"AI Innovation Speeds Up the Identification of High-Efficiency Solar Cell Materials"

“AI Innovation Speeds Up the Identification of High-Efficiency Solar Cell Materials”


**Innovative Advancement in Hole-Transporting Layers Leveraging Machine Learning Moves Perovskite Solar Cells Closer to Record Efficiency**

A significant achievement in the evolution of perovskite solar cells has been realized through the application of artificial intelligence. Scientists from the University of Erlangen–Nuremberg (Germany), Karlsruhe Institute of Technology (Germany), and Ulsan National Institute of Science and Technology (South Korea) have developed a new hole-transporting layer (HTL), an essential element in solar cell architecture, that exhibits nearly record power conversion efficiencies (PCE). This advancement showcases the impact of machine learning in material exploration, greatly expediting the search for high-efficiency materials and revealing fundamental physical principles in solar energy design.

### The Significance of Hole-Transporting Layers in Solar Cells

Solar cells function by transforming sunlight into electrical energy through the excitation of electrons and formation of electron–hole pairs. The HTL plays a vital role in this mechanism, facilitating the movement of “holes” (the positive charges remaining when electrons are excited) to the positive terminal of the device. Its performance has a direct effect on the PCE of solar cells, a crucial metric for determining the efficiency of sunlight conversion into usable electricity.

Currently, only a limited set of HTL materials are in effective use. These have largely been developed through experimental trial-and-error methodologies, primarily focusing on chemical alterations to existing molecular frameworks. However, this strategy often lacks a thorough mechanistic comprehension, rendering the creation of advanced HTL materials a labor-intensive and resource-draining endeavor.

### Leveraging Machine Learning to Investigate the Chemical Landscape

The researchers tackled these obstacles by merging state-of-the-art materials science techniques with sophisticated machine learning (ML) methodologies. Machine learning allows researchers to navigate extensive chemical spaces much more rapidly and cost-effectively than conventional experimental approaches. In this investigation, a team led by Anastasia Barabash (University of Erlangen–Nuremberg) aimed to unveil not only superior HTL materials but also a more profound understanding of the principles that underlie their performance.

“Semiconductors for solar cells are frequently designed by merging donor and acceptor molecular elements,” Barabash elaborates. “We employed the well-known Suzuki reaction to address this challenge, as it facilitates the swift synthesis of conjugated molecules, enabling high-throughput experimentation.”

### A Data-Centric Materials Discovery Pipeline

The researchers initiated their work by compiling an initial dataset of over one million potential HTL candidates. From this vast collection, Jianchang Wu, a fellow researcher at Erlangen–Nuremberg, curated a varied group of 101 molecules, encompassing a wide spectrum of structural and chemical characteristics, including dimensionality (one-, two-, and three-dimensional), mobility, and solubility. These samples were subsequently synthesized and integrated into experimental solar cells, where their PCE and inherent material attributes were assessed.

The data acquired during this phase served as a training basis for the research team’s ML model. This model then pinpointed 24 additional promising candidates, notable for either their high potential efficiency or their capability to yield valuable insights into basic physical relationships. These candidates underwent a semi-automated synthesis and testing regimen, and after several rounds of optimization directed by the algorithm, the team successfully produced HTL materials with PCEs reaching up to 26.2%—close to the current record of 26.7%.

### Advancing Understanding of Solar Cell Materials

In addition to identifying high-efficiency materials, the team was able to generate multiple HTL candidates with comparable PCEs. This redundancy may be crucial for elucidating the physical mechanisms underlying the observed efficiencies. Pascal Friederich from Karlsruhe Institute of Technology underscores this point: “I find it intriguing to investigate whether self-driving laboratories can accomplish more than optimizing materials. Can they assist us in comprehending the fundamental physics at play?”

The capacity to develop multiple near-optimal materials from varying chemical configurations could provide fresh insights into molecular design principles, enabling researchers to fabricate HTLs with predictable and adjustable properties. This fusion of discovery and comprehension signals a substantial leap in materials science.

### Future Steps: Broadening the Machine Learning Toolkit

With their machine-learning-driven strategy confirmed, the researchers are now expanding their focus to other elements of perovskite solar cells, starting with the electron-transport layers (ETLs). These layers perform a function similar to HTLs but are responsible for transporting electrons instead of holes. The overarching aim is to enhance the entire architecture of perovskite solar cells, extending the limits of what the technology can accomplish regarding efficiency, stability, and scalability.

### A Significant Advancement for Sustainable Energy

Authorities in photovoltaics have praised this achievement for its groundbreaking application of machine learning in material discovery. Ted Sargent from Northwestern University points out the significance of this methodology: “The researchers demonstrate how machine learning can unveil hidden connections in material design, setting the stage for more efficient and stable perovskite devices.”

Cheng Liu,