Structural defects and local chemistry across ferroelectric–electrode interfaces in epitaxial heterostructures

Bright field TEM (BF TEM) was employed to investigate the interfacial quality and any structural features present. Figure 1 shows a set of BF TEM images for two PZT 20/80 samples with similar thickness of the ferroelectric layer albeit grown using different techniques. Figure 1a shows a 65-nm thick PZT film grown by HPOAS, whereas Fig. 1b shows a 55-nm thick layer grown by PLD. The 65-nm HPOAS sample displays 90° domains, threading dislocations, and a very rough SRO–PZT interface. In contrast the 55-nm PLD sample shows no domains, sharper interfaces, and a very low density of threading dislocations. This clearly denotes how significantly the growth conditions alone can influence the sharpness and structural defects in the heterostructures. In order to investigate the chemistry at the interface, EFTEM analysis was performed on both samples. Figure 2 shows the EFTEM data for both PZT 20/80 samples (a, c, e, and g represent data for the sputtered sample, whereas b, d, f, and h are for the PLD sample). Figure 2a, b displays the elastic images, whereas the elemental jump-ratio images are shown in Fig. 2c–h. The Sr jump-ratio maps indicate distinct Sr diffusion of ~5 nm into the PZT from the SRO for both samples (Fig. 2c, d). This is a rather unexpected result as it is conventionally thought that Ru, rather than Sr, is more susceptible to transport across the interfaces, as RuO₂ is highly volatile [25, 26]. However, earlier Cross et al. [27, 28] have published SIMS and EDS data for La-doped PZT (PLZT) on SRO that also finds Sr diffusion from SRO into the PLZT layer. They proposed that Sr diffusion is related to the excess of Pb, with such a diffusion being responsible for the high leakage current in their capacitors. It is well known that Sr doping of PZT leads to a marked decrease in the magnitude of the polarization and could also degrade the physical properties such as the dielectric constant and Curie temperature [29, 30]. This finding is critical as it provides a secondary explanation to the observed decrease in the switching polarization with decreasing thickness for Pb-based ferroelectric films [31]. Until now this has been attributed in an overwhelming majority of studies to the depolarization field at the ferroelectric–electrode interface, i.e., it is a physical effect rather than chemical roughness. The data presented in Fig. 2 demonstrate the critical need for careful examination of the chemistry of the interface and the possibility of strong “extrinsic” contributions to the observed degradation. Indeed such extrinsic factors have been thoroughly investigated for ferroelectric systems, which have a large lattice-mismatch with the underlying electrode [4, 5, 32]. The fact that Sr diffusion may occur even in what is perceived to be a perfect interface (e.g., Fig. 2b) is rather surprising.

EDS was performed to investigate Pb inter-diffusion and corroborate the results obtained by EFTEM. Figure 3 displays the EDS elemental line scans acquired across the interfaces for both samples, the dotted lines indicate the interfaces for each layer. Again the left column represents data for the sputtered sample and the right column is for the PLD sample. The EDS line scans for Sr confirm the results previously seen in the EFTEM maps of Fig. 2. They show diffusion of Sr to a distance of approximately 10 nm from the SRO into the PZT layer. Surprisingly, despite the radical (visual) difference in interface roughness between the two samples, Sr diffusion seems to be independent of interfacial sharpness or defect concentration. On the other hand, for Ti and Ru (the perovskite B-site elements) no diffusion could be measured by either technique. Neither the EFTEM results (Fig. 2e–h) nor the EDS line (Fig. 3a, b) scans for Ti or Ru indicates any movement or diffusion of the B-site species from the ferroelectric into the electrode or in the opposite direction.

To better understand the role of interfacial features in the ultimate local chemistry, we also investigated a 200-nm thick PZT 20/80 layer that was grown using PLD albeit with an oxygen-deficient target [32]. This thickness is well beyond the M-B critical thickness and was chosen so as to ensure the presence of 90° domains [18]. Figure 4a is the bright field image for this sample, which shows sharp interfaces and the expected 90° domains. Following the same methodology as above, EDS (Fig. 4b) and EFTEM (Fig. 5) analyses were performed. The EDS line scans in Fig. 4b again indicate Sr diffusion (~5 nm), while chemically sharp interfaces are observed for Ru, Ti and Pb. EFTEM maps confirm Sr diffusion into the PZT (Fig. 5b) and again, chemically sharp interfaces for B-site Ti and Ru (Fig. 5c, d). More interestingly, a closer inspection of the Sr jump-ratio image in Fig. 5b not only reveals Sr diffusion, but also Sr deficiency in a region directly below the 90° domain nucleation zone, an observation hitherto unreported for PZT thin films. This deficiency is indicated by a darker contrast inside the SRO layer and highlighted by the black arrow, whereas the dotted line indicates the SRO–STO interface. Notably, although the 65-nm thick HPOAS PZT 20/80 sample showed 90° domains (Fig. 1a), it did not exhibit Sr-deficient pockets below the 90° domain nucleation zone. This latter result is unique to the thicker PZT 20/80 sample grown using the oxygen-deficient target and underlies the subtle effects of the thin-film processing conditions on the local chemistry at the ferroelectric–electrode interface. These Sr-deficient pockets are currently being examined through aberration-corrected high-resolution TEM and further investigations on the plausible mechanisms behind their origins will be reported in the near future.

It is worthwhile to note that Sr diffusion from the SRO layer into the PZT 20/80 layer is observed irrespective of the growth technique or growth rates. The next step was to quantify the extent of diffusion at the PZT/SRO interface in order to identify the driving force behind the diffusion. In epitaxial thin films, the diffusion process should be considered in terms of the main parameters that describe the mass transport phenomena, namely the diffusivity of the species and the driving force (gradient of total free energy) [29]. To identify the underpinning driving force for the cation diffusion, we examine the diffusion scenario for both Pb and Sr. As a simple approximation using parabolic kinetics typical of diffusion, the diffusion depth (X) was theoretically calculated using the formula, X = √Dt, with the diffusivity (D) calculated by the Arrhenius relationship, D = D_o exp (−Q/RT), where D_o is the diffusivity coefficient, Q the activation enthalpy, R the gas constant, and T the temperature, and t is time. We first consider the self-diffusion of Pb at 600 °C in a PLD sample, similar to that shown in Fig. 4a. Considering a total deposition time of 1 h, for Pb and using the Pb self-diffusion coefficient reported by Freer [33] for PbTiO₃, we find a total length X = 60 nm. This is more than a quarter of the thickness of the film and immediately underlines the need for excess Pb in the precursor during the growth process. More importantly under the typical vacuum and temperature conditions used for the thin-film growth process, Pb volatility is expected to worsen. The lack of dependency of Sr transport on the structural roughness of the interface and relatively tight control of the B-site ions across the interface strongly suggests that the observed Sr diffusion is driven by Pb volatility. Considering the lattice structure of PZT versus SRO, it is likely that Sr (already occupying a 2+ oxidation state A-site) would be the most ideal element to occupy the vacant Pb sites. This would hence create a diffusion process from the electrode into the ferroelectric and thus reveals the importance of chemical stability of individual elements of a heterostructure. We are able to examine this by estimating the vacancy-diffusion kinetics for Sr. The diffusivity coefficient of Sr has been reported previously for perovskite structures by several groups [34‐36]. We take the value reported by Sakaguchi et al. [35] (D_Sr = 3.0128 × 10⁻²¹ m² s⁻¹) given for Sr diffusion into the Lead Magnesium Niobate system. Using the above diffusion coefficient and applying the parabolic law for an SRO layer grown by PLD at 700 °C gives a diffusion depth of X = 9.5 nm. This is in close agreement with the microscopy observations reported in the EFTEM and EDS results shown here. Although these results are quantitative approximations, they not only confirm the influence of the growth process but also the important role played by the individual chemical species.

To compare the effect of lattice mismatch, the interface for PZT 52/48 films was investigated next. The mismatch between PZT 52/48 and SRO is ~3.93% (i.e., thrice larger than in the PZT 20/80 case) and thus a high density of misfit dislocations and other mismatch accommodating defects are expected.

Although both HPOAS and PLD films were investigated, we present here the case for the PLD sample, as both datasets were identical. The bright field image (Fig. 6) for this 200-nm thick PLD PZT 52/48 sample shows that it has a high density of dislocations, very fine 90° domains and significant interface roughness, similar to the HPOAS PZT (20/80) sample. EDS analysis (Fig. 7a) now not only shows chemically sharp interfaces for Ti and Ru (as in the previous system), but also for Sr. The same Sr sharpness is confirmed by EFTEM Sr maps (Fig. 7b). On the other hand, the EDS line scans for Pb clearly show Pb diffusion into the SRO layer. Thus we have a scenario that is found to be completely opposite to that obtained in the PZT 20/80 system where evidently Sr was the element migrating into the PZT and not Pb.

To explain this strong difference, we need to consider the relevant stages of formation of misfit dislocations versus 90° domains within the ferroelectric thin film, both during the deposition process and the subsequent cooling stage. It is critical to note that growth defects such as misfit dislocations primarily nucleate at the high growth temperatures (T_G ~ 575–650 °C) while 90° domains are formed during the process of cooling down. It can be expected that the dislocations are mobile at such high temperatures and moreover the temperatures at which dislocation formation occurs during the growth process would be typically above the Curie temperature (~400 °C for the PZT 52/48 case) and when the film is in a paraelectric state (i.e., no driving force to create ferroelastic domains). Thus for the case of the PZT 52/48 film where the interface has an extremely large density of misfit dislocations (~10¹²/cm²) [5], the dislocation cores can act as perfect channels for the Pb to migrate into the neighboring SRO layer. For the case of the PZT 20/80 system, the primary stress accommodating mechanism is the formation of 90° domains as the lattice mismatch is quite low (~1%). This domain formation process can occur only as the film cools down to undergo a paraelectric to ferroelectric phase transition. Thus it is possible that the nucleation of domains restricts the migration of the Pb into the SRO layer. It is also likely that the Sr penetration into the PZT 20/80 layers is a secondary stress accommodating mechanism as Sr doping into the ferroelectric layer significantly reduces the tetragonality and hence the paraelectric to ferroelectric transformation strain.

Preliminary high angle annular dark field (HAADF)/EDX results (data not shown) on a similar sample confirm several cases of diffusion of lead and titanium into SRO at the site of an interfacial dislocation. Furthermore nanoscale chemical maps of dislocation cores performed with the HAADF sample confirm that the dislocation cores do have excess Pb, with the regions in the PZT matrix directly above the core being Pb-deficient. These results are unlikely due to beam damage as we have acquired the EDS maps at an energy level of 80 to 120 kV and confirmed that the sample is undamaged by acquiring HAADF images before and after EDS analysis. More detailed HAADF imaging of dislocation cores in the 52/48 PZT sample versus the 20/80 case is currently underway and will be reported at a later stage.

Finally, we present analysis of a perfect BTO/SRO/STO heterostructure to elucidate the effect of non-volatile cations. The BTO sample was deposited by HPOAS and has a misfit of 2.6% [37]. The thickness for the BTO layer is ~50 nm and as shown in Fig. 8a has extremely smooth interfaces and no defects except for a few threading dislocations originating from the substrate. The EFTEM elemental jump-ratio images in Fig. 8c–f indicate chemically sharp interfaces for Ba, Ti, Ru, and Sr (i.e., all cation species) and hence these results are very different from the PZT systems presented before. It is worth mentioning that it is only EFTEM that identifies all the elements in the BTO heterolayer, since there is an overlap in the EDS signals for Ba and Ti, as indicated in Table 2. Nevertheless, EDS line scans were also acquired for Sr and Ru (data not shown) and they do agree with the EFTEM data. Notably, although BTO has a misfit almost twice larger than the PZT 20/80 misfit, a defect-free sample was achievable with no evidence of A-site diffusion, thus underlying the critical role of the stability/volatility of the cation species in obtaining chemically sharp interfaces.