Embedded Systems

M/EDGE - Secure Low Power Medical Edge Computing

The M/EDGE project aims to develop and prototype an electronics platform for highly integrated medical edge intelligence for multi-sensor capsule endoscopy and neuroimplants.

Modern medical devices, with their tight integration of programmable microelectronics, sensors, and actuators, have opened fundamentally new diagnostic and therapeutic possibilities. The logical evolution to cloud-networked cyber-medical systems offers revolutionary opportunities for intelligent, careful, and pinpointed medicine by using AI techniques. However, for medical implants or wireless sensor technology, broadband and uninterrupted network connectivity is technically and practically unfeasible. The solution to these challenges is the integration of intelligence and autonomy directly in the medical device, i.e. medical edge computing. For the deployment of machine learning in such cyber-medical edge devices, there is a high demand for high-performance embedded hardware architectures that can execute machine learning for intelligent sensor data processing in an energy-efficient manner without violating application-specific requirements in terms of latency and power consumption. An important prerequisite for this goal is the provision of intelligent edge components that can be embedded in their environment in a well-tailored manner and interact with it autonomously. This requires application-specific AI hardware accelerators that can be flexibly adapted to the specific requirements of medical applications for the analysis and classification of sensor data streams. In this context, the machine learning methods are to be developed and optimized together with the hardware AI accelerators in an automated HW/SW codesign in order to meet the required performance characteristics. This enables efficient intelligent sensor signal processing with an average electrical power consumption well below 1 mW.

Funding

The project M/EDGE is funded by the Federal Ministry of Education and Research (BMBF).

Participating Team Members

Bringmann, Oliver

Gerum, Christoph

Reiber, Moritz

Werner, Julia