There’s always room for improvement

Unintentionally blurred images have become a rare occurrence in the age of smartphone cameras. This has to do with tiny sensors that are built into the smart phones, so-called MEMS (micro-electromechanical systems). Among other things, they detect the trembling of a hand holding the phone or other movements by means of a rotary rate sensor and accelerometer to enable optical image stabilization. But MEMS can do even more: In cars, they are used for safety purposes. Based on the sensor signals, the control unit decides whether the airbag should be triggered or if the electronic stability program (ESP) needs to take action. MEMS components are also used in the healthcare sector, for example in modern diagnostic instruments such as the Vivalytic platform developed by Bosch. It uses a structure manufactured with MEMS. Another example, BioMEMS technology, combines MEMS technology with microbiology.

Bosch is one of the world’s largest suppliers of MEMS sensors for vehicles and consumer electronics. The company’s associates produce more than four million of these miniature sensors every day. This makes Bosch the number one player in the MEMS market. MEMS technology is a challenging field: “Despite their small size, developing and manufacturing MEMS is an incredibly complex task that requires a high degree of precision and expertise,” says Matthias Wenzel, research engineer at Bosch Research, who has been involved in the ongoing development of tools and methodologies in MEMS design for many years. Quantum physicist Wenzel, who holds a doctorate, explains: “These sophisticated mechanical devices have to meet a number of stringent requirements to be successful on the market.” Key parameters that determine their success include development costs as well as the product’s time to market. State-of-the-art algorithms have long been used for designing MEMS. Generative AI (GenAI for short), has the potential to become a new, potent tool. It could speed up development, making the process more efficient in terms of resources and cost.

Matthias Wenzel and his team joined forces with Paul Baireuther, a research engineer at Bosch’s Artificial Intelligence Research department. Together with Bosch MEMS design experts from Mirko Hofmann's department in Reutlingen, Germany, they are developing an automated toolchain for MEMS design and optimization. With the help of modern algorithms, they optimize MEMS structures on the local level and use GenAI to create new topologies for critical parts of the design, i.e., for the basic structure of the sensor. “Our innovative approach promises to revolutionize the design process and offers a glimpse into a future where AI can support human innovation even better and thus significantly increase the quality of results,” says Matthias Wenzel. Why put in all this effort if Bosch is one of the world's largest suppliers? “In terms of the entire MEMS market, which includes more than just MEMS sensors, Bosch is in the leading position,” says Matthias Wenzel, “but we understand that AI has the potential to revolutionize entire industries.” The aim is to maintain our leading position by further increasing the sensors’ performance.

The team focused on a single component of the MEMS to prove the feasibility of their approach. Developing a small piece of silicon like this is a complex task in its own right. It involves weeks or even months of manual topology adjustment and local optimization. As this is the standard approach in research, the team had access to state-of-the-art designs they could compare to the AI-based suggestions. The result: Their tailored generative algorithm designed a completely new topology approach in just a few days, offering new solution spaces to further improve specific product characteristics.

GenAI can also help to greatly improve the design process itself. Conventional optimization methods are well suited to local optimization, but since they are always based on an engineer’s design, they are not very flexible. So when humans exclude a design option from the outset or have not considered it at all, certain development paths might be overlooked, or it might take a lot longer until they are discovered. In contrast, an AI-driven topology generator can create and evaluate thousands of different designs within a short time, accounting for far more possibilities than a human designer can. Based on defined requirements, the algorithm creates a series of optimized designs. The engineers can scrutinize these and take their favorite pick.

At present, Matthias Wenzel and his team’s approach is still in an early stage. They have provided the first proof of concept, and the next steps are clear: The MEMS component generated with the help of their algorithm will be placed on a silicon wafer and will be ready for integration into a product in 2025. If all goes according to plan, it will reach market maturity in 2027. In the meantime, the method will be applied in a broader scope than individual components, drawing on its potential for the entire MEMS spectrum in the near future. “The journey has only just begun,” says Matthias Wenzel, “but GenAI in MEMS design will undeniably change the way we develop these tiny devices that power our modern world.”