(271i) Machine Intelligence-Accelerated Design of Ultrastretchable Electrodeswith Strain-Insensitive Functionalities

Stretchable electronics, with their adaptable functionalities and elastic properties, are increasingly in demand in a range of emerging technologies, including stretchable transistors, stretchable light-emitting diodes, wearable sensors, stretchable energy harvesters, wearable batteries, and intelligent soft machines. Central to these technologies are stretchable electrodes, which are required to endure substantial deformations without sacrificing key functions, such as sensitivity to specific analytes, electrical conductivity, and electrochemical activity. Additionally, the mechanical durability of stretchable electrodes is critical to ensure their longevity for practical exploitations. Conventional electrode materials, including vapor-evaporated metals (like gold and copper), electro-deposited metals (like zinc and platinum), and post-decorated metal oxides (like MnO₂ and indium tin oxide), are inherently brittle and prone to crack upon stretching, leading to the loss of electrode functions. Therefore, to fabricate stretchable electrodes that can withstand rigorous mechanical strains and remain functionally reliable, it is crucial to develop effective strategies that prevent cracking and halt its progression.

Various strategies have been explored in structural engineering for stretchable electrodes, such as percolating networks, serpentine interconnects, wavy structures, and two-dimensional interlayers. All these strategies aim to balance mechanical deformability with functional integrity. However, the complex correlations between the electrode’s configuration - including its composition, thickness, and morphology - and the fabrication parameters, as well as the electrode’s ultimate performance, are often not fully understood. These elusive correlations necessitate extensive experiments to identify the optimal parameter sets for fabricating stretchable electrodes with advanced functionalities and customizable deformability. To accelerate the design process of stretchable electrodes and minimize the reliance on empirical guesswork, the development of a prediction model would be advantageous. The prospect of a prediction model can efficiently explore the parameter space with multiple degrees of freedom (DOFs) and address critical trade-offs (between stretchability, performance, and durability) for stretchable electrodes. By leveraging the model’s predictive capability, the development cycle of stretchable electrodes with strain-insensitive functionalities can be significantly accelerated.

Machine learning (ML), a subset of artificial intelligence (AI), plays a crucial role in building models for making predictions and recommendations. Emerging AI/ML methodologies offer a robust toolkit capable of revealing intricate correlations in various sectors. In the realm of materials science, the integration of AI/ML predictions has expanded considerably, particularly in areas where computational simulations and high-throughput analytical instruments can produce substantial datasets, such as sustainable energy, catalysis, and flexible electronics. However, developing accurate prediction models for the design of stretchable electrodes presents several challenges. First, different research groups tend to use distinct building blocks and establish their specific fabrication and testing protocols, resulting in data inconsistencies. Second, most recent studies have concentrated on optimizing a singular function of stretchable electrodes (e.g., stretchability or conductivity). Last, the acquisition rate of high-quality data points for stretchable electrodes is inherently low, making it difficult to build an AI/ML model that can effectively execute multi-faceted property optimization.

In this work, an integrated workflow that leverages the synergistic capabilities of collaborative robotics, AI/ML predictions, and finite element (FE) simulations was established to accelerate the design of ultrastretchable electrodes with strain-consistent functionalities, in terms of electrical conductivity and electrochemical performance. Four building blocks, including two-dimensional (2D) Ti₃C₂T_x MXene nanosheets, one-dimensional (1D) single-walled carbon nanotubes (SWNTs), zero-dimensional (0D) gold nanoparticles (AuNPs), and polyvinyl alcohol (PVA) binder, were selected for the fabrication of stretchable electrodes. An automated pipetting robot was commanded to prepare 286 aqueous mixtures with varying MXene/SWNT/AuNP/PVA ratios (Figure A). After vacuum filtration, a library of nanocomposites was created, and their conductance values were input to train a support vector machine (SVM) model (Figure B). Next, through active learning loops with data augmentation, 146 different kinds of stretchable electrodes were stagewise fabricated with varying compositions, thicknesses, deformation sequences, and applied pre-strains. The electrical conductance and strain response of these electrodes informed the development of an artificial neural network (ANN)-based model, accurately predicting the electrode’s electrical conductance and strain responses based on a set of fabrication parameters (Figure C). Furthermore, SHapley Additive exPlanations (SHAP) model interpretation was conducted, and the data-driven design principles were validated through FE simulations. As demonstrations, the ML-driven design led to the discovery of two exemplary stretchable electrodes with strain-insensitive functionalities. First, by following the model-suggested parameters, a stretchable gold conductor was fabricated, showcasing metal-like conductivity (2.2×10⁷ S·m), exceptional stretchability with minimal strain responses (<10% resistance increases under 900% strains), and enduring electromechanical stability (600% strains over 50,000 cycles). Second, a stretchable Zn metal battery (with a Zn anode and a MnO₂ cathode) was fabricated, demonstrating >300% stretchability and strain-resilient electrochemical performance. Our hybrid approach, involving robot-assisted experiments, data science, and simulation tools, offers an unconventional design platform to accelerate the invention of strain-insensitive stretchable electrodes with customizable opportunities.