Would you briefly introduce yourself and your activities at Fraunhofer ISIT?  

I head the Innovative Devices group in the business area of power electronics at Fraunhofer ISIT. Together with my colleagues, we research novel devices based on gallium nitride (GaN) and the use of piezoelectric as well as ferroelectric materials in electronics. 
 

With the social media camapgne we want to introduce the term "MEMSification". How would you explain it?

By "MEMSification" I mean the use of materials and processes from the classic MEMS manufacturing toolbox for applications that at first glance have nothing to do with MEMS. Two examples:

• A few years ago, it was discovered at ISIT that doped AlN is not only piezoelectric but also ferroelectric. Today, it is used not only in MEMS mirrors and speakers, but also in transistors to give them programmable properties.  
• During the fabrication of novel vertical GaN transistors, membranes of the material 20 times thinner than a human hair must be exposed and handled. In traditional electronics clean rooms, this is nearly impossible. In MEMS manufacturing, this is commonplace.
   

How would you describe the new Fraunhofer lead project NeurOSmart in a few words?

NeurOSmart involves 5 institutes and well over 50 people working together toward a vision: Sensors should learn to think while being significantly more energy efficient than today's alternatives. 
 

What was the main reason for developing such a sensor and how does it work?

The motivation behind this development is the massively increasing number of sensors in autonomous systems and IoT devices, as well as the accompanying escalating energy requirements of the associated data processing. Already today, mobile supercomputers are used for this purpose, which are impractical, especially for mobile devices.

For this purpose, a Fraunhofer LiDAR system is combined with a complex neuromorphic data analysis as an example. The special thing about neuromorphic - i.e. brain-inspired - hardware is that it requires at least 2 orders of magnitude less power than current computers. The trick is that the boundary between memory and CPU/GPU computations is broken and computations for, say, object recognition in sensor data can be performed directly in memory. In addition to energy efficiency, the computing hardware becomes so compact that it can be integrated directly into the sensor.
 

In which areas can the sensor be applied?

The sensor developed in this framework is used in a cooperation cell between humans and robots and monitors that both do not get in each other's way when handling workpieces. Such cells are used, for example, in the robot-assisted production of cars so that workers are relieved when handling particularly heavy workpieces. In principle, however, we are aiming to transfer the principle to almost any sensor.  
 

What is the current status of the development?

The project was started in January. We are currently developing and manufacturing the sensor components. In parallel, we finalized the concept of the computing hardware last month and are now developing the components. We should have the first prototype ready by the end of next year. It is still missing a few components, but we want to test it already at this time to get feedback from the application.
 

Can you give an outlook for the future of MEMS applications?

MEMS is alive and very much in flux. New materials, processes, and applications keep things evolving, which not only benefits MEMSers, but also provides microelectronics and power electronics with innovative ideas. Especially with MEMS sensors, I think they have to get smarter and smarter to keep up with the times.

Thank you very much for the interview, Michael!