TL;DR: Penn State researchers paired a large language model with expert review to identify a lead-free KNN-BNKZT-SCZ ceramic that reached 440 pC/N and 620 pC/N after texturing, then delivered 705 microwatts per cubic centimeter in an energy harvester.
AI plus experts narrows a slow materials hunt
A tedious process of trial and error is normally required to discover ceramic materials that generate electricity under mechanical stress, according to the Penn State University report titled AI streamlines search for new ceramics that generate electricity.
Those materials are known as piezoelectric ceramics, and the report describes them as valuable next-generation materials that can be used to replace batteries in powering small electronic devices, like sensors.
A team led by materials science researchers at Penn State introduced a new framework that combines an existing artificial intelligence model with human expertise to uncover these materials.
To test the new approach, researchers developed a new piezoelectric composite and incorporated it into an energy harvester, which converts vibrations or movements into electricity.
The announcement was published from University Park, Pennsylvania, on August 31, 2026, by authors Heather Dehnel and Mariah Lucas.
The work is detailed in Nature Communications, in a paper titled Design and realization of high performance textured lead-free piezoelectric ceramics through human-AI collaboration.
The Penn State University account says the new approach comprises a three-phase research process, where an AI tool identifies possible material composites that are assessed by researchers before they put the selected potential composites to the test with hands-on experimentation.
Researchers prompted a large language model to analyze patterns from decades of previously published research and yield a list of new material compositions, the university report states.
The related summary carried by Phys.org under the same headline describes the same sequence: an AI tool identifies possible composites, researchers assess the options, and selected composites move to hands-on experimentation.
Wesley Reinhart, assistant professor of materials science and engineering at Penn State, compared the system to an assistant that had reviewed a large body of prior papers and could suggest promising materials.
“That is essentially what we created, but with scientists providing the critical physics, chemistry and engineering judgment needed to turn those suggestions into real-world technologies,” Reinhart said, as quoted by Penn State University.
Aman Nanda, doctoral student in materials science and engineering and first author of the study, said using AI to identify promising candidates for experimentation frees up researchers to focus their efforts more on validation and optimizing materials for applications.
The university account identifies Nanda as first author and Bed Poudel, research professor of materials science and engineering, as corresponding author.
The publication record lists Aman Nanda, Debjyoti Bhattacharya, Nayeon Kang, Shankar Kunwar, Shashank Priya, Jungho Ryu, Wesley Reinhart, Michael Lanagan and Bed Poudel as contributors.
The same record gives the journal as Nature Communications, volume 17, issue 1, article number 8001, with DOI 10.1038/s41467-026-74568-x.
Why piezoelectric ceramics matter for small devices
Piezoelectric ceramics generate electricity under mechanical stress, meaning movement, vibration or pressure can produce an electrical response.
The Penn State University report says that property makes the materials candidates to replace batteries that power small electronic devices, like sensors.
Energy harvesters are central to that use case because they convert vibrations or movements into electricity, rather than storing charge supplied elsewhere.
The researchers tested their AI-suggested material by incorporating it into an energy harvester, directly linking the composition search to a device demonstration.
Phys.org likewise states that researchers developed a new piezoelectric composite and incorporated it into an energy harvester to test the approach.
The device used in the demonstration was a magneto-mechano-electric energy harvester designed to convert mechanical and magnetic energy into usable electrical power.
Such devices are typically used to power wireless sensors, according to both the university and Phys.org accounts.
Wireless sensors often operate in places where battery replacement is inconvenient, and the reported work positions vibration and motion conversion as an alternative power pathway for those devices.
The study focused on lead-free chemistry, specifically multielement-doped lead-free potassium sodium niobate piezoelectrics, according to the publication abstract.
Lead-free compositions are the explicit materials family examined in the Nature Communications paper, which describes the challenge as complex doping trends and tedious trial-and-error experimentation.
The abstract states that the team presented a human-in-the-loop, artificial intelligence-guided materials design framework that uses large language models to capture implicit structure-property knowledge from prior literature and propose new compositions.
It also states that expert intervention further directs experimental realization based on materials science principles and experiential knowledge, accelerating discovery of targeted compositions.
From literature patterns to KNN-BNKZT-SCZ
After creating a list of possible piezoelectric materials, researchers decided to focus on a potassium sodium niobate-based composition, abbreviated KNN-BNKZT-SCZ.
That selection step followed the AI-assisted review of patterns from decades of previously published research, as described by Penn State University.
The composite has a piezoelectric charge constant of approximately 440 picocoulombs per newton, the university report states.
That value means it can generate roughly three times more electrical output from the same amount of movement or vibration than conventional potassium sodium niobate-based materials.
The publication abstract gives a closely corresponding range for the synthesized random composition, reporting a piezoelectric charge constant d33 of 440–500 pC/N.
The abstract also says performance was further enhanced to 600–620 pC/N through crystallographic texturing and sintering-aid optimization.
The university and Phys.org summaries report the textured result as roughly 620 picocoulombs per newton.
Researchers further improved the material’s performance through crystallographic texturing, a process that aligns grains within the ceramic to maximize its electrical properties, according to Penn State University.
Michael Lanagan, professor of engineering science and mechanics and co-author of the paper, said the AI program identifies patterns across large amounts of scientific literature, while successful discovery still requires understanding of chemistry, crystal structures, processing and characterization.
“That partnership significantly helped develop the strategy for texturing, which showed significant improvement from 440 to 620 picocoulombs per newton,” Lanagan said, as quoted in the university report.
The texturing result therefore represents an increase of about 180 picocoulombs per newton over the initial approximately 440 value reported for the selected composition.
The improved amount of voltage generated from a mechanical vibration made the material’s performance competitive to existing materials, researchers said, as reported by Penn State University.
Phys.org uses closely matching language, stating that the increased voltage generated from mechanical vibration made the material’s performance competitive with existing materials.
Heat tolerance and electrical stability
The textured material maintained its ability to operate in high-temperature environments of up to 320 degrees Fahrenheit, according to Penn State University.
Phys.org gives the same limit as 320°F and adds the metric equivalent of 160°C.
The publication abstract likewise reports steady behavior up to 160°C, stating that the composition maintained steady electromechanical coupling and d31 up to that temperature.
The abstract phrases the temperature finding in the context of an inherently off-morphotropic-phase-boundary condition at room temperature, noting steady coupling despite that degradation point.
That stability matters because the reported device goal involves converting mechanical and magnetic inputs into usable power outside tightly controlled laboratory temperatures.
The university report does not claim continuous operation at higher temperatures, only that the textured ceramic maintained its ability to operate in environments of up to the stated limit.
The textured result maintained high charge response while retaining that temperature capability, rather than trading one reported property entirely for the other.
The abstract lists sintering-aid optimization alongside crystallographic texturing as part of the route to the 600–620 pC/N range.
The university summary focuses its process description on crystallographic texturing as the grain-alignment step used to maximize electrical properties.
Together, the accounts identify composition selection, grain alignment and processing optimization as the experimental stages after the AI-assisted proposal step.
Device test reaches 705 microwatts per cubic centimeter
To demonstrate the material’s practical potential, the researchers incorporated KNN-BNKZT-SCZ into a magneto-mechano-electric energy harvester.
The harvester was designed to convert mechanical and magnetic energy into usable electrical power, according to Penn State University.
The device, which is typically used to power wireless sensors, achieved a power density of approximately 705 microwatts per cubic centimeter.
That result outperformed comparable systems and the material’s non-textured counterpart, the university report states.
Phys.org reports the same 705-microwatt-per-cubic-centimeter power density and the same comparison against comparable systems and the non-textured version.
The publication abstract specifies that a cantilever-based magneto-mechano-electric energy harvester was fabricated and delivered a power density of about 705 microwatts per cubic centimeter at the second harmonic.
The abstract further states that the result outperformed reported lead-free magneto-mechano-electric designs.
“Developing an outstanding material is only the first step; its true value lies in demonstrating reliable performance in a functional device,” Poudel said, as quoted by Penn State University.
Poudel added that validating the material in a functional device gives confidence that AI-guided discovery can accelerate development of high-performance materials for practical applications.
The device result therefore connects three reported stages: AI-assisted composition proposal, laboratory improvement through texturing, and integration into a harvester used for sensor-type power.
The university report presents the harvester test as a practical-potential demonstration, not as mass production or commercial deployment.
Who did the work and how it was supported
The team that created the AI-based framework was led by Penn State materials science researchers, according to the university report.
A group photograph identified Shankar Kunwar, postdoctoral scholar in materials science and engineering; Michael Lanagan, professor of engineering science and mechanics; Bed Poudel, research professor of materials science and engineering; Aman Nanda, doctoral student in materials science and engineering; and Debjyoti Bhattacharya, doctoral student in materials science and engineering.
The photograph credit is listed as Provided by Ashis Jacob.
Other co-authors include Nayeon Kang and Jungho Ryu, both affiliated with Yeungnam University in South Korea.
Another co-author is Shashank Priya, vice president for research and innovation at Michigan State University, who formerly served as Penn State’s vice president for research and earned a doctorate in materials science and engineering from Penn State.
The contributions to the work by Penn State researchers were supported by the Office of Naval Research under award number N00014-22-1-2691.
Additional Penn State support came from the U.S. Army DEVCOM Army Research Laboratory Army Research Office under award number W911NF-23-2-0229.
The university report notes that the full list of funders is available in the paper.
A schematic illustration of the large-language-model-assisted compositional discovery framework for potassium sodium niobate-based piezoelectric ceramics appeared with the research coverage and was credited to Nature Communications in 2026.
What the three-phase method changes
The reported framework does not remove laboratory work; it changes where researchers spend time.
The first phase uses a large language model to review prior literature patterns and propose compositions.
The second phase applies researcher assessment based on physics, chemistry and engineering judgment.
The third phase tests selected composites through hands-on experimentation, optimization and device integration.
Nanda’s stated point is that AI identification of candidates lets researchers concentrate more on validation and application-focused optimization.
Reinhart’s quoted explanation emphasizes scientists supplying physics, chemistry and engineering judgment to convert suggestions into technologies.
Lanagan’s quoted explanation emphasizes chemistry, crystal structures, processing and characterization as the expertise still required for successful discovery.
Poudel’s quoted explanation emphasizes functional-device validation as the measure of practical value.
In this demonstration, those roles produced a specific lead-free composition, a textured variant with higher charge response, documented high-temperature operation, and a harvester power-density measurement above comparable lead-free systems described in the paper.