Explore the vast range of titles available.

Next-generation structural materials are expected to be lightweight, high-strength and tough composites with embedded functionalities to sense, adapt, self-repair, morph and restore. This Review highlights recent developments and concepts in bioinspired nanocomposites, emphasizing tailoring of the architecture, interphases and confinement to achieve dynamic and synergetic responses. We highlight cornerstone examples from natural materials with unique mechanical property combinations based on relatively simple building blocks produced in aqueous environments under ambient conditions. A particular focus is on structural hierarchies across multiple length scales to achieve multifunctionality and robustness. We further discuss recent advances, trends and emerging opportunities for combining biological and synthetic components, state-of-the-art characterization and modelling approaches to assess the physical principles underlying nature-inspired design and mechanical responses at multiple length scales. These multidisciplinary approaches promote the synergetic enhancement of individual materials properties and an improved predictive and prescriptive design of the next era of structural materials at multilength scales for a wide range of applications.This Review discusses recent progress in bioinspired nanocomposite design, emphasizing the role of hierarchical structuring at distinct length scales to create multifunctional, lightweight and robust structural materials for diverse technological applications.

Natural load-bearing materials such as tendons have a high water content of about 70 per cent but are still strong and tough, even when used for over one million cycles per year, owing to the hierarchical assembly of anisotropic structures across multiple length scales 1 . Synthetic hydrogels have been created using methods such as electro-spinning 2 , extrusion 3 , compositing 4 , 5 , freeze-casting 6 , 7 , self-assembly 8 and mechanical stretching 9 , 10 for improved mechanical performance. However, in contrast to tendons, many hydrogels with the same high water content do not show high strength, toughness or fatigue resistance. Here we present a strategy to produce a multi-length-scale hierarchical hydrogel architecture using a freezing-assisted salting-out treatment. The produced poly(vinyl alcohol) hydrogels are highly anisotropic, comprising micrometre-scale honeycomb-like pore walls, which in turn comprise interconnected nanofibril meshes. These hydrogels have a water content of 70–95 per cent and properties that compare favourably to those of other tough hydrogels and even natural tendons; for example, an ultimate stress of 23.5 ± 2.7 megapascals, strain levels of 2,900 ± 450 per cent, toughness of 210 ± 13 megajoules per cubic metre, fracture energy of 170 ± 8 kilojoules per square metre and a fatigue threshold of 10.5 ± 1.3 kilojoules per square metre. The presented strategy is generalizable to other polymers, and could expand the applicability of structural hydrogels to conditions involving more demanding mechanical loading. A strategy that combines freeze-casting and salting-out treatments produces strong, tough, stretchable and fatigue-resistant poly(vinyl alcohol) hydrogels.

This paper explores the knowledge of linguistic structure learned by large artificial neural networks, trained via self-supervision, whereby the model simply tries to predict a masked word in a given context. Human language communication is via sequences of words, but language understanding requires constructing rich hierarchical structures that are never observed explicitly. The mechanisms for this have been a prime mystery of human language acquisition, while engineering work has mainly proceeded by supervised learning on treebanks of sentences hand labeled for this latent structure. However, we demonstrate that modern deep contextual language models learn major aspects of this structure, without any explicit supervision. We develop methods for identifying linguistic hierarchical structure emergent in artificial neural networks and demonstrate that components in these models focus on syntactic grammatical relationships and anaphoric coreference. Indeed, we show that a linear transformation of learned embeddings in these models captures parse tree distances to a surprising degree, allowing approximate reconstruction of the sentence tree structures normally assumed by linguists. These results help explain why these models have brought such large improvements across many language-understanding tasks.

We report a three-orders-of-magnitude variation of carrier lifetimes in exotic crystalline phases of gold nanoclusters (NCs) in addition to the well-known face-centered cubic structure, including hexagonal close-packed (hcp) Au30 and body-centered cubic (bcc) Au38 NCs protected by the same type of capping ligand. The bcc Au38 NC had an exceptionally long carrier lifetime (4.7 microseconds) comparable to that of bulk silicon, whereas the hcp Au30 NC had a very short lifetime (1 nanosecond). Although the presence of ligands may, in general, affect carrier lifetimes, experimental and theoretical results showed that the drastically different recombination lifetimes originate in the different overlaps of wave functions between the tetrahedral Au₄ building blocks in the hierarchical structures of these NCs.

Sensory (or afferent) nerves bring sensations of touch, pain, or temperature variation to the central nervous system and brain. Using the tools and materials of organic electronics, Kim et al. combined a pressure sensor, a ring oscillator, and an ion gel–gated transistor to form an artificial mechanoreceptor (see the Perspective by Bartolozzi). The combination allows for the sensing of multiple pressure inputs, which can be converted into a sensor signal and used to drive the motion of a cockroach leg in an oscillatory pattern. Science , this issue p. 998 ; see also p. 966 Organic flexible electronics mimic the functions of a biological afferent nerve and actuate muscles. The distributed network of receptors, neurons, and synapses in the somatosensory system efficiently processes complex tactile information. We used flexible organic electronics to mimic the functions of a sensory nerve. Our artificial afferent nerve collects pressure information (1 to 80 kilopascals) from clusters of pressure sensors, converts the pressure information into action potentials (0 to 100 hertz) by using ring oscillators, and integrates the action potentials from multiple ring oscillators with a synaptic transistor. Biomimetic hierarchical structures can detect movement of an object, combine simultaneous pressure inputs, and distinguish braille characters. Furthermore, we connected our artificial afferent nerve to motor nerves to construct a hybrid bioelectronic reflex arc to actuate muscles. Our system has potential applications in neurorobotics and neuroprosthetics.

Various natural materials have hierarchical microscale and nanoscale structures that allow for directional water transport. Here we report an ultrafast water transport process in the surface of a Sarracenia trichome, whose transport velocity is about three orders of magnitude faster than those measured in cactus spine and spider silk. The high velocity of water transport is attributed to the unique hierarchical microchannel organization of the trichome. Two types of ribs with different height regularly distribute around the trichome cone, where two neighbouring high ribs form a large channel that contains 1–5 low ribs that define smaller base channels. This results in two successive but distinct modes of water transport. Initially, a rapid thin film of water is formed inside the base channels (Mode I), which is followed by ultrafast water sliding on top of that thin film (Mode II). This two-step ultrafast water transport mechanism is modelled and experimentally tested in bio-inspired microchannels, which demonstrates the potential of this hierarchal design for microfluidic applications.

A bstract In modular-invariant models of flavour, hierarchical fermion mass matrices may arise solely due to the proximity of the modulus τ to a point of residual symmetry. This mechanism does not require flavon fields, and modular weights are not analogous to Froggatt-Nielsen charges. Instead, we show that hierarchies depend on the decomposition of field representations under the residual symmetry group. We systematically go through the possible fermion field representation choices which may yield hierarchical structures in the vicinity of symmetric points, for the four smallest finite modular groups, isomorphic to S 3 , A 4 , S 4 , and A 5 , as well as for their double covers. We find a restricted set of pairs of representations for which the discussed mechanism may produce viable fermion (charged-lepton and quark) mass hierarchies. We present two lepton flavour models in which the charged-lepton mass hierarchies are naturally obtained, while lepton mixing is somewhat fine-tuned. After formulating the conditions for obtaining a viable lepton mixing matrix in the symmetric limit, we construct a model in which both the charged-lepton and neutrino sectors are free from fine-tuning.

We demonstrate that the thermoelectric properties of p-type chalcogenides can be effectively improved by band convergence and hierarchical structure based on a high-entropy-stabilized matrix. The band convergence is due to the decreased light and heavy band energy offsets by alloying Cd for an enhanced Seebeck coefficient and electric transport property. Moreover, the hierarchical structure manipulated by entropy engineering introduces all-scale scattering sources for heat-carrying phonons resulting in a very low lattice thermal conductivity. Consequently, a peak zT of 2.0 at 900 K for p-type chalcogenides and a high experimental conversion efficiency of 12% at Δ T  = 506 K for the fabricated segmented modules are achieved. This work provides an entropy strategy to form all-scale hierarchical structures employing high-entropy-stabilized matrix. This work will promote real applications of low-cost thermoelectric materials. The synergism of entropy engineering and the typical optimization mechanisms in high-entropy-stabilized chalcogenide is unknown. Here, the authors find high-entropy-stabilized composition works as a promising matrix of applying synergistic effect to realize high thermoelectric performance.

Porous structured silicon has been regarded as a promising candidate to overcome pulverization of silicon-based anodes. However, poor mechanical strength of these porous particles has limited their volumetric energy density towards practical applications. Here we design and synthesize hierarchical carbon-nanotube@silicon@carbon microspheres with both high porosity and extraordinary mechanical strength (>200 MPa) and a low apparent particle expansion of ~40% upon full lithiation. The composite electrodes of carbon-nanotube@silicon@carbon-graphite with a practical loading (3 mAh cm −2 ) deliver ~750 mAh g −1 specific capacity, <20% initial swelling at 100% state-of-charge, and ~92% capacity retention over 500 cycles. Calendered electrodes achieve ~980 mAh cm −3 volumetric capacity density and <50% end-of-life swell after 120 cycles. Full cells with LiNi 1/3 Mn 1/3 Co 1/3 O 2 cathodes demonstrate >92% capacity retention over 500 cycles. This work is a leap in silicon anode development and provides insights into the design of electrode materials for other batteries. The authors here construct hierarchical porous CNT@Si@C microspheres as anodes for Li-ion batteries, enabling both high electrochemical performance and excellent mechanical strength. The work highlights the importance of mechanical properties in developing battery materials for practical applications.

We investigate how neural networks can learn and process languages with hierarchical, compositional semantics. To this end, we define the artificial task of processing nested arithmetic expressions, and study whether different types of neural networks can learn to compute their meaning. We find that recursive neural networks can implement a generalising solution to this problem, and we visualise this solution by breaking it up in three steps: project, sum and squash. As a next step, we investigate recurrent neural networks, and show that a gated recurrent unit, that processes its input incrementally, also performs very well on this task: the network learns to predict the outcome of the arithmetic expressions with high accuracy, although performance deteriorates somewhat with increasing length. To develop an understanding of what the recurrent network encodes, visualisation techniques alone do not suffice. Therefore, we develop an approach where we formulate and test multiple hypotheses on the information encoded and processed by the network. For each hypothesis, we derive predictions about features of the hidden state representations at each time step, and train 'diagnostic classifiers' to test those predictions. Our results indicate that the networks follow a strategy similar to our hypothesised 'cumulative strategy', which explains the high accuracy of the network on novel expressions, the generalisation to longer expressions than seen in training, and the mild deterioration with increasing length. This in turn shows that diagnostic classifiers can be a useful technique for opening up the black box of neural networks. We argue that diagnostic classification, unlike most visualisation techniques, does scale up from small networks in a toy domain, to larger and deeper recurrent networks dealing with real-life data, and may therefore contribute to a better understanding of the internal dynamics of current state-of-the-art models in natural language processing.