Revealing ferroelectric switching character using deep recurrent neural networks

Abstract

The ability to manipulate domains underpins function in applications of ferroelectrics. While there have been demonstrations of controlled nanoscale manipulation of domain structures to drive emergent properties, such approaches lack an internal feedback loop required for automatic manipulation. Here, using a deep sequence-to-sequence autoencoder we automate the extraction of latent features of nanoscale ferroelectric switching from piezoresponse force spectroscopy of tensile-strained PbZr0.2Ti0.8O3 with a hierarchical domain structure. We identify characteristic behavior in the piezoresponse and cantilever resonance hysteresis loops, which allows for the classification and quantification of nanoscale-switching mechanisms. Specifically, we identify elastic hardening events which are associated with the nucleation and growth of charged domain walls. This work demonstrates the efficacy of unsupervised neural networks in learning features of a material’s physical response from nanoscale multichannel hyperspectral imagery and provides new capabilities in leveraging in operando spectroscopies that could enable the automated manipulation of nanoscale structures in materials.

Better, Faster, and Less Biased Machine Learning: Electromechanical Switching in Ferroelectric Thin Films

“…[29] Subsequently, the authors applied an NN-based autoencoder to the piezoresponse and contact resonance of the same dataset, and proposed a different interpretation based on contributions from charged domain walls in head-to-head polarization configurations. [31] We show that through careful model parameter selection, appropriate application of the physical understanding and constraints, and based on the knowledge of the system under investigation, we can gain better insights into the fundamental ferroelectric behavior in the datasets described above. Specifically, through the application of clustering methods, we identify and remove from consideration localized outliers and instrumentation contributions.…”

“…Here, we apply easily implementable and computationally inexpensive clustering and DR analysis methods augmented with dimensional stacking to previously reported [29,31] ferroelectric switching data collected via band-excitation piezoresponse force microscopy (BE-PFM) [40] on PbZr 0.2 Ti 0.8 O 3 thin films. Agar et al have previously identified the presence and interplay of both ferroelectric and ferroelastic switching in the piezoresponse through clustering and DR analysis, based solely on the piezoresponse, PR.…”

“…Further information on the characterization and with respect to the initial scan measurements can be found in Figures S1 and S2, Supporting Information and in the original reports. [29,31] A possible distribution of the c and a domains is identified in Figure 1, which is extrapolated from literature [42] and lateral images of another area of the same sample as discussed in Figure S3, Supporting Information.…”

“…All analyses (including the k-means seed, and the stacked DL) were completed within 1 to 2 h on a commercially available ASUS laptop with an 8th Generation Intel Core i5 8250U 1.6 GHz processor and 16 Gb RAM. This can be contrasted with a typical NN analysis, [31] which needs substantially more computational power and an order of magnitude more time to yield physically interpretable results.…”

“…[28,29] Similarly, while simple neural networks (NN) have been mostly used for predictive machine learning and image processing, [30] more advanced variants such as auto-encoders have been also leveraged to decompose complex characterization datasets. [31] While NN-based techniques show significant promise, they not only require substantial computational time and sophisticated hardware, but also have a greater energy cost than the simpler clustering and DR approaches. Additionally, the increasing monetary cost of computations and the trend toward growing monopolies in cloud service providers and data center hosts reduces the feasibility and wide scale adoption of NN approaches for analysis of large datasets.…”