At the heart of advancing these technologies lies the design and optimization of porous electrodes—engineered structures that facilitate ion and reactant transport within electrochemical devices. The intricate micro- and nanoscale architecture of these electrodes creates anisotropic pathways that govern mass transport phenomena. However, predicting and tailoring these transport properties presents a formidable challenge due to the complex, heterogeneous geometry. Traditional modeling and simulation methods are often computationally intensive and slow, creating a bottleneck that significantly hampers progress toward higher energy and power densities in electrochemical devices.

Addressing this critical bottleneck, a pioneering team of researchers has developed an innovative deep learning framework named Electrode Net. This approach leverages recent advances in artificial intelligence to model three-dimensional porous electrode geometries with exceptional accuracy and speed. By representing the electrode’s complex architecture through signed distance fields—a mathematical technique that precisely characterizes the three-dimensional geometry—the method feeds this representation into a sophisticated three-dimensional convolutional neural network (3D CNN). This network is trained to decode the relationship between microstructural features and the resultant anisotropic transport properties.

The researchers constructed a robust dataset comprising 15,433 samples of porous electrode microstructures, each paired with corresponding anisotropic transport data generated by a validated pore-network model. Through extensive training and testing on this dataset, Electrode Net demonstrated remarkable predictive performance, achieving coefficients of determination (R²) exceeding 0.95. This level of accuracy significantly surpasses that of existing advanced models, underscoring the efficacy of their deep learning approach in capturing the nuanced interplay between structure and function in porous media.

One of the most transformative aspects of Electrode Net is its computational efficiency. By harnessing the signed distance field representation and the inherent learning capabilities of the convolutional neural network, the model achieves an astounding reduction in computation time—up to 96% faster than conventional numerical simulations. Where traditional methods might require hours of intensive computation, Electrode Net can deliver predictions within minutes or even seconds. This rapid inference capability empowers researchers and engineers to explore expansive design spaces swiftly, facilitating accelerated screening and optimization of electrode geometries.

Crucial to the model’s utility and versatility, the team validated Electrode Net across multiple electrochemical technology platforms. Fuel cells, water electrolyzers, and redox-flow batteries—each with distinct operating regimes and electrode structures—served as testbeds to gauge the framework’s generalization ability. Impressively, the model sustained its high-level predictive accuracy across these diverse systems. This cross-technology reliability signals that Electrode Net is not confined to a single application but can serve as a universal tool to expedite electrode design across a spectrum of clean energy devices.

Building upon the model’s predictive prowess, the researchers introduced an integrated multiscale design workflow. This process begins with Electrode Net estimating pore-scale anisotropic transport parameters based on three-dimensional microstructural input. These parameters are then upscaled into cell-level simulation models, which account for realistic operational constraints and device physics. Through this hierarchical modeling paradigm, designers can optimize electrode architecture not purely from a theoretical standpoint but within the tangible context of device performance.

To illustrate this workflow’s practical effectiveness, the team applied it to the gas diffusion layer in proton-exchange-membrane fuel cells (PEMFCs), a critical component influencing overall efficiency and durability. Leveraging Electrode Net’s output, the cell-scale simulations guided modifications to the pore and fiber structures of the electrode, culminating in designs exhibiting significantly enhanced limiting power density and limiting current density. Such improvements promise tangible benefits in real-world fuel cell deployments, where maximizing power output while minimizing losses is paramount.

This melding of deep learning with physics-based simulation signifies a watershed moment for electrochemical device development. By directly learning from volumetric, three-dimensional structural data, Electrode Net bypasses long-standing inefficiencies inherent in traditional modeling. The method offers a scalable and generalizable framework capable of accommodating the multifaceted geometries typical of porous electrodes. This capability unlocks new pathways for rapid innovation, reducing the time from conceptual design to functional device.

Moreover, the implications of this research extend beyond immediate applications in renewable energy storage and conversion. The principles underpinning Electrode Net could be adapted to other domains where complex porous structures dictate functional performance, including catalysis, filtration, and biomedical engineering. The demonstrated approach illustrates how coupling advanced computational representations with data-driven methodologies can unravel complex microstructural relationships that have eluded analytical characterization.

In sum, the Electrode Net framework heralds a paradigm shift in porous electrode science and engineering. Through meticulous dataset curation, innovative geometric representation, and deep neural network design, the researchers have delivered a tool that effortlessly bridges microscopic complexity and macroscopic performance. This technology empowers the clean energy sector to surmount a critical bottleneck, accelerating the optimization of electrochemical reactors and ultimately aiding the worldwide transition to sustainable energy futures.

As the urgency for climate action fuels demand for next-generation energy solutions, tools like Electrode Net provide the necessary computational arsenal to keep pace. They enable a new era of materials informatics-driven innovation, fostering the design of power-dense, cost-effective, and durable electrodes. Such progress promises to bolster the viability of electrochemical devices at scale, contributing decisively to global net-zero ambitions and the broader goals of energy sustainability.

Subject of Research: Electrochemical porous electrode design and optimization using deep learning

Article Title: Electrode Net: A Deep Learning Framework for Multiscale Porous Electrode Optimization in Electrochemical Devices

Web References:

• DOI: 10.1016/j.scib.2025.08.026

Image Credits: ©Science China Press

Keywords

Porous electrodes, deep learning, electrochemical devices, fuel cells, water electrolyzers, redox-flow batteries, anisotropic mass transport, signed distance fields, convolutional neural networks, computational modeling, energy storage, renewable energy

At the core of this innovation is the molecular intricacy of dicyandiamide, a molecule endowed with guanidine and cyano functional groups, which drive synergistic effects at the buried interfaces within the perovskite architecture. The research reveals that the guanidine moiety preferentially binds with undercoordinated lead ions (Pb²⁺) and passivates vacancy defects inherent at the interface between perovskite layers. Concurrently, the cyano groups engage in coordination with titanium ions (Ti⁴⁺) in the electron transport layer (ETL), specifically titanium dioxide (TiO₂), alleviating electronic traps caused by oxygen vacancies. This molecular bridging facilitates not only defect passivation but also the robust coupling between the perovskite active layer and the ETL, enhancing charge transfer efficacy.

The implications of this intricate interface modification are profound. Experimentally, the DCD-mediated quasi-2D ACI perovskite solar cells demonstrate a remarkable leap in power conversion efficiency (PCE), reaching 21.54%, a significant jump from the 19.05% efficiency observed in unmodified control devices. This boost arises from the combined impact of reduced nonradiative recombination losses and an optimized phase distribution within the perovskite film. The suppression of low-n phases—known for their trap-mediated recombination—and the promotion of vertically aligned high-n phases facilitate uniform charge-carrier transport pathways, mitigating energetic disorder and interface recombination.

The research team’s comprehensive spectroscopy and theoretical investigations provide a mechanistic understanding that underpins these performance enhancements. X-ray photoelectron spectroscopy (XPS) and Fourier-transform infrared spectroscopy (FTIR) data confirm strong interactions between DCD molecules and both Pb and Ti centers, manifesting a decrease in surface defects. Notably, oxygen vacancy concentrations in TiO₂ are significantly diminished from 48% to 33%, a quantifiable indicator of improved interface quality. Complementary transient absorption (TA) and photoluminescence (PL) analyses further elucidate a homogeneous n-value phase distribution, effectively minimizing energy transfer losses that frequently curtail device efficiency.

Delving deeper into the molecular science, density functional theory (DFT) calculations reveal robust cyano-Ti coordination bonds that underpin the suppressed formation of interfacial traps, bolstered by guanidine-driven vacancy passivation at Pb sites. This dual-binding paradigm simultaneously stabilizes the buried interface and optimizes the ETL contact, an engineering feat that harmonizes the microstructural and electronic landscapes crucial for high-performance solar cells. The outcome is a more resilient perovskite-ETL interface that supports enduring device operation under practical stresses.

The optimized devices showcase outstanding photovoltaic parameters, including an open-circuit voltage (V_OC) of 1.172 V, a short-circuit current density (J_SC) of 23.08 mA/cm², and a fill factor (FF) approaching 79.6%. These metrics are symptomatic of efficient charge extraction and suppressed recombination pathways, conclusions supported by electrical impedance spectroscopy which finds recombination resistance elevated to an impressive 20.68 kΩ. The pronounced decrease in trap density, by over a factor of three, corroborates the enhanced charge carrier dynamics facilitated by the molecular bridging interface.

Beyond efficiency, the DCD-functionalized ACI perovskite solar cells demonstrate formidable operational stability—a critical criterion for commercial viability. The devices sustain 94% of their initial efficiency after prolonged exposure to thermal and environmental stress for 1200 hours, a substantial improvement over the 84% retention recorded in unmodified counterparts. Moreover, the modified cells endure continuous illumination stress for 400 hours without appreciable performance degradation, signaling robust photostability facilitated by the engineered interface.

This molecular bridge strategy achieves a long-sought decoupling of the conventional efficiency–stability trade-off that has historically hindered the advancement of 2D perovskite photovoltaics. By orchestrating the interplay between interface passivation and phase regulation, the approach yields films with reduced defect densities and uniform electronic landscapes, capable of sustained high performance. This methodology represents a paradigm shift, providing a scalable, chemically driven blueprint for next-generation perovskite solar cell fabrication.

Beyond the immediate scope of photovoltaic technology, this versatile interface engineering strategy holds potential applicability across a spectrum of perovskite-based optoelectronic devices, including light-emitting diodes (LEDs) and photodetectors. The molecular design principles exemplified here could be harnessed to tailor interfaces and phase behavior in diverse device architectures, thereby expanding the functional utility of perovskite materials in future photonic applications.

The study’s implications resonate across materials science, device physics, and chemical engineering disciplines, highlighting the power of targeted molecular modifications to optimize complex functional interfaces. This convergence of atomic-level chemical bonding insights and device-level performance improvements underscores the importance of interdisciplinary research for overcoming challenges in emergent energy conversion technologies.

In summary, the innovative work led by Professors Li, Song, and Zhang reveals how the precise incorporation of dicyandiamide molecules at critical perovskite interfaces fundamentally redefines the capabilities of quasi-2D ACI perovskite solar cells. Their molecular bridge strategy delivers record-setting efficiencies coupled with exceptional device stability, charting a promising pathway toward commercially viable and durable perovskite photovoltaics for the sustainable energy landscape of the future.

Article Title: Dicyandiamide-Driven Tailoring of the n-Value Distribution and Interface Dynamics for High-Performance ACI 2D Perovskite Solar Cells

News Publication Date: 23-Jun-2025

Web References: http://dx.doi.org/10.1007/s40820-025-01817-x

Image Credits: Ge Chen, Yunlong Gan, Shiheng Wang, Xueru Liu, Jing Yang, Sihui Peng, Yingjie Zhao, Pengwei Li, Asliddin Komilov, Yanlin Song, Yiqiang Zhang

Keywords: 2D Perovskite, Interface Engineering, Dicyandiamide, Solar Cells, Quasi-2D ACI Perovskites, Defect Passivation, Electron Transport Layer, Phase Regulation, Molecular Bridge, Photovoltaic Stability, Charge Carrier Dynamics