Nazanin Bassiri-Gharb, Harris Saunders, Jr. Chair, Professor of the George W. Woodruff School of Mechanical Engineering, and the School of Materials Science and Engineering (MSE). She has been extremely successful in improving signal recovery for advanced imaging modalities such as HRTEM. Her new research — released this week in the journal Small Methods — elucidates what researchers say is a novel and transformative approach. This approach addresses longstanding issues in Piezoresponse Force Microscopy (PFM), a niche scanning probe microscopy technique.
Key to the research team’s approach has been addressing challenges with low signal-to-noise ratios (SNR), which frequently make effective data collection impossible in PFM. Co-lead authors Kerisha Williams and Henry Shaowu Yuchi are fueling this high-impact effort. Using state-of-the-art statistical machine learning techniques, they have created a framework for accurately reconstructing these elusive signals.
Challenges with Piezoresponse Force Microscopy
Piezoresponse Force Microscopy, or PFM, is a powerful material characterization technique that has become a workhorse, particularly for nanoscale materials investigations. Scientists are often challenged by areas with low SNR, where important information can be hidden behind the noise. This limitation results in untrustworthy results and prevents progress toward bringing new material science to fruition.
Bassiri-Gharb and her colleagues saw an opportunity to develop more rigorous analytical methods to provide a higher level of assurance about the reliability of PFM data. By comparing their calculated heating rates to known plasma behaviors, they tracked down errors that were a result of traditional signal-fitting procedures. These missteps pass over the most important lessons that we can learn from the underlying data.
To overcome these hurdles, the research team developed BayeSMG, a groundbreaking algorithm that recovers hidden signals with high accuracy. With the locomotion of prediction, this tactic narrates missing information to amplify appearance inaccurately. Further, it benefits researchers with perspective on how much confidence they can put in their findings through estimation of levels of uncertainty.
Promoting Transparency and Reproducibility
A big takeaway from Bassiri-Gharb’s research is the importance of transparency and reproducibility in all scientific studies. She supports widely sharing metadata, code, and processing details about her findings. She’s not alone in that mission—Kerisha Williams, her co-lead author, joins her in that goal. This promise goes hand in hand with the F.A.I.R. principles, which advocate for Findable, Accessible, Interoperable, and Reusable data in scientific research.
The desire for openness stems from a broader objective to serve as a model for the scanning probe microscopy (SPM) community and beyond. Bassiri-Gharb and her team are opening up their methodologies to the public. Their goals are to obtain other researchers to follow in their footsteps to implement practices that substantiate the credibility of scientific inquiry.
The use of statistical machine learning methods provides a more systematic analysis of the data. Researchers have been able to successfully recover lost signals with these cutting-edge techniques. They can identify bad data versus good data at a much higher level.
Collaborators and Future Implications
The study provides important lessons learned from a number of these collaborative partners. Among them are Kevin Ligonde, a Ph.D. student at the Woodruff School, and Mathew Repasky, a former Ph.D. student from the H. Milton Stewart School of Industrial and Systems Engineering (ISyE). Yao Xie, Coca-Cola Foundation Chair and Professor in ISyE, contributed his expertise to this innovative and path-breaking research.
The effects of this research reach far beyond PFM and into the larger field of materials sciences. By refining data recovery techniques and advocating for open research practices, Bassiri-Gharb’s work stands to benefit numerous applications in material characterization and analysis.