🔬 materials science

Accelerating discovery of infrared nonlinear optical materials with large shift current via high-throughput screening

This study employs a high-throughput screening strategy on over 154,000 materials to identify 32 infrared nonlinear optical candidates with strong shift current responses, revealing that layered structures with $C_{3v}$ symmetry and heavy $p$ -block elements are particularly promising for next-generation optoelectronic applications.

Original authors: Aiqin Yang, Dian Jin, Mingkang Liu, Daye Zheng, Qi Wang, Qiangqiang Gu, Jian-Hua Jiang

Published 2026-07-28

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Aiqin Yang, Dian Jin, Mingkang Liu, Daye Zheng, Qi Wang, Qiangqiang Gu, Jian-Hua Jiang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world of light not just as something that lets us see, but as a powerful river of energy waiting to be turned into electricity. For decades, we've relied on a specific kind of "dam" to catch this energy: the solar panel. These panels work by using a special junction (like a wall between two different materials) to push electrons one way, creating a current. But there's a catch: this old-school method has a hard ceiling on how much voltage it can produce, and it mostly ignores the invisible, warm part of the light spectrum known as infrared.

Enter a different, more magical way to catch light called the "bulk photovoltaic effect." Instead of needing a wall or a junction, this effect happens inside a single, perfectly symmetrical crystal that lacks a center point. When light hits it, the electrons don't just jump; they "shift" their position in real space, like a dancer taking a sudden, coordinated step to the side. This creates a steady flow of electricity without needing any external battery or voltage. Scientists call this "shift current." The big mystery has been: where are the best materials to do this, especially for infrared light? Finding them has been like searching for a needle in a haystack, usually by testing one material at a time and hoping for the best.

The Great Digital Hunt

In this study, a team of researchers decided to stop guessing and start searching with a super-powered net. They didn't just look at a few materials; they cast a massive digital net over a database containing more than 154,000 known inorganic compounds. Think of this database as a giant library of every known chemical recipe. The team built a clever, multi-step filter to sort through this library, looking for crystals that were stable, non-magnetic, and had the right kind of "lopsided" symmetry to allow electrons to shift.

After filtering out the impossible and the unstable, they were left with 2,519 promising candidates. They then ran high-speed computer simulations on all of them to see how well they could generate this "shift current." It was a marathon of calculation, but the results were a goldmine. They found 32 materials that showed a very strong response, but the real treasure hunt was just beginning.

The Infrared Champions

The team wanted to know which of these 32 champions could work with infrared light—the kind of light that carries heat and is used in night-vision goggles and remote controls. After running even more precise and detailed simulations (using a more accurate mathematical method to double-check their work), they narrowed it down to 9 special materials.

The star of the show is a compound called Sn5Ge2(SbTe5)2. In their simulations, this material showed a massive shift current response of 616 µA/V². To put that in perspective, the paper notes this is four times stronger than the best infrared material previously known. Other top performers included BiSb and Ge(SbTe2)2, which also showed huge responses in the infrared range.

The researchers discovered a secret recipe for success. The best materials weren't random; they all shared a specific "personality." They tended to have a layered structure (like a stack of pancakes) and a specific type of symmetry called C3v. Most importantly, they were packed with heavy elements from the "p-block" of the periodic table, specifically Tellurium (Te), Antimony (Sb), Germanium (Ge), and Selenium (Se). The paper suggests that these heavy atoms have electron clouds that are large and directional, allowing them to take that big "dance step" when hit by light, creating a powerful current.

From Simulation to Smart Prediction

But the team didn't just stop at finding these 9 materials. They realized that checking 154,000 materials one by one is slow, even for computers. So, they used their new data to train an artificial intelligence (AI) model. They taught the AI to look at the shape and ingredients of a crystal and predict how good it would be at generating shift current.

They tested this AI on materials it had never seen before, including some with very large and complex structures that are hard to simulate directly. The AI successfully identified two new, large-crystal candidates (Cs(Bi2Te3)2 and Sn(BiTe2)2) that looked promising. This proves that the dataset they created is a powerful tool that can help scientists find future materials much faster, without needing to simulate every single one from scratch.

What This Means for the Future

The paper concludes that while these materials are currently identified through computer simulations, they look very promising for real-world use. The researchers point out that many of these chemical combinations are similar to materials that scientists already know how to make in labs, suggesting that building these materials isn't just a dream.

If these materials can be made and tested in the real world, they could revolutionize how we interact with infrared light. Imagine self-powered sensors that don't need batteries, infrared cameras that are incredibly sensitive, or new ways to harvest energy from heat and light. The paper doesn't claim these devices exist yet, but it has provided the map and the coordinates for the next generation of scientists to build them. By combining high-speed computer screening with smart AI, the team has turned a needle-in-a-haystack problem into a targeted treasure hunt, revealing a new class of materials ready to power the future of light-based technology.

Technical Summary: Accelerating Discovery of Infrared Nonlinear Optical Materials with Large Shift Current via High-Throughput Screening

Problem Statement
The discovery of nonlinear optical (NLO) materials capable of generating strong shift current responses, particularly within the infrared (IR) regime, remains a significant challenge for next-generation optoelectronics. While the bulk photovoltaic effect (BPVE) offers an interface-free mechanism for generating steady direct-current photocurrents and above-bandgap photovoltages, practical applications are currently constrained by low photocurrents and poor energy conversion efficiency. Existing high-throughput efforts have been limited to specific structural families or small databases, leaving the vast chemical and structural space of the Materials Project (MP) database largely unexplored for high-performance IR shift-current materials. Furthermore, the lack of high-quality, large-scale datasets for shift current responses creates a bottleneck for developing AI-driven predictive and generative models.

Methodology
The authors employed an unbiased, multi-stage high-throughput screening workflow applied to the MP database, which initially contained over 154,000 inorganic materials. The screening process involved the following steps:

Filtering: A series of criteria were applied to identify non-centrosymmetric, thermodynamically reasonable, and computationally tractable candidates. Filters included symmetry (non-centrosymmetric), electronic structure (PBE bandgap $E_g > 0.1$ eV, non-magnetic), thermodynamic stability (energy above convex hull $E_{hull} \le 50$ meV/atom), elemental composition (exclusion of partially filled 3d shells and lanthanide/actinide elements), and unit cell size (fewer than 20 atoms). This reduced the pool to 2,519 candidates.
First-Principles Calculations (PBE Level): Electronic structures and optical responses for all 2,519 candidates were computed using the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional within the ABACUS package. The shift current conductivity tensor $\sigma_{abc}$ was calculated using the PYATB package based on the DFT Hamiltonian.
Refined Calculations (HSE Level): The top 32 candidates, defined by a PBE shift current response $|\sigma_{abc}|_{max} > 100$ µA/V $^2$ , underwent further evaluation using the Heyd-Scuseria-Ernzerhof (HSE) hybrid functional to obtain more accurate bandgaps and NLO responses.
Machine Learning Validation: To demonstrate the utility of the generated dataset for AI-driven discovery, a prototype machine learning model, DPA3-σ, was developed. This model utilized transfer learning from the pre-trained DPA3 graph neural network (originally designed for interatomic potentials) to predict the maximum shift current response from structural embeddings.

Key Contributions

Comprehensive Dataset: The study establishes one of the most comprehensive first-principles datasets for shift current response to date, covering 2,519 non-centrosymmetric materials with detailed PBE and HSE level calculations.
Design Principles: The work identifies specific structural and chemical design principles for enhancing shift current: materials with $C_{3v}$ symmetry (specifically the trigonal $R3m$ space group) containing heavy p-block elements (notably Te, Sb, Ge, and Se) exhibit superior performance.
AI Pipeline Demonstration: The paper validates a transfer learning strategy (DPA3-σ) for screening large-scale databases, demonstrating the model's ability to generalize from small training sets to larger, out-of-domain systems (20–60 atoms) with high recall and precision.

Results

Candidate Identification: From the 2,519 screened materials, 32 candidates with strong shift current responses ( $|\sigma_{abc}|_{max} > 100$ µA/V $^2$ ) were identified at the PBE level.
IR Region Discovery: After HSE refinement, 9 materials were confirmed to exhibit pronounced shift current peaks within the IR region. The strongest candidate, Sn $_5$ Ge $_2$ (SbTe $_5$ ) $_2$ (mp-1219067), displays a peak shift current of 616 µA/V $^2$ at 0.08 eV (HSE level), which is approximately four times greater than the previously reported topological semimetal TaAs.
Material Properties: The identified IR candidates include both thermodynamically stable and metastable compounds. Most crystallize in layered trigonal structures with broken inversion symmetry. The shift current response is dominated by in-plane tensor components (e.g., $xxy$, $yxx$, $-yyy$) for $C_{3v}$ materials.
Spectral Distribution: Among the top 32 candidates, 9 exhibit IR response, 18 show visible (VIS) response, and 5 show ultraviolet (UV) response. The HSE correction generally reduced the magnitude of the shift current peaks and blue-shifted the resonant energies compared to PBE results.
Machine Learning Performance: The DPA3-σ model achieved a recall of 80.0% and precision of 80.0% when screening for candidates with $\sigma > 100$ µA/V $^2$ . When applied to a subset of the MP database containing complex systems (>30 atoms), the model successfully identified two new high-response candidates, Cs(Bi $_2$ Te $_3$ ) $_2$ and Sn(BiTe $_2$ ) $_2$ .

Significance
The paper claims that this work significantly expands the material landscape for bulk photovoltaic effects from the visible/ultraviolet spectrum into the infrared region. By identifying materials with strong IR shift current responses, the study provides promising platforms for applications such as IR photodetection, self-powered sensing, and energy harvesting without external bias. Furthermore, the establishment of a high-quality, large-scale first-principles dataset addresses a critical gap in NLO research, serving as a foundational resource for future data-driven discovery and the development of predictive machine learning models. The successful demonstration of the DPA3-σ pipeline illustrates the feasibility of using transfer learning to accelerate the screening of complex, large-unit-cell materials that are computationally expensive to evaluate with standard DFT methods. The authors emphasize that while spin-orbit coupling (SOC) was not included in the current screening, the identified design principles (polar $C_{3v}$ symmetry and heavy p-block chemistry) offer a clear microscopic rationale for future, more detailed investigations and experimental validation.

More like this