At the heart of this innovation is the concept of topological robustness inherent in skyrmions, which are swirling configurations originally conceived in magnetic materials but here realized within plasmonic electromagnetic fields. These configurations are defined by their nontrivial topology, rendering them stable against continuous deformations and perturbations. This stability is particularly vital for information carriers in noisy, turbulent environments, where conventional signals degrade rapidly. The platform introduced by the research team leverages this property to create electromagnetic fields with encoded topological information that can endure harsh channel conditions without loss of integrity.

The key challenge addressed concerns the programmability and tunability of skyrmions. Conventional methods typically produce fixed skyrmion states limited by the geometrical constraints of the nanostructures or strict material properties. The presented solution overcomes this by synthesizing what the authors term “harmonic skyrmions” in the temporal domain through ultrafast coding techniques. This dynamic approach enables on-demand generation of skyrmions with various topological types, such as Néel-type skyrmions and merons, which are half-skyrmions characterized by distinct magneto-optical patterns. The versatility thus achieved paves the way for intricate topological “alphabets” manipulable in real time.

The temporal encoding strategy involves modulating the plasmonic field’s phase and amplitude at ultrafast timescales, effectively creating skyrmion waveforms that vary harmonically with time. This not only permits rapid switching between different topologies but also expands the bandwidth and channel capacity of communication systems employing skyrmion carriers. The capacity to encode information in such multidimensional topological structures fundamentally redefines the architecture of robust wireless signaling, providing resilience against interference and environmental disruptions that plague existing technologies.

One particularly compelling demonstration is the application of programmable skyrmions in multi-channel wireless communication experiments. The research team showed that distinct topological modes could serve as independent parallel channels, each capable of transmitting data with high fidelity. This multiplexing capability, combined with the robustness of the skyrmions themselves, offers a promising route to boost data throughput in challenging scenarios such as turbulent atmospheric or underwater conditions. Such resilience is crucial for next-generation communication networks where reliability is paramount.

Beyond communication, the platform’s adaptability extends to intelligent sensing domains, where topological patterns can enhance the recognition and classification of complex objects. The integration of a convolutional neural network (CNN) with the skyrmion-generating system enabled high-accuracy recognition of twenty different animal figurines based on their interaction with encoded skyrmion fields. This synergy between topological photonics and artificial intelligence signifies a powerful paradigm for advanced sensing systems, marrying physical robustness with computational intelligence.

The fundamental physics underlying the programmable platform combines sophisticated plasmonic engineering with ultrafast optics. Surface plasmon polaritons (SPPs), which are collective electron oscillations coupled to electromagnetic fields at metal-dielectric interfaces, serve as the playground for skyrmion creation. By precisely tailoring the spatial and temporal distribution of these SPPs, the researchers synthesized complex field configurations exhibiting nontrivial topological charges and textures. This meticulous control required innovations in both the design of metasurfaces—a class of engineered nanostructured materials—and the ultrafast modulation techniques implemented through tailored laser pulse sequences.

Furthermore, the exploration of Néel-type skyrmions and merons within plasmonic environments uncovers new fundamental insights into how topological phases evolve in photonic systems. The Néel-type skyrmions, characterized by a radially symmetric spin texture, and the merons, topological modes with half-integer winding numbers, represent distinct classes of quasiparticles whose electromagnetic analogues can be selectively programmed. This richness adds to the taxonomy of topological photonics and opens avenues for exploring the interplay between topology, time-dependent fields, and material responses in nanoscale light-matter interaction regimes.

The implications of programmable plasmonic skyrmions also resonate with the broader pursuit of topological photonic computing and communication. By encoding information robustly in the spatial and temporal topology of light fields, this approach circumvents key limitations imposed by noise, scattering, and material imperfections. This enhancement in robustness is not merely incremental but transformative, as it introduces fundamentally new degrees of freedom for designing resilient systems at optical frequencies. It also aligns with trends towards integrated photonic chips capable of operating under variable and extreme environmental conditions, intensifying the push towards real-world deployments.

Crucially, the research demonstrates programmability at ultrafast timescales, effectively bridging the gap between static topological states and dynamic information processing. The temporal dimension of skyrmion encoding enables real-time modulation, which is essential for high-speed communication and adaptive sensing. This feature could allow devices to dynamically switch operational modes, respond to environmental changes, or encode multiplexed information streams without requiring physical reconfiguration of the underlying materials.

Moreover, the integration of machine learning with topological photonic platforms marks a milestone in active, intelligent photonic systems. The convolutional neural network employed interprets subtle variations in the skyrmion-encoded signals, achieving current-state-of-the-art recognition accuracy for diverse objects. This confluence suggests a future where topological photonic devices serve as front-end sensors feeding directly into AI processors, creating compact, robust, and autonomous sensing modules for applications ranging from environmental monitoring to biomedical diagnostics.

On the technical front, the stable generation of skyrmions in plasmonic fields involves overcoming intrinsic dissipative losses and maintaining coherence over relevant timescales and spatial extents. The authors tackled these challenges by optimizing metasurface designs and carefully calibrating the ultrafast excitation protocols, achieving stable skyrmion lifetimes sufficient for communication and sensing tasks. Such engineering finesse points towards practical device architectures that can strike a balance between complexity and scalability.

The potential real-world impacts of this technology extend broadly. The ability to encode information topologically with programmable electromagnetic skyrmions could revolutionize wireless communication by facilitating multichannel data transmission with exceptional noise immunity. Similarly, in sensing applications, the enhanced recognition capabilities enabled by topological encoding and AI processing could find uses in robotics, surveillance, and environmental science, where robustness and precision are critical. Furthermore, the underlying methodology could catalyze innovations across photonic computing, metrology, and secure information transfer.

Looking forward, this platform offers a tantalizing glimpse into future photonic technologies that harness the deep principles of topology and ultrafast dynamics to transcend current limitations. As demand grows for devices that can sustain high performance in unpredictable or harsh environments—such as space exploration, underwater communication, or disaster response—the robustness and programmability of plasmonic skyrmions provide an elegant and effective solution. The seamless integration of topological design, ultrafast laser control, nanofabrication, and machine learning portends a fertile ground for interdisciplinary advances.

The research thus represents a landmark achievement in the field of topological photonics and plasmonics, pushing the boundaries of what electromagnetic quasiparticles can achieve in practical applications. By transforming the theoretical concept of skyrmions into a highly controllable and multifunctional platform, the work stands to reshape the landscape of optical communication and sensing. It underscores the power of combining advanced materials, ultrafast technologies, and computational intelligence to address longstanding challenges in information science and photonics.

This discovery also invites further exploration into the fundamental interplay between topology and dynamics, hinting at new classes of topological excitations that may be realized through temporal modulation schemes. Such possibilities could unlock exotic functionalities, including topological protection amid non-equilibrium conditions, dynamic switching between topological phases, and topological quantum-like effects in classical photonics. The insights gained here may thus inspire a rich lineage of research bridging physics, engineering, and computing.

In sum, the programmable platform for plasmonic skyrmions developed by Chen, Li, Shen, and colleagues marks a transformative step towards harnessing the full potential of topological photonics for next-generation technology. Combining nanoscale engineering, ultrafast modulation, and artificial intelligence, it offers a robust, tunable, and intelligent framework for encoding and decoding information in ways that promise to revolutionize wireless communication and sensing. The implications are profound, positioning this approach at the forefront of future photonic innovations set to impact numerous scientific and technological domains.

Subject of Research: Plasmonic skyrmions and their applications in wireless communication and intelligent sensing.

Article Title: Programmable skyrmions for communication and sensing.

Article References:
Chen, L., Li, X.Y., Shen, Y. et al. Programmable skyrmions for communication and sensing. Nat Electron (2026). https://doi.org/10.1038/s41928-026-01611-6

DOI: https://doi.org/10.1038/s41928-026-01611-6

Optical processors have long been recognized for their potential to dramatically accelerate information processing speeds and reduce energy consumption compared to their electronic counterparts. However, their practical deployment has been hindered by difficulties in training these systems to perform complex tasks. Conventional methods demanded accurate forward models of the optical system to dictate parameter adjustments, a requirement that proved infeasible in many real-world scenarios due to system imperfections and environmental variables.

The research conducted by Li, Chen, Gong, and colleagues introduces an innovative solution by removing the dependency on explicit models. Their approach exploits direct interaction with the physical optical system through a reinforcement learning framework that continuously updates system parameters in situ, refining performance iteratively without prior knowledge of the underlying physics. This represents a significant methodological shift towards adaptive optical computing, where the processor learns autonomously from real-world feedback.

At the core of this advancement lies proximal policy optimization (PPO), a state-of-the-art reinforcement learning algorithm known for its stability and efficiency in continuous control tasks. By integrating PPO with optical hardware, the researchers enabled the system to adjust its internal configuration—such as phase modulations across spatial light modulators—to achieve desired computational objectives. This tight coupling between learning algorithm and physical system underlines an essential synergy for future adaptive photonic technologies.

The experimental setup involved a sophisticated arrangement where the optical processor’s parameters were iteratively tuned based on observed output performance. Rather than relying on a pre-calibrated mathematical model, the reinforcement learning agent received environmental feedback, gauged the quality of its outputs through a reward signal, and made incremental policy updates. This closed-loop paradigm facilitated rapid convergence towards optimized performance metrics, demonstrating robustness in the face of noise and parameter drift.

One of the remarkable outcomes of the study was the processor’s ability to solve complex computational problems such as image classification directly in the optical domain. This task, traditionally conducted by electronic neural networks, was effectively accomplished without requiring an explicit model of the optical transformations involved, showcasing the system’s practical viability. Such achievements underscore the potential for deploying compact, low-power optical processors in applications ranging from edge computing to autonomous systems.

The model-free framework also shines in its adaptability to varying environmental conditions. Optical systems often suffer from fluctuations due to temperature variations, alignment shifts, and component aging. By continuously learning from in situ feedback, the proposed method inherently compensates for such perturbations, maintaining high performance without manual recalibration. This self-tuning capability addresses a major obstacle in deploying photonic processors outside controlled laboratory settings.

Moreover, the approach opens new frontiers for complex optical tasks that defy accurate modeling, including those with non-linear or chaotic characteristics. By embedding intelligence at the hardware level, optical processors can extend their functional repertoire beyond pre-defined algorithms, embracing a level of autonomy previously unseen in photonics. This shift hints at an exciting convergence between optical hardware and artificial intelligence paradigms.

The team’s integration of deep reinforcement learning represents a powerful fusion of modern AI techniques with physical layer computation. Unlike traditional software-based AI, this method exploits the inherent parallelism and speed of light-based processing, potentially achieving orders of magnitude faster inference times while minimizing energy consumption. This dual advantage positions optical processors as compelling candidates for future high-throughput data centers and real-time decision-making platforms.

Despite these promising results, challenges remain in scaling the system for broader commercial adoption. Current implementations are constrained by device resolution, speed of modulation elements, and the complexity of reward function design. However, ongoing advancements in spatial light modulators, photonic integrated circuits, and algorithmic efficiency are expected to bridge these gaps in the coming years, accelerating the maturation of model-free optical computing.

Industry experts anticipate that such adaptive optical processors will revolutionize sectors requiring rapid data analysis and low-latency responses, including telecommunications, autonomous vehicles, and medical imaging. By embedding learning directly within hardware, these devices herald a paradigm shift from static, hardcoded processors to dynamically evolving computation platforms capable of autonomous problem-solving.

Furthermore, the research highlights the broader trend of marrying hardware advances with machine learning to overcome fundamental barriers in computational sciences. As devices become smarter and more context-aware, the boundary between physical systems and algorithmic intelligence continues to blur, giving rise to multifunctional platforms that can self-optimize, self-heal, and adapt in real time to their operational environment.

Another critical consequence of this study is its potential impact on the design of neuromorphic systems, which aim to mimic biological neural architectures. The use of in situ reinforcement learning within optical processors moves such technologies closer to the goal of creating brain-inspired computing machines with unmatched efficiency and agility, enabling applications previously relegated to theoretical exploration.

In sum, the introduction of model-free optical processors armed with proximal policy optimization via in situ reinforcement learning marks a crucial step towards truly intelligent photonic computation. This work not only provides a practical pathway to surmount the limitations of model-dependent training but also unlocks a new dimension of adaptability and performance for optical technologies.

As this line of research progresses, one can envision a future where optical processors autonomously learn from and react to their environment, continuously refining their operation without human intervention. Such capabilities could transform the very fabric of computational hardware, leading to smarter, faster, and more energy-efficient machines across diverse scientific and industrial domains.

This pioneering research by Li and colleagues underscores the transformative potential of integrating advanced reinforcement learning algorithms directly within optical hardware, signaling the dawn of a new era in model-free, self-optimizing computation. As the field advances, the fusion of photonics and AI promises to catalyze revolutionary shifts in technology, fundamentally altering how we compute, perceive, and interact with information.

Subject of Research: Model-free optical processors employing in situ reinforcement learning with proximal policy optimization for adaptive photonic computation.

Article Title: Model-free optical processors using in situ reinforcement learning with proximal policy optimization.

Article References:
Li, Y., Chen, S., Gong, T. et al. Model-free optical processors using in situ reinforcement learning with proximal policy optimization. Light Sci Appl 15, 32 (2026). https://doi.org/10.1038/s41377-025-02148-7

Image Credits: AI Generated

DOI: 10.1038/s41377-025-02148-7

Keywords: Optical processors, in situ reinforcement learning, proximal policy optimization, model-free computation, photonic computing, adaptive systems, deep reinforcement learning, spatial light modulators, optical neural networks, autonomous hardware learning.

Expanding beyond data storage, memristors present transformative opportunities for real-time, energy-conscious data stream processing. By integrating processing functions directly into memory elements, these devices enable ultrafast computations with lower costs and reduced power demands. Flexibly embedded within artificial intelligence architectures, memristor-based neuromorphic processors have the potential to empower neural networks to learn dynamically on compact platforms, including handheld mobile devices. Researchers worldwide continue to innovate diverse memristor configurations aiming to support advanced applications such as machine vision, acoustic and speech recognition systems, and sophisticated biointerfacing technologies.

The performance and reliability of memristors are heavily dependent on intrinsic material characteristics, contact engineering, and device miniaturization. Here, lead halide perovskites have garnered considerable attention as a promising semiconductor class owing to their versatile applications ranging from photovoltaic cells and photodiodes to light-emitting diodes and sensors. These materials exhibit excellent semiconductor behavior suitable for use in memristors, both with and without the formation of conductive filaments, which are crucial for switching resistance states. The ability of halide perovskites to endure environmental fluctuations while maintaining their electronic properties underscores their potential for durable, efficient memristive devices.

Notwithstanding significant progress, understanding and achieving stable electrical characteristics in perovskite memristors under prolonged voltage stress remains a formidable challenge. The inherent formation of conductive filaments, coupled with metal-ion penetration from electrodes into the perovskite layer under ambient humidity and temperature conditions, threatens device longevity and operational stability. Key hurdles involve enhancing cycling endurance to optimize switching performance, minimizing power consumption, and achieving device miniaturization without compromising function. Thorough investigation into defect dynamics and ion migration within the perovskite lattice is necessary to unravel the complex switching mechanisms underpinning device behavior.

In a groundbreaking study led by Prof. Sergey Makarov’s research group at ITMO University and Harbin Engineering University, scientists have engineered one of the smallest memristors based on a cesium lead bromide (CsPbBr3) perovskite nanocube, recognized for its chemical stability compared to other lead-halide variants. These monocrystalline nanocubes were carefully synthesized directly onto indium tin oxide (ITO) substrates, sandwiched between chemically inert contacts—ITO at the bottom and boron-doped diamond (BDD) at the top. This unique configuration allows precise control and coupling of the perovskite crystal with the electrode, enabling direct measurement of current-voltage (J-V) characteristics over thousands of cycles, confirming remarkable stability.

Critical insights emerged from exploring how the nanocube’s thickness influenced resistive switching and power usage. The team achieved an unprecedented low power consumption for switching state, registering between 70 and 80 nanowatts for crystals sized between 130 to 150 nanometers. This ultra-low power threshold is among the smallest values reported for memristors, enhancing their viability for highly energy-efficient electronics. Furthermore, the distinct current ratio exceeding 10^5 between the binary ON and OFF states promises clear computational readability and reliability, vital for practical memory and logic applications.

To decode the memristive behavior of their devices, researchers employed an advanced drift-diffusion model that uniquely incorporated dual mobile ionic species—both anions and cations—within the perovskite lattice, a marked departure from traditional semiconductor models focusing solely on electron and hole transport. These ions, possessing slower mobility relative to electronic carriers, dynamically modulate the internal electrical potential, thereby impacting the current flow dependent on the applied voltage history. This ionic drift introduces a temporal memory effect, locking the device’s conduction state based on previously applied voltages.

The study elucidated that memory functionality arises not only from ion redistribution but also from critical ion-contact interactions at the interfaces. These interactions notably decrease energy barrier heights, fostering enhanced electronic tunneling across the perovskite-contact interface. This synergistic interplay between the mobile ions and the chemical contacts is essential to supporting the observed non-volatile resistive switching, setting a foundation for robust device design strategies that optimize both ionic and electronic transport mechanisms.

The work’s leading authors bring diverse expertise supporting this breakthrough. Abolfazl Mahmoodpoor, currently a PhD candidate deeply engaged in numerical modeling of ion migration phenomena in perovskite semiconductors, contributed significant theoretical and simulation insights. Prokhor Alekseev, with a strong background in scanning probe microscopy and semiconductor physics, facilitated precise characterization of the nanoscale electrical properties. Meanwhile, Ksenia Gasnikova has recently embraced advanced microscopy techniques to probe the intricate nanostructures. Prof. Sergey Makarov’s extensive background in nanophotonics and solution-processed devices, combined with Prof. Aleksandra Furasova’s leadership in perovskite device fabrication and experimental analysis, orchestrated this comprehensive investigation.

Taken together, these advancements affirm the promise of halide perovskite-based nanomemristors as next-generation building blocks for ultralow-power neuromorphic computation and memory technologies. The unique chemical robustness and tunable ionic-electronic transport dynamics foster new paradigms in memory retention and device scalability. This pioneering research opens pathways for practical memristive devices that can seamlessly integrate into portable electronics, AI hardware, and biointerfaces, propelling the boundaries of sustainable and intelligent computing.

For further exploration and technical depth, the full research article is accessible via the journal “Opto-Electronic Advances,” providing an exhaustive account of experimental methods, theoretical modeling, and implications for future semiconductor device innovations.

Subject of Research: Development and characterization of ultralow-power halide perovskite nanomemristors based on CsPbBr3 monocrystalline nanocubes interfaced with chemically inert electrodes.

Article Title: Halide perovskite volatile unipolar Nanomemristor

News Publication Date: 14-Oct-2025

Web References: https://www.oejournal.org/oea/article/doi/10.29026/oea.2025.250110

Image Credits: Abolfazl Mahmoodpoor, Aleksandra Furasova

Keywords: memristor, halide perovskite, CsPbBr3, nanocube, neuromorphic computing, ionic drift, resistive switching, low power consumption, boron-doped diamond electrode, indium tin oxide, drift-diffusion modeling, energy-efficient memory