PT2 - AI and other Trends in MEMS Industry change Foundry Models
- Event
- EUROSENSORS XXXVII
2025-09-07 - 2025-09-10
Wroclaw - Band
- Lectures
- Chapter
- Plenary Talk
- Author(s)
- S. Majoni - Bosch, Reutlingen (Germany)
- Pages
- 22 - 23
- DOI
- 10.5162/EUROSENSORS2025/PT2
- ISBN
- 978-3-910600-07-2
- Price
- free
Abstract
AI requires drastically increase in computing power, which is not primary limited by the processors, but two other factors: 1.) Core to core communication, which will switch from electronic to optical data transfer to overcome current issues. Here a fundamental change from silicon- and silicon nitride-photonic to lithium niobate and similar is happening. The later one is incompatible with CMOS technology, shifting photonic from ASIC to MEMS foundries. The according customers request design- and product-support, increasing the foundry responsibility beyond pure wafer processing, which is a significant step for MEMS found-ries. 2.) Computing power is throttled due to overheating, requiring new cooling concepts. Beside water cooling of the racks, decentralized local cooling will be necessary. Here MEMS will play a key role due to low cost, power consumption and size, resulting in a multitude of designs and process flows for indi-vidual solution per processor as discrete element or integrated with the µprocessors. Here and in many other cases, it is obvious, that the standardized process approach as copied from ASIC found-ries is not successful. The wafer process reflects the mutual optimum of customers-design and -re-quirements and foundry capabilities. The trend for integrated MEMS-cooling, inertials with an ASIC cap, pMUT and medical fluidics on ASIC, etc reflects the merge of CMOS, MEMS and assembly pro-cessing [1]. This drives pure-play MEMS foundries to extend their scope. Additionally, MEMS-on-ASIC accelerates the 300mm transition of MEMS to be compatible with the advanced CMOS nodes.