Overview
Perovskite solar cells can deliver high power-conversion efficiency, but operational stability remains a central barrier to deployment. Composition, defects, charge-selective interfaces, heat, light and moisture interact, so improving one metric in isolation can shift failure elsewhere.
A 2026 Science study reported an AI-guided workflow in which four specialized agents coordinated evidence collection and design decisions. The resulting devices used a FA0.92Cs0.08PbI3 absorber, a UV-resistant MeO-DPPACz molecular layer and metal-oxide interfaces. The champion device achieved 25.0% power-conversion efficiency and a steady-state value of 24.9%.
A Multi-Objective Stability Problem
Long-lived devices require simultaneous control of several degradation pathways. Halide migration, interfacial reactions, imperfect crystallization, mobile ions, electrode diffusion and environmental exposure can reinforce one another. The most useful AI system therefore needs to reason across material layers rather than optimize only an absorber formula.
Champion efficiency
Maximum reported power-conversion efficiency for the AI-guided device.
Steady-state efficiency
Reported stabilized performance, providing a more conservative device metric.
Performance retained
After 1,000 hours at 100°C under one sun, nitrogen and maximum-power-point tracking.
Damp-heat retention
Reported after 1,150 hours at 85°C and 85% relative humidity under the study's test conditions.
How the Four-Agent Framework Organized Design
| Agent | Primary role | Decision contribution |
|---|---|---|
| Data agent | Collects and structures published and experimental evidence. | Builds the evidence base and exposes missing or conflicting measurements. |
| Central agent | Coordinates objectives, constraints and recommendations. | Connects absorber, molecular and device-layer choices into one plan. |
| Composition agent | Searches absorber formulations and evaluates stability–performance trade-offs. | Prioritized the FA/Cs composition used in the validated device. |
| Interface agent | Targets molecular contacts, passivation and transport layers. | Guided selection of a UV-resistant molecular layer and dual metal-oxide protection. |
This structure is important because it makes the optimization modular without treating each layer as independent. A central coordinator can reject a locally attractive recommendation when it conflicts with a device-level constraint such as band alignment, chemical compatibility or process temperature.
What Was Designed
Absorber composition
The selected FA0.92Cs0.08PbI3 composition balances the optoelectronic advantages of formamidinium-rich iodide perovskites with cesium-assisted structural control. The article reports a trap density of 2.07 × 1010 cm−3 for the optimized Cs8 composition, consistent with improved film quality and reduced non-radiative loss.
Molecular contact
MeO-DPPACz was developed as a UV-resistant self-assembled molecular material. Molecular-level design at this interface affects energetic alignment, charge extraction, surface coverage and susceptibility to photochemical change.
Dual-sided inorganic interfaces
Metal-oxide layers on both sides of the absorber were used to stabilize contacts and limit detrimental interfacial reactions. The approach treats the perovskite as part of a complete device stack rather than as an isolated film.
Reading the Validation Metrics Correctly
| Test | Reported outcome | What it establishes |
|---|---|---|
| 100°C, one sun, N2, maximum-power-point tracking | >97% retained after 1,000 h | Strong operational heat stability under controlled accelerated conditions. |
| 85°C / 85% relative humidity | 98% retained after 1,150 h | Resistance to combined heat and moisture under the reported protocol. |
| Outdoor exposure | No obvious loss reported over 1,000 h | Useful real-environment evidence, but not a substitute for multi-season, multi-site qualification. |
| Lifetime projection | Approximately 30 years at 30°C | A model-based extrapolation from accelerated testing, not an elapsed field lifetime. |
Stress-test results depend on encapsulation, spectral distribution, bias condition, device area, humidity, atmosphere and failure definition. Comparisons between studies are defensible only when those conditions and uncertainty are reported consistently.
What AI Adds—and What Still Requires Experiments
The study demonstrates a practical role for AI as a research coordinator: integrate heterogeneous evidence, suggest interacting design changes and prioritize experiments that test the largest uncertainties. This can shorten iteration cycles when the search space includes composition, molecules and process parameters.
- Computational triage: electronic structure and molecular calculations can eliminate incompatible interfaces before fabrication.
- Uncertainty-aware experimentation: candidate selection should maximize information gain, not only predicted efficiency.
- Protocol discipline: device metadata and stress conditions must be machine-readable to prevent misleading cross-study comparisons.
- Scale-up evidence: champion small-area cells do not by themselves establish module yield, uniformity, encapsulation reliability or manufacturing cost.
- Independent replication: performance and longevity require confirmation across batches, laboratories and relevant field environments.
How CD ComputaBio Can Support Energy-Materials Research
A perovskite or interface-design program may require several computational levels, from rapid property comparison to detailed electronic and dynamical analysis. CD ComputaBio can organize these methods around a defined candidate-selection or mechanism question.
| Research need | Related support | Connection to the workflow |
|---|---|---|
| Evaluate candidate electronic structures | Electronic Property Analysis Service | Supports analysis of electronic descriptors and energy-level relationships. |
| Model ground-state structures and energies | DFT Calculation Service | Provides first-principles calculations for selected compositions and interfaces. |
| Compare molecular interface candidates | Quantum Chemistry Service | Characterizes molecular electronic properties relevant to contact design. |
| Quantify chemical descriptors | Chemical Property Calculations | Supports structured ranking across a defined candidate series. |
| Explore structural response | Molecular Dynamics Simulation Service | Examines modeled interactions and configuration changes over time. |
Contact Us
Developing an absorber, molecular interface or device-material shortlist? CD ComputaBio can help translate the design question into a tiered computational workflow with clearly defined outputs and validation gates.
References
- Liu Y, et al. AI-guided design of efficient perovskite solar cells operationally stable at 100°C. Science. 2026;392(6799):724–728. https://doi.org/10.1126/science.aef1620
- Khenkin MV, Katz EA, Abate A, et al. Consensus statement for stability assessment and reporting for perovskite photovoltaics based on ISOS procedures. Nature Energy. 2020;5:35–49. https://doi.org/10.1038/s41560-019-0529-5
For Research Use Only. This page summarizes published research and describes computational research services. Accelerated stability results and lifetime extrapolations are not guarantees of field lifetime, manufacturability or commercial performance.
