Nanoscience News (<front>)

Sheaf-Based Diffusion for Multimodal Graph Learning

Multimodal Graph Learning (MGL) is an emerging area in machine learning that focuses on graphs whose nodes carry information from different modalities, such as text and image. A central challenge in MGL is integrating these heterogeneous data types, which are not directly comparable. Standard Graph Neural Networks (GNNs) struggle in multimodal contexts because they assume homogeneity in node features and tend to merge modalities too early, leading to the loss of valuable, modality-specific information. Existing solutions address this by processing each modality independently and fusing their predictions at the output level. However, recent studies show that these late-fusion strategies underperform compared to general-purpose GNNs. To address this limitation, we introduce MMSheaf, a family of sheaf-based neural network architectures that preserve modality separation before diffusion and introduce structured, learnable mechanisms for cross-modal interaction during message passing. As a first contribution, we show that Sheaf Neural Networks (SNNs) outperform standard GNNs like GCN or GAT on multimodal graphs, proving to be an appropriate tool for this context. Building on this insight, our MMSheaf architecture further improves performance by explicitly modeling cross-modal interactions. We evaluate MMSheaf on synthetic multimodal datasets where successful classification requires integrating modalities in a non-trivial way. Additional experiments on the real-world Ele-Fashion dataset showcase the model’s effectiveness in practical multimodal settings. Overall, our findings establish sheaf-based diffusion as a powerful and expressive framework for Multimodal Graph Learning. Future work will apply this approach to diverse domains such as biomedicine and recommender systems.

Meet link: meet.google.com/wtt-wydt-hfk

• Speaker: Mar Gonzàlez i Català (Universitat Politècnica de Catalunya)
• Tuesday 13 May 2025, 16:00-16:45
• Venue: Lecture Theatre 2, Computer Laboratory, William Gates Building.
• Series: Foundation AI; organiser: Pietro Lio.

Add to your calendar or Include in your list

ECFGel is a multifunctional hydrogel engineered to treat infection-associated ECFs. ECFGel demonstrates outstanding mechanical and biological properties, facilitating easy application, reliable occlusion, and sterilization, while promoting effective healing of infected fistula tracts. A choline and geranate-based ionic liquid hydrogel is used to concurrently enhance mechanical performance and confer antimicrobial functionality.

Abstract

Enterocutaneous fistulas (ECFs) profoundly impact patients’ quality of life, contributing to high morbidity rates and increased mortality due to ineffective treatment options. To address this challenge, ECFGel, a multifunctional, tissue adhesive injectable hydrogel, designed to occlude, sterilize, and promote healing of ECF tracts, is developed. ECFGel is formulated using gelatin and oxidized dextran (O-Dex) as base components, which form chemical crosslinks within the hydrogel and with surrounding biological tissues, ensuring tissue adhesiveness. A choline and geranate-based ionic liquid (IL) is incorporated to provide dual functionality, potent antimicrobial activity, and mechanical enhancement. By optimizing IL concentration, ECFGel achieves rapid gelation, enhanced mechanical strength, and improved elastic recoverability. Additionally, iohexol (IOH) is added for radiopacity, enabling real-time imaging and further strengthening the hydrogel's mechanical properties. ECFGel demonstrates antiswelling properties, biodegradability, and effective tract occlusion in porcine soft tissues. It shows strong antimicrobial activity against highly resistant, patient-derived pathogens isolated from clinical ECF cases. In a porcine perianal fistula model, ECFGel enables rapid occlusion and complete healing, promoting tissue maturation, reducing bacterial load, and increasing markers of cell proliferation and vascularization compared to untreated controls. These promising results highlight ECFGel's potential as a new therapeutic option for treating infected ECFs.

Sustainable acrylic optically clear adhesives (OCAs) for foldable displays are developed by incorporating ultraviolet (UV)-triggered debondability and lipoic acid–based degradability through visible-light-induced bulk polymerization. These OCAs enable residue-free removal and degrade into small oligomers or recoverable monomers under mild conditions, supporting substrate recycling and adhesive recovery in next-generation flexible electronics.

Abstract

In advanced applications such as flexible displays, reusing components is essential for achieving sustainability. However, the removal of acrylic pressure-sensitive adhesives (PSAs), which bond these components, remains a major challenge due to residue formation and the non-degradable C─C backbone. Here, the development of new acrylic PSA alternatives for foldable displays is reported by introducing ultraviolet (UV)-triggered debondability, degradability through lipoic acid analogs, and a visible-light-curing process. PSAs composed of 60 mol% lipoic acid ethyl ester (LpEt) and 3 mol% UV-cross-linkable benzophenone-functionalized acrylic monomers exhibit viscoelastic properties comparable to those of conventional acrylic PSAs, while also enable clean removal from substrates after use. Following removal, the PSAs efficiently degrade into small molecular units in the presence of a green reductant or can be recovered as monomers under controlled conditions. This strategy offers a promising pathway toward sustainable PSAs, enables the recycling of valuable substrates from flexible display modules while simultaneously allows adhesive recovery, thus presents a viable alternative to conventional acrylic adhesives.

Mimicking energy transduction in prototissue assemblies remains a challenge of bottom-up synthetic biology. In this work, prototissues integrating protocells with photothermal gold nanoparticle proto-organelles and a thermoresponsive polymeric proto-cortex are developed. These prototissues exhibited light-controlled reversible contractions, programmable motions, and dynamic control of enzymatic metabolism by restricting substrate access, pioneering a route toward bioinspired materials with emergent energy transduction behaviors.

Abstract

Despite recent significant advances in the controlled assembly of protocell units into complex 3D architectures, the development of prototissues capable of mimicking the sophisticated energy transduction processes fundamental to living tissues remains a critical unmet challenge in bottom-up synthetic biology. Here a synthetic approach is described to start addressing this challenge and report the bottom-up chemical construction of a photonastic prototissue endowed with photo-mechano-chemical transduction capabilities. For this, novel protocells enclosing photothermal transducing proto-organelles based on gold nanoparticles and a thermoresponsive polymeric proto-cortex are developed. These advanced protocell units are assembled into prototissues capable of light-induced reversible contractions and complex motions, which can be exploited to reversibly switch off a coordinated internalized enzyme metabolism by blocking the access of small substrate molecules. Overall, the work provides a synthetic pathway to constructing prototissues with sophisticated energy transduction mechanisms, enabling the rational design of emergent behaviors in synthetic materials and addressing critical challenges in bottom-up synthetic biology and bioinspired materials engineering.

Room-temperature air-blade coating facilitated the fabrication of large-area organic solar modules using eco-friendly solvents via controlling liquid-to-solid transition with directional gas flow, achieving efficient opaque and colorful semitransparent modules.

Abstract

Organic solar modules (OSMs) hold potential for building-integrated photovoltaics, yet facing challenges to fabricate uniform and large-area active layers over non-halogenated solvent coating. In this work, room-temperature air-blade assisted (RT/A) coating is presented that helps obtaining uniform active layers under ambient and non-halogenated solvent processing. It is revealed that RT/A coating mitigates the film inhomogeneity that is commonly observed during hot-substrate coating. Different to the thermal gradients-induced inhomogeneous liquid-to-solid transition of hot-substrate coating, RT/A strategy enables the control of transition time on film formation via directional gas flow to yield high-quality active layer blends at ambient coating. Large-area active layers from RT/A coating exhibit good consistency and uniformity. The resultant OSMs achieve high efficiencies with the certified PCE of 14.5% at 19.31 cm2 area (recorded in solar cell efficiency tables, version 60). By further integrating Fabry–Pérot transparent electrodes, colorful and semitransparent modules with PCEs of 12.80% are successfully developed. Overall, this work provides a promising method on the scalable fabrication of organic photovoltaics.

The work develops a binary aerogel-hydrogel system (SHA-HVG), achieving efficient water desalination (2.75 kg m−2 h−1) and high power density (56.86 µW cm−2) through enhanced interfacial evaporation and ionic concentration gradients. This study presents a cost-effective HVG strategy for water desalination and electrical energy harvesting, which is expected to promote the development of distributed energy, smart agriculture, and offshore ecosystems.

Abstract

Hydrovoltaic generators (HVGs) convert abundant water energy into distributed electricity to promote the Internet of Things. Realizing low-cost yet high-performance HVG remains challenging, hindering its commercialization and application. Inspired by the xylem conduits in plants, which transport water and nutrients, an aerogel-hydrogel binary-component system (SHA-HVG) is developed. It consists of a photothermal graphite-doped polyvinylidene fluoride (G-PVDF) aerogel, infilled with a thermosensitive wettability-switchable sulfonic acid-modified polyisopropylacrylamide hydrogel (S-PNIPAM) by in situ polymerization, which significantly promotes water/ion transporting and boosts electricity output. SHA-HVG demonstrates all-weather high output by cooperating power generation mechanisms of thermosensitive hydrogel-promoted surface photothermal evaporation during the daytime and sulfonic group-enhanced ion concentration gradient at nighttime, resulting in efficient water desalination (2.75 kg m−2 h−1) and a 2669% increase in power density (56.86 µW cm−2) compared to single-component HVG of G-PVDF. SHA-HVG is chemically stable and can be reactivated/recycled to improve its power generation efficiency to ∼130% by increasing its built-in ionic environment. A marine/offshore cultivation system is demonstrated using an SHA-HVG array, realizing an autonomous greenhouse for water desalination, self-irrigation, and self-powered environment monitoring. This work presents a cost-effective HVG strategy for efficient seawater desalination and electricity harvesting, envisioning the development of distributed energy, smart agriculture, and offshore planting.

A charge lock-free TENG is proposed, which has excellent electrical output performance and durability. This study provides a charge-lock-free strategy to effectively release the charge locked on the interface, which realizes efficient electrical energy extraction from sliding frictional interfaces.

Abstract

Sliding-mode triboelectric nanogenerators (TENGs) generate electricity by utilizing dynamic friction on the surface of dielectric materials, demonstrating notable potential for low-frequency mechanical energy harvesting. However, conventional device structures are limited by interface charge locking and low inherent capacitance, which results in a relatively low power density. Herein, the charge-locking mechanism is explored, and propose an innovative strategy for the dynamic interface shift between two materials with different polarities. When the interfaces overlap in the same materials, the locked interface charges are fully released. Based on this concept, a charge lock-free TENG (LF-TENG) is constructed. The output energy of the LF-TENG is 4.44 times that of the traditional sliding TENG. The rotating LF-TENG achieves an output charge and a current of 5 µC, 100 µA at 60 rpm, and an average power density of 9 W m−2 Hz−1 at 70 MΩ. Furthermore, the equivalent circuit model is analyzed and it is found that the capacitance change of the LF-TENG is twice that of traditional devices, which is the main factor for the increase in the output power. This study provides valuable insights into efficient electrical energy extraction from sliding frictional interfaces and paves the way for the development of low-frequency mechanical energy-collection technologies.

The wafer-scale hexagonal boron nitride (h-BN) epitaxial film exhibits pronounced anisotropy in optical absorption and carrier transport stemming from its distinct m-plane surfaces, greatly favoring its polarized vacuum ultraviolet (VUV) detection. The device demonstrates excellent polarization-sensitive performance under 188 nm linearly polarized light, along with a remarkable polarization ratio, thus extending the short-wavelength limit of existing lensless polarization-sensitive photodetection technologies based on anisotropic semiconductor materials.

Abstract

Vacuum ultraviolet (VUV) detection plays an essential role in space science, radiation monitoring, electronic industry, and fundamental research. Integrating polarization characteristics into VUV detection enriches the comprehension of the target attributes and broadens the signal dimensionality. Polarization detection has been widely developed in visible and infrared regions; however, it is still relatively unexplored in VUV light due to the lack of photoactive materials with low-symmetry structures, VUV selective response and radiation resistance. Here, the wafer-scale hexagonal boron nitride (h-BN) epitaxial films with the distinct m-plane surfaces are demonstrated that exhibit significant anisotropy due to space symmetry breaking, instead of the routinely obtained high-symmetry c-planes governed by the most thermodynamically stable growth mode. This results in notable anisotropy in light absorption and charge density distributions, yielding a dichroic ratio greater than 10 and a carrier transport efficiency ratio (μτa -axis/μτc -axis) of 24. The h-BN based detector achieves a high polarization ratio of 6.2 for 188 nm VUV polarized light, reaching the short-wavelength limit of the reported polarization-sensitive photodetectors. This work presents an effective strategy for designing polarized VUV photodetector from h-BN, and paves the road towards the novel integrated optoelectronics, photonics and electronics based on traditional 2D materials.

This review summarizes the recent progress of all-inorganic Sn-containing PSCs in the aspects of efficiency and stability, including the basic properties and degradation mechanisms/pathways of pristine Sn and mixed Sn-Pb perovskites, as well as various strategies to improve the photovoltaic performance of devices, discuss the existing challenges in this field, and look forward to the prospects for further improvement.

Abstract

All-inorganic tin (Sn)-containing perovskites have emerged as highly promising photovoltaic materials for single-junction and tandem perovskite solar cells (PSCs), owing to their reduced toxicity, optimal narrow bandgap, and superior thermal stability. Since their initial exploration in 2012, significant advancements have been achieved, with the highest efficiencies of single-junction and tandem devices now surpassing 17% and 22%, respectively. Nevertheless, the intrinsic challenges associated with the oxidation susceptibility of Sn2+ and the uncontrolled crystallization dynamics impede their further development. Addressing these issues necessitates a comprehensive and systematic understanding of the degradation mechanisms inherent to all-inorganic Sn-containing perovskites, as well as the development of effective mitigation strategies. This review provides a detailed overview of the research progress in all-inorganic Sn-containing PSCs, with a particular focus on the basic properties and degradation pathways of both pristine Sn and mixed Sn-Pb perovskites. Furthermore, various strategies to improve the efficiency and stability of Sn-containing PSCs are thoroughly discussed. Finally, the existing challenges and perspectives are provided for further improving the photovoltaic performance of eco-friendly PSCs.

This review examines the critical role of neighboring active motifs in heterogeneous catalysts, emphasizing how their synergistic cooperativity can be rationally designed to modulate catalytic properties. The discussion centers on the modulation of electronic and geometric structures, categorizing the catalytic systems into metal-acid (Brønsted/Lewis) pairs, metal-base pairs, and metal-metal pairs. The goal is to provide a systematic overview that offers insights into the design and characterization of next-generation multifunctional catalysts.

Abstract

Understanding the intricate interplay between catalytically active motifs in heterogeneous catalysts has long posed a significant challenge in the design of highly active and selective reactions. Drawing inspiration from biological enzymes and homogeneous catalysts, the synergistic cooperation between neighboring active motifs has emerged as a crucial factor in achieving effective catalysis. This synergistic control is often observed in natural enzymes and homogeneous systems through ligand coordination. The synergistic interaction is especially vital in reactions involving tandem or cascade steps, where distinct active motifs provide different functionalities to enable the co-activation of the reaction substrate(s). Situated within a 3D spatial domain, these catalytically active motifs can shape favorable catalytic landscapes by modulating electronic and geometric characteristics, thereby stabilizing specific intermediate or transition state species in a specific catalytic reaction. In this review, we aim to explore a diverse array of the latest heterogeneous catalytic systems that capitalize on the synergistic cooperativity between neighboring active motifs. We will delve into how such synergistic interactions can be utilized to engineer more favorable catalytic landscapes, ultimately resulting in the modulation of catalytic reactivities.

Dynamic imprint and seismic signature of mineral phase transitions in the Earth's deep mantle

Abstract not available

• Speaker: Isabel Papanagnou
• Friday 16 May 2025, 16:00-17:00
• Venue: Tea Room, Old House.
• Series: Bullard Laboratories Tea Time Talks; organiser: David Al-Attar.

Add to your calendar or Include in your list

Nature Materials, Published online: 12 May 2025; doi:10.1038/s41563-025-02232-8

Epithelial cell–cell contact resistance to mechanical stress is defined by a regulated viscous dissipation through the cytoskeleton, where the toughness of the cell junction is set by the balance between cortical tension and cell shape recovery time.

Nature Materials, Published online: 12 May 2025; doi:10.1038/s41563-025-02242-6

The orbital-to-spin moments ratio for individual atomic planes of an iron crystal is studied, revealing local variations at subatomic scales.

Nature Energy, Published online: 12 May 2025; doi:10.1038/s41560-025-01776-y

The increasing energy density and size requirements for batteries demand better safety technologies, but size limits and high costs hinder effective testing. Now, accelerated rate calorimetry tests on small batteries with an optimal thermal runaway factor enable rapid screening and provide early-stage feedback for improving safety features.

Title to be confirmed

Abstract not available

• Speaker: Arduin Findeis (University of Cambridge)
• Tuesday 17 June 2025, 13:00-14:00
• Venue: Lecture Theatre 2, Computer Laboratory, William Gates Building.
• Series: Artificial Intelligence Research Group Talks (Computer Laboratory); organiser: Mateja Jamnik.

Add to your calendar or Include in your list

Instrumentation and Iteration in Massively Online Live Service Gaming

Massively multiplayer online games (MMOs) bring together millions of players in shared online worlds. The games industry eclipsed film by revenue in the mid ‘90s. These huge financial opportunities introduced incentives both for developers to instrument their games to better understand player behaviour, and for malicious parties to exploit these online games for their own personal gain. Jagex launched RuneScape, their flagship MMO , in 2001. RuneScape players inhabit a virtual world in which they socialise, work, trade, and fight—a microcosm of human behaviour. Through near-weekly updates adding new features and game content, it has cemented itself as a long-lived and immensely popular game. This talk describes the data collection and analysis that maintains the feedback loop between these regular updates and the player-base, allowing game developers to delight their users, as well as the tools for identifying bad in-game actors. We will describe how modern games are instrumented, and then cover three main topics: preventing cheating, advertising, and in-game economics. Finally, we will discuss how to model ‘emergent behaviours’ as game players learn more about the game world they inhabit. No prior knowledge of the games industry is expected.

• Speaker: Mark Kerr, Jagex
• Thursday 15 May 2025, 14:00-15:00
• Venue: JDB Seminar Room, Department of Engineering and online (Zoom).
• Series: CUED Control Group Seminars; organiser: Fulvio Forni.

Add to your calendar or Include in your list

Harnessing the Power, Managing the Risk: The Future of AI, Data, and Cybersecurity

AI has the potential to transform learning, creativity, and productivity. It represents a profound platform shift in technology – like the internet, or the shift to mobile. As with any transformational shift, AI will be (and already is) used for good and for malicious purposes. This talk will cover some of the practical examples of threats we see and risks we are preparing for, but also the incredible opportunities available to us to use AI to change the game in cybersecurity and ultimately make the world safer.

Evan Kotsovinos is the Vice President of the Privacy, Safety and Security team at Google, which is the central engineering function that builds and scales the foundational technology that keeps billions of people safe online. His team is focused on cybersecurity threat detection, analysis and counterabuse, building advanced AI/ML security technologies, and protecting privacy, identity and data. He continues to work on planet-scale solutions that protect the security and privacy of Google’s billions of users around the world.

Evan was previously Global Head of Infrastructure at American Express, responsible for the company’s data centers, networks, compute, storage, hybrid cloud, data infrastructure, and SRE teams. Prior to that he served as Asia CIO at Morgan Stanley, where he managed all technology services and resources in the region.

Evan began his career as a Senior Research Scientist with Deutsche Telekom Laboratories and is a recognized leader in cloud computing, having led the team that developed one of the first cloud computing systems in the early 2000s at the University of Cambridge. He holds a Doctorate in Computer Science from the University of Cambridge and a Master’s in Finance from London Business School.

• Speaker: Evan Kotsovinos, Vice President Privacy, Safety, Security, Google
• Tuesday 13 May 2025, 14:00-15:00
• Venue: Webinar & LT2, Computer Laboratory, William Gates Building..
• Series: Computer Laboratory Security Seminar; organiser: Anna Talas.

Add to your calendar or Include in your list

Title to be confirmed

Abstract not available

• Speaker: Aitor Iribar Lopez, ETH Zurich
• Wednesday 21 May 2025, 14:15-15:15
• Venue: CMS MR13.
• Series: Algebraic Geometry Seminar; organiser: Dhruv Ranganathan.

Add to your calendar or Include in your list

Energy Environ. Sci., 2025, Accepted Manuscript
DOI: 10.1039/D5EE01604E, PaperXiaoqi Yu, Jintao Zhu, Lin Xie, Haotian Hu, Tongqiang Liu, Pengfei Ding, Xueliang Yu, Jinfeng Ge, Chengcheng Han, Wei Song, Ziyi Ge
The ternary strategy is an effective approach to enhancing luminescent properties and mitigating non-radiative recombination. However, achieving a simultaneous reduction in non-radiative recombination without sacrificing charge generation in organic solar...
The content of this RSS Feed (c) The Royal Society of Chemistry

Superposition in GNNs

Existing mechanistic interpretability efforts seem severely threatened by superposition, an effect in which a neural network represents more “features” than it has neurons. Previous papers have used toy models with MLP architectures to study both representational superposition (caused by passing higher dimensional data through a lower dimensional hidden layer) and computation in superposition. Here, for the first time, toy models are used to study how superposition arises in graph neural networks (GNNs). We demonstrate, (i) that superposition in GNNs can arise similarly to MLPs through compression, yet different aggregation functions distinctly impact this phenomenon, with max pooling notably discouraging superposition; (ii) we find that the inherent topology of graphs enables the construction of toy models where superposition arises even in the absence of compression and we discuss the algorithms the model finds to do this; (iii) we identify that graph isomorphism networks (GIN) can lead to the emergence of superposition within a lower-dimensional subspace of a larger embedding, suggesting that superposition inadvertently creates metastable minima; and (iv) we look at how superposition emerges in real life binary classification datasets.

• Speaker: Lukas Pertl
• Monday 12 May 2025, 17:00-17:45
• Venue: Lecture Theatre 2, Computer Laboratory, William Gates Building.
• Series: Foundation AI; organiser: Pietro Lio.

Add to your calendar or Include in your list