Application-orientated design of micro thermoelectric coolers using a Machine learning-driven reverse framework

Micro thermoelectric coolers (μ-TECs) represent promising solution for “hot-spot” local cooling within confined spaces. However, systematically conducting the full-parameter optimization design of μ-TECs tailored to specific applications remains a formidable challenge for researchers, attributed to the intricate influencing parameters and manufacturing processes. Herein, a novel knowledge-data-driven research paradigm is proposed for the multifactorial multi-objective optimization of customized μ-TECs capitalizing the rational combination of the numerical analysis, finite element method (FEM) and deep neural networks (DNNs). Specifically, this approach is applied to the customized design of two distinct μ-TECs that function as heat pumps and coolers design, respectively, providing more comprehensive device-level insights. Regarding the most rational multifactorial multi-objective optimization strategy, an evaluation criterion for parameter priority is presented to guide the materials and interfaces development based on the application-orientated optimization profits. Furthermore, a database-enabled reverse framework is developed for deducing optimal interfacial and geometric design that align with specific materials, scenarios, and requirement. Consequently, the batch-produced 18-pair modules designed through the reverse framework achieve the maximum cooling density and temperature of 29.1 W cm⁻² and 70.3 K, respectively. Notably, our customized devices provide a superior cooling density of 8.4 W cm⁻² in the target applications requiring stable cooling of 50 K, surpassing most advanced general-purpose products. This study offers a high-efficient knowledge-data-driven methodology for the rapid and reverse design of high-performance μ-TECs with tailored application scenarios and requirements.