Improvement of Ferroelectricity with Two-Dimensional Enwrapping Structure

Abstract

The effect of the layered InSe intercalation structure on the ferroelectric properties of HfO₂ was investigated. At low crystallization temperatures, the ferroelectric phase is formed more easily. Stronger polarization and better reliability can be achieved. This result indicates that the InSe intercalation structure is promising for engineering the ferroelectric properties of HfO₂.

1. Introduction

Recently, HfO₂ has received attention as a novel ferroelectric oxide due to its smaller permittivity and ultrathin nanometer-scale thickness in comparison to conventional perovskite materials. It is highly interesting that the ferroelectricity of HfO₂ can exist in an ultrathin state, which holds great promise for scaled integrated circuits. In the meantime, HfO₂ is well known for its many important microelectronics applications, such as high k dielectric, non-volatile memory, etc., and due to its good compatibility with COMS industrial processes and highly developed ultrathin conformal fabrication technology (such as atomic layer deposition). Consequently, this novel type of ferroelectric material discovered in HfO₂ will significantly expand the application area of the polarization field. Ferroelectric formation is typically attributable to non-centrosymmetric crystal structures [1]. According to previous research, it is the orthorhombic phase that contributes to the ferroelectric properties of ultrathin HfO₂ films [2]. Moreover, defects (such as oxygen vacancies) have been reported as typical causes for the production of the orthorhombic phase in HfO₂ ferroelectric films [3,4]. Though HfO₂ ferroelectric films should possess favorable chemical stability and robust polarization under repeated electrical stress for application purposes, a HfO₂ oxide layer with relative oxygen vacancies is usually susceptible to oxidation, and random oxidation limits the advantages of HfO₂ as a ferroelectric material and inhibits the ferroelectric performance, making it neither stable nor robust.

A HfO₂ oxide layer with a relatively large amount of oxygen vacancies is usually in an oxygen-deficient state. Though ferroelectric behavior may be more easily developed in the HfO₂ films with defects such as oxygen vacancies than completely stoichiometric ones, the control of defects such as oxygen vacancies is difficult due to the inter-diffusion or outer oxidation issues when HfO₂ comes into contact with other layers. There are a number of methods to control oxygen vacancies, including the use of oxygen-scavenging electrodes such as TaN, nitrogen annealing, interface-induced film stress, electrical cycling, doping with cations of different valences, and modification by cations with strong oxygen affinity, such as Zr. However, these methods either degrade the electrical stability or alter the internal chemical structure for doping or electrical stressing. Therefore, instead of methods of doping or stressing, an ultrathin layer of InSe was used in this study to enwrap the core layer of HfO₂.

It is interesting that stacked InSe is technologically desired for improving the ferroelectric properties of HfO₂. The sandwich-stack structure of InSe-HfO₂(oxygen-deficient)-InSe leads to an improvement in remnant polarization (P_r) when compared to the control samples without InSe enwrapping. In addition to maintaining the ferroelectric phase, it also improves the ratio of the orthorhombic phase in the oxygen-deficient HfO₂ (OD-HfO₂) layer in the sandwich-stack structure. Since previous research has shown that particular encapsulation confinement encourages the formation of a ferroelectric HfO₂ layer, the sandwich-stack structure might be a promising engineering method to modify the OD-HfO₂ to achieve improved ferroelectricity. The role of the enwrapping layer is not only to protect, but also to sufficiently confine the enwrapped OD-HfO₂ layer [5,6,7], and this suggests potential for a variety of applications, including as transistors, NVM devices, and other related applications [8,9,10].

2. Experimental

In this work, a sandwich InSe/OD-HfO₂/InSe stack was fabricated with HfO₂ covered by InSe layers both at the top and bottom sides, denoted as SS (Se-covered sample) samples. First, the bottom electrode, composed of 150 nm Ti, was patterned on Si substrates with 200-nm-thick SiO₂ layers prior to the fabrication of the device. InSe was initially purchased from the SUNANO Co., Ltd. and was synthesized by chemical vapor transportation (CVT) from mixtures of elemental InSe precursor powder within a quartz tube. An exfoliation-transfer process was then undergone to place a InSe nanoflake on the Ti surface. The OD HfO₂ thin film was subsequently fabricated using a magnetron sputtering system with a high argon/oxygen ratio (9:1), targeting pure metal Hf (99.99%). The corresponding operating pressure and power were 1.13 × 10⁻³ Torr and 100 W. X-ray analysis confirmed an Hf:O ratio of 1:1.25 in the HfO₂ layer, indicating an internal oxygen deficiency in the HfO₂. Afterwards, the other 150 nm Ti top electrode was deposited on top of the HfO₂/InSe heterostructure. Ti (150 nm) top electrodes (TEs) with a square pattern of 150 × 150 μm² size were obtained by photolithography and dry etching [11]. Finally, all samples were subjected to rapid thermal annealing (RTA, or RTP) for 40 s at 400–700 °C in a N₂ environment. In addition, control samples (denoted as CS below) without InSe enwrapping were also prepared with the same procedure for comparison.

To analyze the crystal structure, we performed grazing incidence X-ray diffraction (GIXRD) measurements. The chemical structure was also analyzed by X-ray photoelectron spectroscopy. XPS spectra were collected at the interface with Ar⁺ (2 keV) etching to obtain comparative information on the valence states at a certain depth. The polarization–voltage (P–V) characteristics were measured using an FE tester (TF Analyzer 2000, Aixacct Systems). In this experiment, ferroelectric polarization measurements were performed using the positive-up and negative-down (PUND) method. The 1 kHz frequency with triangular bipolar pulses was used in the PUND measurements. A fatigue pulse with an amplitude of 3.0 V and a frequency of 100 kHz was used to test its durability. To obtain the switching I–V curve, electrical tests were performed using a Keithley 4200 in voltage sweep mode. Electrical characteristics were also evaluated using a Keithley 4200 parameter analyzer and a PMU ultrafast current–voltage module.

3. Results and Discussion

Figure 1 shows the typical polarization vs. bias curve of SS samples compared with control samples (CS) under different RTP temperatures. Significantly improved ferroelectric properties could be observed from the P–V curves. It is noted that the samples underwent wake-up cycling, since the batches of samples were measured many times before the fully developed curves were obtained for the analysis. Meanwhile, after the ferroelectric phase was formed, there was no weakening or disappearance after annealing in oxygen for 2 h. In addition, for the CS samples, a slight distortion of the hysteric P–V curve was present at a low temperature of 450 °C, implying a non-negligible leakage, which is consistent with an earlier report [1].

We independently measured the InSe samples supplied by the national manufacturer, which exhibited no ferroelectric polarization response curve at the fundamental level. If this is the case in isolation, the ferroelectric characteristics could have been obtained only from the hafnium oxide rather than InSe, since it is sandwiched with ferroelectric HfO₂. However, we note that the ferroelectric characteristics of In₂Se₃ have been reported recently, which may be attributable to various In-Se ratios.

As indicated in Figure 2a, the polarization of the SS samples exhibits less change between 450 and 700 °C. Meanwhile, it is lower due to the constraints of the heat budget on the development of HfO₂ ferroelectricity in the SS sample than in the CS sample. Figure 2a displays the extracted data rather than the measured data. The average of p_r (positive) and p_r (negative) was used as the basis for the extraction in order to facilitate visualization and analysis. Additionally, the overall results were averaged to obtain statistical patterns for trend analysis in light of the fact that we performed analyses on several batches of samples, as well as various wake cycles on the samples. Since the samples were obtained from different batches, both fresh and after undergoing wake-up cycling, when using the statistical average, low temperatures, such as 450 °C, show a higher fluctuation error. The greater temperature of SS allows for a higher ferroelectric phase share, which is indicative of its benefit, as opposed to a low temperature.

Both samples were subjected to the wake-up/fatigue measurement at the lowest temperature of 400 °C, as depicted in Figure 2b. The fatigue cycling performed here used bipolar cycling, and the frequency was higher than the effective frequency of the hysteresis measurement. The endurance was investigated with triangular pulses at a frequency of 100 kHZ. The variations in 2P_r changed with the increase in fatigue pulses. The CS exhibited poor cycling endurance characteristics, with around 10⁶ endurance times, compared to approximately 10⁹ endurance times for the SS. The lower durability of the CS sample is attributable to the uncompensated ferroelectric FE-bound charge, which causes additional field effects on the FE layer, suggesting the presence of non-FE components [10], whereas the P_r of the SS sample increased steadily and did not show a fast decrease up to 10⁹ cycles, which can be ascribed to the good control of oxygen vacancies in OD-HfO₂. Generally, due to their high mobility and internal voltage, oxygen ions can travel without an external voltage [2,11]. This movement can be accelerated by the concentration gradient of oxygen vacancies from higher-density sites to lower-density ones, including into the adjacent layers, relying on the powerful transportation ability of oxygen species [12], which would randomly break the ferroelectric nucleus sites around oxygen vacancies in the HfO₂ layer if there was no protection by InSe.

It could not be guaranteed that the samples were able to entirely follow the high frequency, as consistent with that at the effective frequency. In fact, the ferroelectric values tested at high frequency were lower than those at low frequency. This may be related to the fact that the sample cannot follow the high-frequency field changes. However, one can observe a clear difference in the fatigue curves between SS and CS samples. SS shows much stabler endurance performance, even at a high frequency, which is strong evidence that the ferroelectricity in SS is much more stable. At a low frequency, both samples show better endurance characteristics with the same result that SS is superior to CS at the cost of a higher cycling number. The endurance cycle tests were performed for various frequencies, and we chose the high-frequency sample at a low temperature, as shown in Figure 2b (aiming to demonstrate that SS can still show better performance in a disadvantageous environment than at high temperatures, considering the previously observed stability and the enhanced ferroelectric phase due to high temperatures).

It is the o-phase among the following three phases, e.g., m-phase (P2₁/c monoclinic), o-phase (FE Pca2₁ orthorhombic), and t-phase (P4₂/nmc tetragonal), that leads to the ferroelectric properties of HfO₂. As shown in Figure 3, the ferroelectric o-phase is present in both SS and CS samples. The clear ferroelectric o-phase peak appears at 31°. Here, the values of the o-phase fraction are 84.3% and 73.7% for SS and CS samples, respectively. Previous reports [1,13] proposed a model for ferroelectric phase transition during cooling after rapid thermal climbing [14,15,16]. The t-phase formed during RTP is generally considered as the precursor phase, which can transform into either the m-phase or the o-phase [17]. However, the driving force of the former transformation is higher than that of the latter, and, therefore, the probability of the formation of the latter ferroelectric phase is occupied by the former. It is difficult to guarantee the formation of the pure t-phase because the surface energy of the t-phase is smaller than that of the m-phase, which makes it difficult to be stabilized at low RTP temperatures. However, despite the relatively high bulk free energy of the t-phase, it has been reported that oxygen vacancies can reduce the bulk free energy of the t-phase more effectively than the m-phase [18]. Therefore, the InSe layer in SS can inhibit the formation of the m-phase in HfO₂. The larger amount of oxygen vacancies in SS samples results from the efficient immobilization and protection by the sandwich structure, which in turn leads to an increase in o-phase formation. It is generally believed that phase transitions tend to form a monoclinic m-phase, thus occupying the probability of other phase transitions (e.g., t → o-phase) [2,19]. Thus, by proper stabilization of oxygen vacancies and perfect confinement during thermal processes, the amorphous state without crystallinity promotes the o-phase transition by suppressing the m-phase transition.

Figure 4 shows the calibrated XPS spectra of the SS sample and the CS sample. From the XPS spectra, the binding energies of Hf 4f in both SS and CS samples show a slight shift to higher energies in the deposited state, indicating the presence of anoxic states [20,21]. The chemical states of the Hf-O bonds in the SS samples at lower binding energies further suggest anoxic states. Moreover, these states are significantly reduced in the CS samples, which suggests that the Hf-O bonds are easily oxidized without sufficient constraints. In addition, the CS samples show an increase in Hf-O bonding even after low-temperature RTA (450 °C) as the oxygen-deficient bonds are converted to Hf-O bonds by thermal energy, in contrast to the change (smaller but stable) in SS. The non-centrosymmetric phase transition in the SS samples is accelerated with the help of the low temperature as the oxygen-deficient bonds act as nucleation centers when provided with thermal energy during annealing.

To identify the oxygen transportation into/out of the Ti electrode and the redox interaction between HfO₂ and Ti, an XPS study was carried out on the SS and CS structures. All of the spectra were classified as metallic titanium (Ti) and titanium dioxide (TiO₂). The corresponding subpeak positions were located at 562 and 567 eV. Figure 5 illustrates the XPS spectra for two types of samples with the results of Gaussian fitting after various RTA processes.

Prominent chemical changes in the ratio of Ti:TiO₂ can be observed in the Ti spectra of CS samples, which imply fast oxidation at the expense of reduced oxygen vacancies. From the spectra, Ti is readily oxidized at low temperatures. This proves that the oxidation of CS samples is rapid and difficult to control. This is responsible for the severe disordered oxidation of the electrode due to the disordered/random migration of oxygen under voltage stimulation with direct contact between Ti and HfO₂ [13,22]. However, the Ti 2s spectra of the SS device showed a relatively stable ratio of Ti:TiO₂ in the interface region, indicating a good interface with less oxygen migration into the Ti layer.

4. Conclusions

In conclusion, the changes in the ferroelectricity of HfO₂ due to InSe introduction were experimentally analyzed. The SS samples underwent a complete o-phase transition at low temperatures and effectively increased the relative proportion of the ferroelectric o-phase. SS is an optimized engineering approach to improve the ferroelectricity of HfO₂.