Smart and compact sensor technology thanks to edge AI

Artificial intelligence systems have to process large amounts of data as quickly as possible. The project of the Technical University of Applied Sciences Wildau, the Leibniz Institute for Innovative Microelectronics (IHP) and the Fraunhofer Institute for Photonic Microsystems IPMS aims to enable complex calculations directly at the point of origin of the data, for example directly at the sensor itself.

Data processing using AI is currently often carried out via centralized cloud computing solutions. The data is calculated on central servers, which means that large amounts of data are transmitted over long distances. This can repeatedly lead to data leaks and thus to opportunities for unauthorized third parties to attack. Decentralized data processing not only improves data protection, but also enables real-time capability of the systems, as data transfers over long distances are avoided.

Advanced sensor technology to solve material and integration problems

The 'Integrated Silicon Systems' branch of the Fraunhofer IPMS in Cottbus is working on the functional expansion and integration of existing MEMS sensors for edge AI applications. Signal processing is integrated directly into the sensor and data can be collected directly where it is generated. The aim is to increase the adaptability of sensors to different application scenarios without having to replace the underlying hardware.

Ultrasonic sensor chip (L-CMUT) from Fraunhofer IPMS © Fraunhofer IPMS

One of the first key areas of development at Fraunhofer IPMS is gas analysis using ion mobility spectrometers (IMS). An IMS makes it possible to detect ionizable analyte substances directly in the air, even at very low concentrations. Existing approaches lack sufficient miniaturization. A first IMS demonstrator, which is based on a FAIMS approach (field asymmetric-waveform ion mobility spectrometry), has flexible electrode spacing, making it possible to overcome this hurdle.

Furthermore, the goal of a data-supported evaluation of photodetectors for the near-infrared wavelength range is being pursued. These are used, for example, in material analysis and recycling and even make it possible to analyze through packaging. In particular, the focus here is on improving an Al-TiN-Si Schottky detector component with cylindrical pyramidal structures for higher sensitivity and better scalability using cheaper materials.

A third area deals with the adapted use of capacitive micromechanical ultrasonic transducers (CMUTs) for improved imaging. CMUTs are highly sensitive ultrasound receivers due to their size and capacitive operating principle. Signal evaluation close to the sensor enables faster imaging. "In the future, it will be possible to carry out very precise analyses of hand movements using an ultrasound signal based on that of bats, as well as measuring blood sugar levels using ultrasound," explains Dr. Sebastian Meyer, Head of the Integrated Silicon Systems division at Fraunhofer IPMS.

The TH Wildau and the Leibniz IHP will then use the generated data to train edge AI systems for fast and precise data processing. The results of the project will enable further steps to be taken towards more intelligent and compact sensor systems.