GPU-enabled microfluidic design automation for concentration gradient generators

• Published in: Engineering with computers Vol. 39; no. 2; pp. 1637 - 1652
• Main Authors: Hong, Seong Hyeon, Shu, Jung-Il, Ou, Junlin, Wang, Yi
• Format: Journal Article
• Language: English
• Published: London Springer London 01.04.2023; Springer Nature B.V
• Subjects: Automation; CAE) and Design; Calculus of Variations and Optimal Control; Optimization; Classical Mechanics; Computer Science; Computer-Aided Engineering (CAD; Concentration gradient; Control; Edge computing; Genetic algorithms; Global optimization; Laminar flow; Math. Applications in Chemistry; Mathematical and Computational Engineering; Microfluidics; Nonuniformity; Optimization; Original Article; Parallel processing; Systems Theory; Workstations; CUDA; Concentration gradient generator; Genetic algorithm; GPGPU; Microfluidics
• ISSN: 0177-0667; 1435-5663
• DOI: 10.1007/s00366-021-01548-8

A GPU-enabled design framework is presented to automate the global optimization process of microfluidic concentration gradient generators (µCGGs). The optimization finds operational parameters (inlet concentrations and pressures) of CGGs to produce desired/prescribed concentration gradient (CG) profiles. To enhance optimization speed, the physics-based component model (PBCM) in the closed form is employed for simulation in lieu of the expensive CFD model. A genetic algorithm (GA) including the migration to mitigate the pre-maturation issue is developed. A new approach to include a penalty term that minimizes pressure non-uniformity in CGGs and the chance of violating physical assumptions used by PBCM is proposed. The entire process of PBCM evaluation and GA optimization is implemented on the GPU platform to utilize its massive computing parallelization. Two different laminar flow diffusion-based microfluidic CGGs: triple-Y and double-Ψ are used to verify the framework. Various shapes of desired CGs are examined for triple-Y and double-Ψ, and the average mean relative errors between the optimal designs and desired CGs are found to be 3.85% and 3.97%, respectively. The optimization is completed within 150 s on a GPGPU workstation and within 25 min on a GPU-embedded, small form-factor edge computing device, leading to about 130 × and 11–12 × speedup over the CPU process, respectively. The present research for the first time demonstrates the potential of applying microfluidic design automation on the edge-computing platform in laboratory environments.