Piezoelectric inkjets have enabled highly precise, customized material deposition for the micro and nanofabrication industries. Inkjet performance relies on a delicate balance between hardware manufacturing precision and tolerancing, material rheology, and the associated actuation-waveform for drop ejection. For a multi-nozzle inkjet system, the important performance metrics include drop volume resolution, inter-nozzle drop volume variation, overall drop-placement accuracy, jetting frequency (related to throughput), and long-term reliability. Cutting-edge applications typically require minimizing drop volume to 1 pL or less; drop placement accuracy of <10 mm, 3-σ; maintaining high jetting frequency of >10 kHz; and achieving reliable long-term operation, e.g., >1 month of continuous operation without any faults. Here, we strive towards a global optimum in drop resolution while adhering to constraints related to placement accuracy and reliability, at a fixed jetting frequency of 14.2 kHz. Our approach is to explore novel genetic algorithms that generate highly complicated waveforms (>100 parameter waveforms) to achieve our global optimum goals. Our automated genetic algorithm (a) starts from rudimentary waveforms that are easily constructed based on empirical knowledge and manual tuning, and (b) systematically increases complexity to generate highly sophisticated waveforms with increasingly greater number of parameters, an approach that is unattainable through manual tuning processes. This approach was developed and implemented using a commercial MEMS-based Fujifilm Samba G3L piezo-inkjet system with 2048 sub-20μm nozzles, with an aqueous diethylene glycol-based ink supplied by Fujifilm. The genetic algorithm operated based on a variable-length chromosome which started with 10 parameter waveforms and evolved into waveforms defined by > 100 parameters, resulting in drop-volume resolution of <600 fL, and maintaining drop placement accuracy to within 10 μm, 3-σ with reliable jetting over at least 30 print fields with the same waveform, each field nominally composed of 340 drops. The best result presented yielded drops with an average volume of 568 fL, which exceeded the manufacturer's specification for drop volume of 2.4 pL by 4.23 times. In summary, we have developed a novel framework for machine-in-the-loop waveform optimization in multi-nozzle inkjet systems to achieve previously unattained drop resolution with high reliability and accuracy.
Read full abstract