Fig. 3. Examples of ML-assisted workflows for design automation of microfluidic devices. (A) Direct approaches use ML reverse models that convert the desired performance to microfluidic design parameters. (A.i) ML-assisted direct design automation were demonstrated for flow sculpting in inertial flows, figure reproduced from Stoecklein et al., 2017 (ref. 84) licensed by CC BY 4.0; and (A.ii) for generating user-specified concentration gradients, figure redrawn from Hong et al., 2020 (ref. 76) licensed by CC BY 4.0. (B) Iterative design automation uses ML forward models that convert microfluidic design parameters to the predicted performance in conjunction with a iterative search algorithm to convert the user specified desired performance to the necessary design parameters. Iterative design automation were demonstrated for (B.i) design automation of output solute concentrations and flow rates in randomly designed 2-inlet, 3-outlet grid micromixers,⁷¹ random mixer figure reproduced from Wang et al., 2016 (ref. 83) with permission from the Royal Society of Chemistry; and (B.ii) for droplet diameter and generation rate in flow-focusing microfluidic droplet generators, figure reproduced from Lashkaripour et al., 2021 (ref. 13) licensed by CC BY 4.0.