ICRA 2026 highlighted a decisive shift – future systems must be designed not only for autonomy and performance, but also for energy efficiency, material responsibility and sustainable lifecycles.
For decades, resilience in power systems has been framed as an event-based outcome: the response of a system to an individual extreme weather event. Climate change makes that framing untenable. As extreme weather evolves and its uncertainties compound across decades, resilience measured event by event becomes variable and difficult to interpret, and offers no reliable basis for long-term planning. In this Review we argue that resilience must be redefined as a long-horizon, event-independent property that governs power system performance across the range of futures a system may encounter. We formulate this concept of long-horizon resilience, set it against the event-based paradigm it replaces and develop a unified framework for studying, gauging and reinforcing it. We then set out the scientific challenges this shift raises and a research agenda to address them. This reframing provides a conceptual basis for the climate adaptation of future power systems. Power system resilience has conventionally been measured against individual extreme weather events. As climate change moves that weather beyond the historical record, this Review argues for resilience redefined over long horizons and sets out a framework and research agenda for studying it.
Neuromorphic engineering continuously draws inspiration from features of the nervous system, such as synaptic computation, spiking communication, dendritic processing and neuroplasticity. Its growing interest led to the diversification of neuromorphic architectures, complicating cross-platform comparison due to the lack of unified metrics and figures of merit. In this Review, we identify key metrics for neuromorphic devices and architectures, assess their potential applications and create meaningful comparison. Different levels of biological computation are identified, and their artificial counterparts are evaluated by their emulated function. The key metrics and specific strengths of different architectures are then presented via bioplausibility, energy efficiency or scalability, providing a unified framework for evaluating emerging neuromorphic technologies and guiding their optimization for targeted applications. By establishing common evaluation criteria, this Review supports the design of next-generation neuromorphic systems and accelerates progress towards real-world, low-power computing solutions. This Review posits biological relevance as key metrics for evaluating neuromorphic devices. These metrics provide a structured framework to navigate the growing diversity of devices and platforms, guiding the optimization of design trade-offs according to the targeted application.
Large-scale artificial intelligence (AI) workloads expose the structural limits of network-based distributed systems relying on software to coordinate separate devices. As AI models scale across hundreds or thousands of accelerators, latency spread increasingly limits system-wide execution. In this Review, we introduce a one-chip-like standpoint for understanding how datacenter infrastructures are evolving to support AI workloads. Tightly coupled AI computation over a distributed environment defines the requirements for extending chip-level execution rules across physically separated resources. Compute express link (CXL) serves as a basis for chip-like execution by enabling memory pooling and sharing under fabric. However, realizing predictable execution with CXL-based fabric requires more architectural considerations than standard-defined functionalities. It requires hardware mechanisms that control the sources of latency variability along traversal paths and hierarchical deployment structures that organize resources in chip-like arrangements. Looking forward, emerging interconnect technologies, including optical links and CXL-over-optics, can further extend one-chip-like execution across broader physical domains. This Review highlights how one-chip-like datacenter extends chip-level semantics to rack scale by combining compute express link fabrics with hardware mechanisms, thereby stabilizing latency variability along traversal paths to enable large-scale AI systems to operate cohesively beyond electrical limits.
The brain remains among the hardest systems in biology to model. Whole-brain simulations, neuromorphic systems and predictive surrogates are among the main approaches, but each captures isolated aspects of brain function rather than a specific living brain that can be updated over time. Digital twin brains (DTBs) reframe the goal as individualized, data-coupled models for which structure and dynamics are constrained by subject-specific measurements. In this Review, we argue that the fidelity a DTB can reach is set by the resolution, completeness and updatability of those measurements, not by neuron count — a measurement-defined emulation scale. On this scale, today’s DTBs are best understood as partial, simulation-based counterparts: they reconstruct structure and reproduce dynamics, yet persistent updating, closed-loop interaction and embodiment remain long-term goals. Building DTBs at biological scale will depend as much on data integration, validation and governance as on raw computation. DTBs are therefore emerging as instruments for health care, discovery and brain-inspired artificial intelligence. Digital twin brains aim to mirror a living brain in software, updating as new measurements arrive. This Review argues that a model’s accuracy is limited by what we can measure, not by sheer scale, and charts a path from reconstructing brain structure to adaptive, interactive models.
Solid-state batteries (SSBs) promise higher energy density and intrinsic safety than lithium-ion batteries, but field-scale deployment is blocked by life cycle-wide barriers that materials-centric research alone cannot resolve. Interfacial degradation, poor real-world reliability and complex end-of-life recovery differ across oxide, sulfide and polymer chemistries. These barriers call for a system-level rather than a materials-level response. Electrical engineering supplies the missing toolkit: multi-physics sensing, machine-learning analytics and adaptive control coupled to the life cycle digital twin. These capabilities form a continuous information loop across the SSB’s life cycle. The life cycle intelligence framework treats the cell as a cyber-physical system rather than a sealed storage device. In this Review, we examine the framework’s components, identify the engineering and institutional challenges that hinder scale-up, and outline a three-phase roadmap towards commercial deployment. Materials advances have brought solid-state batteries close to commercial deployment, but the systemic barriers that remain — interfacial degradation, real-world reliability and end-of-life recycling — call for electrical engineering, not materials chemistry alone. This Review shows how sensing, machine learning and digital twins reshape the cell across its life cycle.
Synergistic integration of a resonant RF antenna and a microring electro-optic modulator breaks the size–efficiency trade-off in monolithic photonic RF receivers, supporting diverse communications, radar and electronic warfare applications.
Xiwen Gong, assistant professor of chemical engineering of the University of Michigan, shares her career path and vision as the early-career winner of the 2026 Sony Women in Technology Award with Nature.
Power can be harvested in space today. The real challenge is building, operating and certifying a complete energy system in orbit — under extreme conditions, with no one nearby to repair it. Whether space joins the world’s energy supply will depend on that challenge, not just on generation.
Functional ultrasound imaging (fUSI) offers a non-invasive, high-resolution path to brain–computer interfacing (BCI) but moving from laboratory prototypes to clinical use demands solutions to technical, clinical and scalability challenges. This Comment examines those challenges and outlines directions for bringing fUSI-BCIs out of the lab.
Future cyber–physical systems face a gap between surging data traffic and finite resources under stringent latency and reliability demands. Throughput-driven and latency-driven communication systems assign the same scheduling priority to all packets regardless of their relevance to the downstream task, which can undermine decision quality. Conversely, goal-oriented communications transmit only goal-relevant information, protecting control systems from destabilization due to indiscriminate data flooding. In this Review, we analyse how goal-oriented communications reshape cyber–physical system design across four tiers: networked sensing and control, decision-making, distributed intelligence and multimodal intelligence. The analysis reveals that task-level metrics, such as control error, inference accuracy and decision utility, rather than bit-level communication metrics, are key to ensuring reliable sensing, control and decision-making. We reframe cyber–physical communication from bit-level optimization toward task-oriented design, where communication decisions are driven by downstream task requirements. Finally, we posit a framework that will enable engineers and researchers to build intelligent cyber–physical systems. This framework will power smarter grids, safer autonomous vehicles and more adaptive industrial automation. This Review shows that transmitting targeted, goal-oriented bits rather than more data reshapes communication–performance trade-offs across four tiers of cyber-physical complexity, from control stability to multimodal intelligence.
Topolectrical circuits are electrical networks that encode topological band theory. Their physical behaviour is set by how the components are wired together, not by where they are placed. This wiring-centred design delivers defect tolerance, directional signal flow without bulky magnetic parts, and amplified response at boundaries. Conventional electronics obtain these only by adding isolators, feedback or calibration after the circuit is laid out. In this Review, we identify which capabilities are engineering-ready and which remain confined to the laboratory. We cover the underlying physics, the resulting sensors and engineering applications, and cross-disciplinary uses in quantum simulation, artificial intelligence-assisted design and curved-space geometries. We then examine how co-integration with memristive devices makes the platform adaptive and history-dependent. We close with the hardware routes towards deployable electronics, from monolithic chips to flexible substrates and body-worn systems. Topolectrical circuits route signals one way without magnets and keep working when components fail. These capabilities come from how the components are wired together, not from added parts. This Review covers their physics, applications across sensing, wireless power and communications, and the path towards integrated, flexible and adaptive electronics.
Zhen Xu, Li Ka Shing Endowed Professor of Biomedical Engineering and professor of Radiology and Neurosurgery at the University of Michigan, speaks to Nature Reviews Electrical Engineering about the journey that led to her success in inventing and advancing histotripsy to help patients, and shares her vision as the mid-career winner of the 2026 Sony Women in Technology Award with Nature. Zhen Xu reflects on the journey that led to her success in inventing and advancing histotripsy to help patients, and shares her vision as the mid-career winner of the 2026 Sony Women in Technology Award with Nature.
Stark effect-induced bandgap tuning in black phosphorus (BP) provides precise control over carrier density and switching behaviour, enabling adjustable amplifier gain and bandwidth, facilitating compact binary and ternary logic implementations, and advancing BP transistor arrays for next-generation circuit applications.
Alkaline water electrolysis supplies most installed water-electrolyser capacity, yet the technology and its perceived limits have not fundamentally changed in a century. Rather than inherent chemistry, we argue that these limits are the consequence of traditional operating conditions at near-atmospheric pressure, low current density, and steady state. The performance gap with proton exchange membrane systems persists across a coupled hierarchy of losses: kinetic and ohmic losses at the electrodes and separator in concentrated KOH; shunt and reverse currents along the manifolds of bipolar stacks; and power-conversion, compression and thermal-management losses at the plant level. In this Review, we outline how closing this gap calls for coordinated advances in electrode and separator materials, stack architecture, power electronics and plant-level integration, evaluated under realistic industrial conditions of concentrated alkali, elevated temperature and pressure, and dynamic loads. By pursuing these advances, we can rebuild alkaline electrolysis from first principles into a flexible workhorse for low-carbon hydrogen production. Alkaline water electrolysis dominates global water-electrolyser capacity, but its architecture has remained largely unchanged for a century. This Review examines how losses at the cell, stack and system levels conventionally limit efficiency, and highlights how rethinking these constraints can unlock substantial gains for industrial hydrogen production.
Peripheral nerve interfaces provide a bidirectional electrical link between implanted electronics and the peripheral nervous system, supporting therapies for sensory restoration, motor control and chronic disease management. Their long-term performance depends less on how well signals are transmitted at implantation and more on how the electrode–tissue relationship changes, as impedance, threshold, signal fidelity and selectivity drift over months to years. This Review describes these changes through dynamic biophysical coupling, in which mechanical, geometric, electrochemical, biochemical and biological variables co-evolve at the electrode–tissue interface and together determine how the device functions. In this view, keeping coupling within a useful operating window is not a one-time goal set at implantation but a state that must be maintained through coordinated design across materials, device structures, anatomy-specific implantation and system design. We then consider how coupling is measured, modelled and evaluated over the long term, and identify the shared metrics that would make next-generation peripheral nerve interfaces predictable and clinically scalable. Peripheral nerve interfaces typically fail not from acute signal loss but from slow drift in the electrode–tissue relationship. This Review reframes that drift through dynamic biophysical coupling, a unifying view of how materials, structures, anatomy and feedback control together keep interfaces stable, selective and predictable across years of implantation.
A study in Nature uses deep learning to passively monitor resting heart rates via front-facing smartphone cameras, yielding highly accurate cardiovascular tracking across all skin tones without the need for wearables.
Two-dimensional (2D) semiconductors have emerged as strong candidates for beyond-silicon electronics. Their integration into complementary metal–oxide–semiconductor (CMOS) technologies, however, requires both n- and p-type field-effect transistors with competitive performance, scalability and reliability. Although n-type devices remain more mature, p-type 2D field-effect transistors are beginning to support complementary circuits and monolithic three-dimensional integration. This Review examines the physical and technological factors governing CMOS-grade parity in p-type 2D field-effect transistors. Emphasis is placed on contact engineering, including Schottky barrier control, contact scaling and edge-contact strategies compatible with advanced technology nodes. The Review also considers top-gate integration, threshold-voltage engineering for circuit design, and reliability evaluation for scaled devices. Channel growth, selective doping and defect reduction are discussed as essential materials and process requirements for manufacturable technologies. Finally, reporting protocols are outlined to enable consistent benchmarking across studies and to guide scalable complementary and vertically integrated 2D electronics. Moving beyond silicon electronics with n- and p-type field-effect transistors requires competitive performance, scalability and reliability. This Review examines how the gap between p-type two-dimensional transistors and their n-type counterparts for complementary metal–oxide–semiconductor (CMOS)-grade complementary and three-dimensional integrated electronics is narrowing.
