Armin Knoll

Self-driving laboratories can optimize experiments, but whether they can reason scientifically about what to test remains unclear. Bayesian optimization (BO) provides uncertainty-aware search without explicit scientific interpretation, whereas language-model agents can formulate hypotheses but lack calibrated uncertainty. Here we introduce maLLaMar , which combines these capabilities in a physical self-driving laboratory. Using the Belousov–Zhabotinsky reaction, we compare BO-only, agent-led and cooperative experiment selection under matched experimental budgets. Agent-led selection raised the fraction of hour-long oscillations from 24% to 64%, while cooperation yielded the largest number of near-optimal conditions, eleven compared with two for either strategy alone. The best conditions sustained oscillations for the full hour with mean periods up to 4.7 times shorter than the highest-scoring literature formulation re-measured on our platform. Hypotheses recorded prospectively during experimentation were evaluated against an exhaustive 294-condition reference map that was never available to the agent. The map supported or refined several predictions and falsified generalizations drawn from sparsely sampled regions. Statistical optimization and explicit scientific reasoning thus contribute complementary capabilities, allowing an autonomous laboratory both to find improved conditions and to generate claims that independent experiments can test.

We integrate fluidic networks filled with the oscillatory Belousov–Zhabotinsky (BZ) reaction in silicon microchips to investigate the potential of these networks for on-chip chemical computation. Using nanofluidic confinement and networks of thousands of microreactors (5–30 µm) connected by nanochannels, we demonstrate stable oscillations down to 300 nm vertical confinement with an estimated lower-bound chemical energy cost of ≈20 fJ per cycle when rescaled to the minimum operational volume, well below that of CMOS and VO ₂ oscillators. Reaction–diffusion waves propagate signals with diffusion lengths of ≈30–50 µm and speeds up to 123 µm/s, while light-sensitive modulation enables dynamic control for programmable architectures. These features support large fan-out, weighted connections, and reconfigurable pathways, laying the foundation for chemical logic, neuromorphic networks, and adaptive systems. Our results position oscillatory chemistry as a promising route toward ultra-low-power computing and bio-inspired architectures.

Chemical oscillators such as the Belousov–Zhabotinsky (BZ) reaction offer a promising foundation for energy-efficient, biologically inspired chemical information processing. While optical and mechanical manipulation of BZ has been extensively studied, electrochemical control has remained largely elusive and lacks fundamental understanding, in spite of its superior energy efficiency. Here, we investigate how a platinum electrode influences the dynamics of a ruthenium-catalyzed BZ reaction. We observe a pronounced acceleration of the oscillation frequency that scales with the electrode’s surface-to-solution-volume ratio. In addition, we demonstrate that small DC currents (<0.4 mA) further modulate the effect up to a full recovery of the natural period. To explain these findings, we develop a quantitative model based on mixed-potential theory, which accurately captures the observed behavior. This framework reveals how the low-volume electrochemical conversion of molecules at the electrode’s surface gets amplified by the chemistry and finally causes a pronounced impact on the entire system. This surprising effect, controllable through small applied currents, offers new strategies for designing programmable chemical systems for sensing and computing applications.

Chemical oscillators such as the Belousov-Zhabotinsky (BZ) reaction offer a promising foundation for energy-efficient, biologically inspired chemical information processing. While optical and mechanical manipulation of BZ has been extensively studied, electrochemical control has remained largely elusive and lacks fundamental understanding, in spite of its superior energy efficiency. Here, we investigate how a platinum electrode influences the dynamics of a ruthenium-catalyzed BZ reaction. We observe a pronounced acceleration of the oscillation frequency that scales with the electrode’s surface to solution-volume ratio. In addition, we demonstrate that small DC currents (<0.4 mA) further modulate the effect up to a full recovery of the natural period. To explain these findings, we develop a quantitative model based on mixed-potential theory, which accurately captures the observed behavior. This framework reveals how the low-volume electrochemical conversion of molecules at the electrode's surface gets amplified by the chemistry and finally causes a pronounced impact on the entire system. This surprising effect, controllable through small applied currents, offers new strategies for designing programmable chemical systems for sensing and computing applications.

Chemical oscillators such as the Belousov-Zhabotinsky (BZ) reaction offer a promising foundation for energy-efficient, biologically inspired chemical information processing. While optical and mechanical manipulation of BZ has been extensively studied, electrochemical control has remained largely elusive and lacks fundamental understanding, in spite of its superior energy efficiency. Here, we investigate how a platinum electrode influences the dynamics of a ruthenium-catalyzed BZ reaction. We observe a pronounced acceleration of the oscillation frequency that scales with the electrode’s surface to solution-volume ratio. In addition, we demonstrate that small DC currents (<0.4 mA) further modulate the effect up to a full recovery of the natural period. To explain these findings, we develop a quantitative model based on mixed-potential theory, which accurately captures the observed behavior. This framework reveals how the low-volume electrochemical conversion of molecules at the electrode's surface gets amplified by the chemistry and finally causes a pronounced impact on the entire system. This surprising effect, controllable through small applied currents, offers new strategies for designing programmable chemical systems for sensing and computing applications.

Numerous metabolic processes in nature are governed by extrinsic stimuli such as light and pH variations, which afford opportunities for synthetic and biological applications. In developing a multisensor apparatus, we have integrated submicrometer purple membrane patches, each harboring bacteriorhodopsin, onto a surface. Bacteriorhodopsin is a light-driven proton pump. We conducted monitoring of the interactions between this system and a pH-responsive supramolecular hydrogel to evaluate fibrous matrix growth. Initial photostimulation induced localized reductions in pH at the membrane surface, thereby catalyzing fibrogenesis within the hydrogel. Utilizing liquid atomic force microscopy alongside confocal laser scanning microscopy, we observed the hydrogel's morphogenesis and structural adaptations in real time. The system adeptly modulated microscale pH environments, fostering targeted fibrous development within the hydrogel matrix. This elucidates the potential for engineering responsive materials that emulate natural bioprocesses.

The ability to control the location of nanoscale objects in liquids is essential for fundamental and applied research from nanofluidics to molecular biology. To overcome their random Brownian motion, the electrostatic fluid trap creates local minima in potential energy by shaping electrostatic interactions with a tailored wall topography. However, this strategy is inherently static; once fabricated, the potential wells cannot be modulated. Here, we propose and experimentally demonstrate that such a trap can be controlled through a buried gate electrode. We measure changes in the average escape times of nanoparticles from the traps to quantify the induced modulations of 0.7 k_BT in potential energy and 50 mV in surface potential. Finally, we summarize the mechanism in a parameter-free predictive model, including surface chemistry and electrostatic fringing, that reproduces the experimental results. Our findings open a route toward real-time controllable nanoparticle traps.

We developed a process to fabricate nanoscale metallic gate electrodes on scanning probe cantilevers, including on the irregular surface of protruding cantilever tips. The process includes a floating-layer technique to coat the cantilevers in an electron-beam resist. We demonstrate gate definition through a lift-off process and through an etching process. The cantilevers maintain a high force sensitivity after undergoing the patterning process. Our method allows the patterning of nanoscale devices on fragile scanning probes, extending their functionality as sensors.