Today, we are excited to speak with Tomasz Kołcon, one of the leading authors of the scientific paper titled “AI-based Positioning on a Micrometer Scale.” The paper, published as part of the AI-PRISM project, explores innovative approaches to enhancing the precision of semiconductor manufacturing through the integration of AI-driven human-robot collaboration. In this study, Tomasz and his team present a novel method using a motorized XY stage and AI-based detection to automate the positioning of semiconductor elements and defects detection, which traditionally relied on manual procedures.
The project aims to reduce human strain, improve production efficiency, and maintain high standards in the quality of infrared detectors. Today, we will delve into Tomasz’s insights on the research process, the challenges encountered, and the broader implications of their findings.
What inspired your team to explore AI-based solutions for micrometer-scale positioning in semiconductor manufacturing, and how did this idea evolve into the research presented in your paper?
The idea of using AI to support certain stages of semiconductor manufacturing came to our minds some time ago. We were looking for new applications for AI in the fields we specialize in. At the same time, our colleagues at VIGO Photonics were continuously working on improving production processes. This coincided with the time when we were submitting an application for the AI-PRISM project. After a brainstorming session, we selected several ideas, and one of them is currently being implemented in practice as part of one of the pilots in the aforementioned project.
Can you elaborate on the specific challenges of manual semiconductor positioning and defects detection that your AI-driven approach aims to solve, and how does this approach impact the role of human operators in the production process?
Let’s start with the fact that at VIGO Photonics, some of the work is done manually for various objective reasons. We decided to improve one of such processes, i.e. gluing the stick to the chip (needed in the grinding and polishing process) as part of the AI-PRISM project. Before we began the strictly technical work, it was preceded by thorough research on human aspects conducted with our partner, Cranfield University. In our approach, we don’t aim to replace the human operator but to support them in the most stressful stages of the entire process. These elements were identified, and that’s what we focused on in the subsequent work. Currently, the chip positioning process, which is observed under a microscope, is supported by AI. Additionally, defects in the chip structure can be very easily detected during this process. This greatly relieves the operator, improves their mental comfort, and also allows for faster work and a reduction in the number of production rejects.
The paper mentions using a motorized XY table and a microscope camera for precise positioning. Could you explain the technical aspects and key innovations behind this setup, particularly the AI components involved?
Due to the small dimensions of the chip itself (from 1x1mm to approx. 3x3mm) and the even smaller dimensions of individual elements visible on its structure (even in the order of single micrometers), instead of a classic robotic arm we had to use precise tables and a microscopic camera. Of course, the station itself is more complicated, but these are its most important components, which allow us, on the one hand, to observe details with dimensions below 1um, and on the other hand, enable precise positioning with such a small step. One could say that detecting the “center” of the chip can be done using classic computer vision methods, but using AI we can do it more effectively due to the fact that even in the same series, chips can differ from each other. Additionally, the use of AI allows for the detection of structural defects at a fairly early stage of production, which provides additional benefits.
In your study, you discuss the importance of human-robot collaboration. How does keeping the final decision-making power with human operators enhance the acceptance and effectiveness of AI in industrial settings?
The entire process is very important in the entire production chain, so a lot of attention is paid to it. We assumed from the outset that we would focus on cooperation between humans, robots and AI to support us, but it is the humans who are to make the final decision. In principle, as it has been done so far, except that the most burdensome stages of the processes are done for us by the machine. There is no room for mistakes here, and correct positioning or detecting defects is not always so easy. Relieving the operator by automating certain stages and leaving him basically only the final decision means that at the end of the working day he will not be so tired and will certainly reduce the number of errors.
Looking ahead, what do you see as the next steps or future developments for AI-based positioning systems in manufacturing? Are there plans to extend this technology to other applications or industries?
The technology being developed opens many new doors for us. Firstly, for Łukasiewicz-PIAP it is a descent to the micro scale, which undoubtedly expands the group of potential customers. Secondly, the use of AI means that we will be pioneers in terms of technology. The AI-PRISM project is still ongoing, so we will try to develop this technology as much as possible. We also believe that the current fruitful cooperation with VIGO Photonics will continue within other projects or commercial orders.
Thank you for answering our questions, and good luck with your further research! The whole paper can be found here: