 7 V bias to the sample while the probe was grounded [Fig. 2(a)]. First, we compare the leakage current of our layer using C-AFM. Collecting a dense array of I–V characteristics in point-spectroscopy, we obtain large variations in the leakage current, reflecting in the current maps [Fig. 1(d)] as regions with local fluctuations in surface conductivity. In Fig. 1(c), the current profile is overimposed on the surface morphology, proving the absence of a clear correlation between the two. Grain boundaries can significantly affect the ferroelectric properties of polycrystalline materials like Si:HfO₂. In our nano-PUND experiments, the tip-induced nanocapacitor area is on the order of the grain size, which means that the measured response often includes contributions from both grain interiors and grain boundaries. Grain boundaries can act as preferential sites for defect accumulation, which may pin domain walls and inhibit polarization switching, or conversely, they can facilitate switching due to local stress and electric field concentration. In our case, the data suggest that the grain boundaries do not impact the leakage current, as compared with individual grains that present a higher leakage current.

In order to investigate the tip-induced polarization charge, we applied the nano-PUND method. The goal is the detection of the ferroelectric switching current (I_FE) isolated from the other current components involved in the switching event, namely the displacement current (I_D) and the leakage current (I_L). This is obtained by sensing the total current flowing in the tip–sample system (I_TOT = I_FE + I_D + I_L) while a specific waveform is applied [Fig. 2(a)]. Multiple triangular unipolar waveforms are used, starting with a pre-polarization pulse that is followed by two consecutive pulses of the same polarity (i.e., P and U). The underlined idea is that the first voltage pulse will generate a current constituted of all the active transport mechanisms, while the second pulse will contain only non-switching contributions. In this way, the contribution of I_L and I_D remains the same on subsequent P-U and N-D pulses, while the exact value of I_FE can be extracted by a simple subtraction of the first and second pulses, i.e., I_P − I_U and I_N − I_D. A detailed description of the PUND principles can be found elsewhere.⁹ Considering the impact of the electrode scaling on the sum of switching and leakage current, reducing the tip–sample capacitance is key to recover a useful signal.^9,11–13 In fact, while the I_FE and I_L scale with the electrode size (i.e., tip–sample junction area), the stray capacitance and, therefore, the I_D are determined by the geometrical probe’s parameters (i.e., apex, lever, and tip-body), inducing I_D values that can be up to two to three orders of magnitude higher compared to the I_FE, as previously measured and simulated.^12,13 Therefore, the selection of a tip that reduces the impact of the stray capacitance is required, and it must be combined with a procedure for capacitance compensation. In this work, we select full Pt probes (25PtIr300B, Rocky Mountain Nanotechnology) as their tip radius (i.e., ∼20 nm) provides good mechanical contact during the electrical stress, and they offer a relatively high cantilever height (d) as in the inset of Fig. 2(a), thus minimizing the overall capacitance coupling. For the capacitance compensation, we rely on the calibration of the parasitic capacitance proposed by Martin and co-workers, where a thick insulating sample (i.e., mica) is used as a reference to quantify the displacement current associated with the stray capacitance of the tip–sample system.⁹ The remaining current contributions can be ascribed to I_L and I_FE, respectively, and are shown in Fig. 2(b) (as absolute values). As visible experimentally [Fig. 2(b)], higher current values are repeatedly obtained for the P and N pulses compared to the U and D, as explained by the absence of the I_FE contribution in the U and D pulses. Notably, the use of an asymmetric stack where doped silicon is used as a bottom electrode results in non-linear I–V characteristics with reduced current when the Si is negatively biased (i.e., P and U cycles); this is consistent with previous observations.¹¹ 

Using a dense array of spectroscopic I-Vs, thousands of nano-PUND measurements are acquired on the surface, showing similar results in multiple locations [i.e., four datasets are shown in Fig. 2(b)]. However, it is worth noting that most of the tested locations show a monotonic increase of the I_TOT with cycling, indicating the absence of polarization current and the occurrence of conventional dielectric degradation processes.¹⁴ This indicates a large presence of grains that show a paraelectric response and relatively low permittivity among the oxide surfaces probed. In addition, the C-AFM is used to map the distribution of leakage current on the surface [Fig. 3(c)] and selectively investigate the response of nano-PUND in different locations to correlate the response with grain resistance. However, although different permittivity is clearly reflected in leakage fluctuations of the C-AFM maps [Fig. 1(d)], we do not find a correlation between regions with local high resistance and ferroelectric switching. This would indicate that the dielectric strength of FE-HfO₂ grains cannot be used alone to identify which grains will show FE-response. It is worth mentioning that in our experiments, we do not apply any wake-up conditioning to the material, and any top-electrodes, capping, and interfacial layers are also absent. Therefore, our nano-PUND results should be considered representative only for the response of the pristine material, i.e., the individual HfO₂ grains under the probe. In our nano-PUND experiments, the tip-induced nanocapacitor area is on the order of the grain size, which means that the measured response often includes contributions from both grain interiors and grain boundaries. We found that it is really difficult to isolate a single grain boundary without a substantial contribution from the grain itself. The presence of grain boundaries is known to have a substantial impact on the ferroelectric properties of polycrystalline films. In our nano-PUND measurements, the tip-induced nanocapacitor area often encompasses multiple grains and their boundaries. The orthorhombic phase’s predominance within individual grains facilitates a robust ferroelectric response.

It is also pertinent to consider that a portion of the total current (I_TOT) observed during cycling may indeed be attributable to the wake-up effect. This phenomenon, characterized by an increase in the ferroelectric switching current (I_FE) over repeated cycles, is a well-documented behavior in ferroelectric materials. The wake-up effect arises from changes in domain wall mobility or a decrease in pinning, which enhances the material’s polarization response with each successive cycle. Our experimental setup and data analysis, as currently constituted, do not allow us to definitively distinguish between leakage current increases due to dielectric degradation and increases in switching current due to the wake-up effect. Recognizing this limitation, we surmise that some portion of the monotonic increase in I_TOT could be associated with evolving ferroelectric behavior, potentially masking or occurring alongside dielectric degradation. This aspect underscores the complexity of interpreting I_TOT changes and the need for a more refined analytical approach to differentiate between these concurrent processes in future work.

Finally, using the switching current obtained by nano-PUND, it is possible to convert it into polarization charge by integrating over time the I_P-U if the value of the switching area is known (Fig. 3). Considering the nominal value of the tip radius and load-force applied during the I-Vs, a precise estimate of the effective contact area can be attempted but remains unprecise for our purpose. However, we note that after repeated I-Vs, a small surface protrusion is observed on the oxide as the result of a rich electrochemical tip–sample interaction [Fig. 3(b)]. The nature of these surface features and their dependence on environmental conditions have been treated in numerous reports.^15,16 Here, following the simulations of Ievlev and co-workers,¹⁷ we extract the footprint of the tip-induced surface modification to quantify the FE-HfO₂ surface area exposed to an electric field during nano-PUND cycling. Applying a water-threshold method to the morphology map, we obtain a value of 0.785 µm² for the tip-induced nanocapacitor area. The relatively large number should not surprise: (a) for the high pressure applied between tip and sample and the blunted tip-apex and (b) for the long contact time during prolonged cycling, thus resulting in possible drift. Considering both the strong field confinement under the tip and the size of the FE-HfO₂ grains, the value extracted here can be representative of the response of a few individual grains under the probe. Under these assumptions, we obtain an estimation of the polarization charge for Si:HfO₂ of Q_r = 17 ± 5 µC cm⁻². Note that due to the asymmetry of the electrodes in our stack (i.e., Pt tip and p-type Si), the calculations are performed on the N–D cycles, yielding a higher current. Notably, when compared to non-ferroelectric materials (<0.5 fC) and perovskites piezoelectric (>3 fC),⁹ the values obtained for Si:HfO₂ are within an intermediate range. Nonetheless, we observe a reasonable agreement with results from averaged P–E measurements, generally in the range of 10–30 µC cm⁻².

In summary, we discussed nano-PUND for the analysis of ferroelectricity in doped binary oxides. Tip-induced FE-switching is demonstrated with this method for the first time on the bare surface of polycrystalline silicon-doped HfO₂. Due to the distinctive PUND response, our results are a solid demonstration of the local ferroelectric response obtained from individual grains. However, the large uniformity of the observed response suggests a high content of paraelectric grains and highlights the main role of strain and interfacial layers in the stabilization of FE-phases. Being based on actual current probing, this method is immune to most electrostatics artifacts that plague other SPM-based techniques. Therefore, considering the widespread application of FE-HfO₂, the development of dedicated characterization methods can have a fundamental role in tuning material development and device response.

Regarding the experimental setup, we utilized the Keysight 5500 Scanning Probe Microscope (SPM) for our measurements. This model was selected for its advanced features, including its high-resolution imaging capabilities and sensitive measurement systems, which are crucial for accurately capturing the nanoscale electrical properties of our samples. Additionally, we employed a Transimpedance Amplifier (TIA) from Analog Devices, specifically the ADA4530-1 model, known for its low noise and high precision, to amplify the current signal detected by the SPM. This combination of the Keysight 5500 SPM and the ADA4530-1 TIA amplifier was instrumental in providing the high-quality data required for our study. For capacitance compensation in our measurements, we utilized a Transimpedance Amplifier (TIA) with an RC (Resistor-Capacitor) feedback loop. This configuration stabilizes the feedback loop of the amplifier within the required data bandwidth. The RC feedback loop is specifically designed to effectively manage the capacitive currents resulting from the tip–sample interaction. This approach is pivotal in our experimental setup as it allows for accurate differentiation between the capacitive background and the actual ferroelectric response. By employing this method, we ensure that our measurements are reflective of the intrinsic ferroelectric properties of the material, minimizing the influence of extraneous capacitive effects.

U.C. acknowledges Mihaela Ioana Popovici and Sean McMitchell (imec) for synthesizing the Si doped HfO₂ and acknowledges the partial funding provided by IMEC’s Industrial Affiliation programs.

The authors have no conflicts to disclose.

U.C. characterized the Si-doped HfO₂ material. A.G. and U.C. carried out the acquisition of nano-PUND datasets. A.G. and U.C. participated in the discussion and interpretation of the results. All authors contributed to the writing of the manuscript.

Andres Gomez: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – original draft (equal). Umberto Celano: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – original draft (lead); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.