Machine learning without a processor: Emergent learning in a nonlinear analog network.

Standard deep learning algorithms require differentiating large nonlinear networks, a process that is slow and power-hungry. Electronic contrastive local learning networks (CLLNs) offer potentially fast, efficient, and fault-tolerant hardware for analog machine learning, but existing implementations are linear, severely limiting their capabilities. These systems differ significantly from artificial neural networks as well as the brain, so the feasibility and utility of incorporating nonlinear elements have not been explored. Here, we introduce a nonlinear CLLN-an analog electronic network made of self-adjusting nonlinear resistive elements based on transistors. We demonstrate that the system learns tasks unachievable in linear systems, including XOR (exclusive or) and nonlinear regression, without a computer. We find our decentralized system reduces modes of training error in order (mean, slope, curvature), similar to spectral bias in artificial neural networks. The circuitry is robust to damage, retrainable in seconds, and performs learned tasks in microseconds while dissipating only picojoules of energy across each transistor. This suggests enormous potential for fast, low-power computing in edge systems like sensors, robotic controllers, and medical devices, as well as manufacturability at scale for performing and studying emergent learning.

Materials for electronically controllable microactuators.

Abstract: Electronically controllable actuators have shrunk to remarkably small dimensions, thanks to recent advances in materials science. Currently, multiple classes of actuators can operate at the micron scale, be patterned using lithographic techniques, and be driven by complementary metal oxide semiconductor (CMOS)-compatible voltages, enabling new technologies, including digitally controlled micro-cilia, cell-sized origami structures, and autonomous microrobots controlled by onboard semiconductor electronics. This field is poised to grow, as many of these actuator technologies are the firsts of their kind and much of the underlying design space remains unexplored. To help map the current state of the art and set goals for the future, here, we overview existing work and examine how key figures of merit for actuation at the microscale, including force output, response time, power consumption, efficiency, and durability are fundamentally intertwined. In doing so, we find performance limits and tradeoffs for different classes of microactuators based on the coupling mechanism between electrical energy, chemical energy, and mechanical work. These limits both point to future goals for actuator development and signal promising applications for these actuators in sophisticated electronically integrated microrobotic systems.

Colloidal robotics.

Robots have components that work together to accomplish a task. Colloids are particles, usually less than 100 µm, that are small enough that they do not settle out of solution. Colloidal robots are particles capable of functions such as sensing, computation, communication, locomotion and energy management that are all controlled by the particle itself. Their design and synthesis is an emerging area of interdisciplinary research drawing from materials science, colloid science, self-assembly, robophysics and control theory. Many colloidal robot systems approach synthetic versions of biological cells in autonomy and may find ultimate utility in bringing these specialized functions to previously inaccessible locations. This Perspective examines the emerging literature and highlights certain design principles and strategies towards the realization of colloidal robots.

Emergent microrobotic oscillators via asymmetry-induced order.

Spontaneous oscillations on the order of several hertz are the drivers of many crucial processes in nature. From bacterial swimming to mammal gaits, converting static energy inputs into slowly oscillating power is key to the autonomy of organisms across scales. However, the fabrication of slow micrometre-scale oscillators remains a major roadblock towards fully-autonomous microrobots. Here, we study a low-frequency oscillator that emerges from a collective of active microparticles at the air-liquid interface of a hydrogen peroxide drop. Their interactions transduce ambient chemical energy into periodic mechanical motion and on-board electrical currents. Surprisingly, these oscillations persist at larger ensemble sizes only when a particle with modified reactivity is added to intentionally break permutation symmetry. We explain such emergent order through the discovery of a thermodynamic mechanism for asymmetry-induced order. The on-board power harvested from the stabilised oscillations enables the use of electronic components, which we demonstrate by cyclically and synchronously driving a microrobotic arm. This work highlights a new strategy for achieving low-frequency oscillations at the microscale, paving the way for future microrobotic autonomy.

Microscopic robots with onboard digital control.

Autonomous robots-systems where mechanical actuators are guided through a series of states by information processing units to perform a predesigned function-are expected to revolutionize everything from health care to transportation. Microscopic robots are poised for a similar revolution in fields from medicine to environmental remediation. A key hurdle to developing these microscopic robots is the integration of information systems, particularly electronics fabricated at commercial foundries, with microactuators. Here, we develop such an integration process and build microscopic robots controlled by onboard complementary metal oxide semiconductor electronics. The resulting autonomous, untethered robots are 100 to 250 micrometers in size, are powered by light, and walk at speeds greater than 10 micrometers per second. In addition, we demonstrate a microscopic robot that can respond to an optical command. This work paves the way for ubiquitous autonomous microscopic robots that perform complex functions, respond to their environments, and communicate with the outside world.

Cilia metasurfaces for electronically programmable microfluidic manipulation.

Cilial pumping is a powerful strategy used by biological organisms to control and manipulate fluids at the microscale. However, despite numerous recent advances in optically, magnetically and electrically driven actuation, development of an engineered cilial platform with the potential for applications has remained difficult to realize1-6. Here we report on active metasurfaces of electronically actuated artificial cilia that can create arbitrary flow patterns in liquids near a surface. We first create voltage-actuated cilia that generate non-reciprocal motions to drive surface flows at tens of microns per second at actuation voltages of 1 volt. We then show that a cilia unit cell can locally create a range of elemental flow geometries. By combining these unit cells, we create an active cilia metasurface that can generate and switch between any desired surface flow pattern. Finally, we integrate the cilia with a light-powered complementary metal-oxide-semiconductor (CMOS) clock circuit to demonstrate wireless operation. As a proof of concept, we use this circuit to output voltage pulses with various phase delays to demonstrate improved pumping efficiency using metachronal waves. These powerful results, demonstrated experimentally and confirmed using theoretical computations, illustrate a pathway towards fine-scale microfluidic manipulation, with applications from microfluidic pumping to microrobotic locomotion.

Micrometer-sized electrically programmable shape-memory actuators for low-power microrobotics.

Shape-memory actuators allow machines ranging from robots to medical implants to hold their form without continuous power, a feature especially advantageous for situations where these devices are untethered and power is limited. Although previous work has demonstrated shape-memory actuators using polymers, alloys, and ceramics, the need for micrometer-scale electro-shape-memory actuators remains largely unmet, especially ones that can be driven by standard electronics (~1 volt). Here, we report on a new class of fast, high-curvature, low-voltage, reconfigurable, micrometer-scale shape-memory actuators. They function by the electrochemical oxidation/reduction of a platinum surface, creating a strain in the oxidized layer that causes bending. They bend to the smallest radius of curvature of any electrically controlled microactuator (~500 nanometers), are fast (<100-millisecond operation), and operate inside the electrochemical window of water, avoiding bubble generation associated with oxygen evolution. We demonstrate that these shape-memory actuators can be used to create basic electrically reconfigurable microscale robot elements including actuating surfaces, origami-based three-dimensional shapes, morphing metamaterials, and mechanical memory elements. Our shape-memory actuators have the potential to enable the realization of adaptive microscale structures, bio-implantable devices, and microscopic robots.

Electronically integrated, mass-manufactured, microscopic robots.

Fifty years of Moore's law scaling in microelectronics have brought remarkable opportunities for the rapidly evolving field of microscopic robotics1-5. Electronic, magnetic and optical systems now offer an unprecedented combination of complexity, small size and low cost6,7, and could be readily appropriated for robots that are smaller than the resolution limit of human vision (less than a hundred micrometres)8-11. However, a major roadblock exists: there is no micrometre-scale actuator system that seamlessly integrates with semiconductor processing and responds to standard electronic control signals. Here we overcome this barrier by developing a new class of voltage-controllable electrochemical actuators that operate at low voltages (200 microvolts), low power (10 nanowatts) and are completely compatible with silicon processing. To demonstrate their potential, we develop lithographic fabrication-and-release protocols to prototype sub-hundred-micrometre walking robots. Every step in this process is performed in parallel, allowing us to produce over one million robots per four-inch wafer. These results are an important advance towards mass-manufactured, silicon-based, functional robots that are too small to be resolved by the naked eye.

Bidirectional Self-Folding with Atomic Layer Deposition Nanofilms for Microscale Origami.

Origami design principles are scale invariant and enable direct miniaturization of origami structures provided the sheets used for folding have equal thickness to length ratios. Recently, seminal steps have been taken to fabricate microscale origami using unidirectionally actuated sheets with nanoscale thickness. Here, we extend the full power of origami-inspired fabrication to nanoscale sheets by engineering bidirectional folding with 4 nm thick atomic layer deposition (ALD) SiNx-SiO2 bilayer films. Strain differentials within these bilayers result in bending, producing microscopic radii of curvature. We lithographically pattern these bilayers and localize the bending using rigid panels to fabricate a variety of complex micro-origami devices. Upon release, these devices self-fold according to prescribed patterns. Our approach combines planar semiconductor microfabrication methods with computerized origami design, making it easy to fabricate and deploy such microstructures en masse. These devices represent an important step forward in the fabrication and assembly of deployable micromechanical systems that can interact with and manipulate micro- and nanoscale environments.

Capillary Origami with Atomically Thin Membranes.

Small-scale optical and mechanical components and machines require control over three-dimensional structure at the microscale. Inspired by the analogy between paper and two-dimensional materials, origami-style folding of atomically thin materials offers a promising approach for making microscale structures from the thinnest possible sheets. In this Letter, we show that a monolayer of molybdenum disulfide (MoS2) can be folded into three-dimensional shapes by a technique called capillary origami, in which the surface tension of a droplet drives the folding of a thin sheet. We define shape nets by patterning rigid metal panels connected by MoS2 hinges, allowing us to fold micron-scale polyhedrons. Finally, we demonstrate that these shapes can be folded in parallel without the use of micropipettes or microfluidics by means of a microemulsion of droplets that dissolves into the bulk solution to drive folding. These results demonstrate controllable folding of the thinnest possible materials using capillary origami and indicate a route forward for design and parallel fabrication of more complex three-dimensional micron-scale structures and machines.

Atomic Layer Deposition for Membranes, Metamaterials, and Mechanisms.

Bending and folding techniques such as origami and kirigami enable the scale-invariant design of 3D structures, metamaterials, and robots from 2D starting materials. These design principles are especially valuable for small systems because most micro- and nanofabrication involves lithographic patterning of planar materials. Ultrathin films of inorganic materials serve as an ideal substrate for the fabrication of flexible microsystems because they possess high intrinsic strength, are not susceptible to plasticity, and are easily integrated into microfabrication processes. Here, atomic layer deposition (ALD) is employed to synthesize films down to 2 nm thickness to create membranes, metamaterials, and machines with micrometer-scale dimensions. Two materials are studied as model systems: ultrathin SiO2 and Pt. In this thickness limit, ALD films of these materials behave elastically and can be fabricated with fJ-scale bending stiffnesses. Further, ALD membranes are utilized to design micrometer-scale mechanical metamaterials and magnetically actuated 3D devices. These results establish thin ALD films as a scalable basis for micrometer-scale actuators and robotics.

Graphene-based bimorphs for micron-sized, autonomous origami machines.

Origami-inspired fabrication presents an attractive platform for miniaturizing machines: thinner layers of folding material lead to smaller devices, provided that key functional aspects, such as conductivity, stiffness, and flexibility, are persevered. Here, we show origami fabrication at its ultimate limit by using 2D atomic membranes as a folding material. As a prototype, we bond graphene sheets to nanometer-thick layers of glass to make ultrathin bimorph actuators that bend to micrometer radii of curvature in response to small strain differentials. These strains are two orders of magnitude lower than the fracture threshold for the device, thus maintaining conductivity across the structure. By patterning 2-[Formula: see text]m-thick rigid panels on top of bimorphs, we localize bending to the unpatterned regions to produce folds. Although the graphene bimorphs are only nanometers thick, they can lift these panels, the weight equivalent of a 500-nm-thick silicon chip. Using panels and bimorphs, we can scale down existing origami patterns to produce a wide range of machines. These machines change shape in fractions of a second when crossing a tunable pH threshold, showing that they sense their environments, respond, and perform useful functions on time and length scales comparable with microscale biological organisms. With the incorporation of electronic, photonic, and chemical payloads, these basic elements will become a powerful platform for robotics at the micrometer scale.

Measuring and Manipulating the Adhesion of Graphene.

We present a technique to precisely measure the surface energies between two-dimensional materials and substrates that is simple to implement and allows exploration of spatial and chemical control of adhesion at the nanoscale. As an example, we characterize the delamination of single-layer graphene from monolayers of pyrene tethered to glass in water and maximize the work of separation between these surfaces by varying the density of pyrene groups in the monolayer. Control of this energy scale enables high-fidelity graphene-transfer protocols that can resist failure under sonication. Additionally, we find that the work required for graphene peeling and readhesion exhibits a dramatic rate-independent hysteresis, differing by a factor of 100. This work establishes a rational means to control the adhesion of 2D materials and enables a systematic approach to engineer stimuli-responsive adhesives and mechanical technologies at the nanoscale.

Turning statistical physics models into materials design engines.

Despite the success statistical physics has enjoyed at predicting the properties of materials for given parameters, the inverse problem, identifying which material parameters produce given, desired properties, is only beginning to be addressed. Recently, several methods have emerged across disciplines that draw upon optimization and simulation to create computer programs that tailor material responses to specified behaviors. However, so far the methods developed either involve black-box techniques, in which the optimizer operates without explicit knowledge of the material's configuration space, or require carefully tuned algorithms with applicability limited to a narrow subclass of materials. Here we introduce a formalism that can generate optimizers automatically by extending statistical mechanics into the realm of design. The strength of this approach lies in its capability to transform statistical models that describe materials into optimizers to tailor them. By comparing against standard black-box optimization methods, we demonstrate how optimizers generated by this formalism can be faster and more effective, while remaining straightforward to implement. The scope of our approach includes possibilities for solving a variety of complex optimization and design problems concerning materials both in and out of equilibrium.

Optimizing a reconfigurable material via evolutionary computation.

Rapid prototyping by combining evolutionary computation with simulations is becoming a powerful tool for solving complex design problems in materials science. This method of optimization operates in a virtual design space that simulates potential material behaviors and after completion needs to be validated by experiment. However, in principle an evolutionary optimizer can also operate on an actual physical structure or laboratory experiment directly, provided the relevant material parameters can be accessed by the optimizer and information about the material's performance can be updated by direct measurements. Here we provide a proof of concept of such direct, physical optimization by showing how a reconfigurable, highly nonlinear material can be tuned to respond to impact. We report on an entirely computer controlled laboratory experiment in which a 6×6 grid of electromagnets creates a magnetic field pattern that tunes the local rigidity of a concentrated suspension of ferrofluid and iron filings. A genetic algorithm is implemented and tasked to find field patterns that minimize the force transmitted through the suspension. Searching within a space of roughly 10^{10} possible configurations, after testing only 1500 independent trials the algorithm identifies an optimized configuration of layered rigid and compliant regions.

Evolving design rules for the inverse granular packing problem.

If a collection of identical particles is poured into a container, different shapes will fill to different densities. But what is the shape that fills a container as close as possible to a pre-specified, desired density? We demonstrate a solution to this inverse-packing problem by framing it in the context of artificial evolution. By representing shapes as bonded spheres, we show how shapes may be mutated, simulated, and selected to produce particularly dense or loose packing aggregates, both with and without friction. Moreover, we show how motifs emerge linking these shapes together. The result is a set of design rules that function as an effective solution to the inverse packing problem for given packing procedures and boundary conditions. Finally, we show that these results are verified by experiments on 3D-printed prototypes used to make packings in the real world.

Particle shape effects on the stress response of granular packings.

We present measurements of the stress response of packings formed from a wide range of particle shapes. Besides spheres these include convex shapes such as the Platonic solids, truncated tetrahedra, and triangular bipyramids, as well as more complex, non-convex geometries such as hexapods with various arm lengths, dolos, and tetrahedral frames. All particles were 3D-printed in hard resin. Well-defined initial packing states were established through preconditioning by cyclic loading under given confinement pressure. Starting from such initial states, stress-strain relationships for axial compression were obtained at four different confining pressures for each particle type. While confining pressure has the largest overall effect on the mechanical response, we find that particle shape controls the details of the stress-strain curves and can be used to tune packing stiffness and yielding. By correlating the experimentally measured values for the effective Young's modulus under compression, yield stress and energy loss during cyclic loading, we identify trends among the various shapes that allow for designing a packing's aggregate behavior.

Self-organization into quantized eigenstates of a classical wave-driven particle.

A growing number of dynamical situations involve the coupling of particles or singularities with physical waves. In principle these situations are very far from the wave particle duality at quantum scale where the wave is probabilistic by nature. Yet some dual characteristics were observed in a system where a macroscopic droplet is guided by a pilot wave it generates. Here we investigate the behaviour of these entities when confined in a two-dimensional harmonic potential well. A discrete set of stable orbits is observed, in the shape of successive generalized Cassinian-like curves (circles, ovals, lemniscates, trefoils and so on). Along these specific trajectories, the droplet motion is characterized by a double quantization of the orbit spatial extent and of the angular momentum. We show that these trajectories are intertwined with the dynamical build-up of central wave-field modes. These dual self-organized modes form a basis of eigenstates on which more complex motions are naturally decomposed.

Adapting granular materials through artificial evolution.

Over 200 years after Coulomb's studies, a general connection between the mechanical response of a granular material and the constituents' shape remains unknown. The key difficulty in articulating this relationship is that shape is an inexhaustible parameter, making its systematic exploration infeasible. Here we show that the role of particle shape can, however, be explored efficiently when granular design is viewed in the context of artificial evolution. By introducing a mutable representation for particle shapes, we demonstrate with computer simulation how shapes can be evolved. As proof of principle, we predicted motifs that link shape to packing stiffness, discovered a particle that produces aggregates that stiffen-rather than weaken-under compression, and verified the results using three-dimensional printing. More generally, our approach facilitates the exploration of the role of arbitrary particle geometry in jammed systems, and invites the discovery and design of granular matter with optimized properties.

Droplet formation and scaling in dense suspensions.

When a dense suspension is squeezed from a nozzle, droplet detachment can occur similar to that of pure liquids. While in pure liquids the process of droplet detachment is well characterized through self-similar profiles and known scaling laws, we show here the simple presence of particles causes suspensions to break up in a new fashion. Using high-speed imaging, we find that detachment of a suspension drop is described by a power law; specifically we find the neck minimum radius, r(m), scales like near breakup at time τ = 0. We demonstrate data collapse in a variety of particle/liquid combinations, packing fractions, solvent viscosities, and initial conditions. We argue that this scaling is a consequence of particles deforming the neck surface, thereby creating a pressure that is balanced by inertia, and show how it emerges from topological constraints that relate particle configurations with macroscopic Gaussian curvature. This new type of scaling, uniquely enforced by geometry and regulated by the particles, displays memory of its initial conditions, fails to be self-similar, and has implications for the pressure given at generic suspension interfaces.