Next Generation Computing in the Edge

How can computing be managed close to the sensor rather than in the cloud in the future? And how can machine learning be implemented on distributed systems in such a setup? Researchers at Fraunhofer EMFT and eleven other Fraunhofer institutes are addressing these highly topical issues in the SecLearn Arrival innovation project. The focus here is on neuromorphic, energy-efficient hardware components and AI algorithms for decentralized learning as well as data protection.

Neuromorphic Computing
© Fraunhofer IGD
Neuromorphic Computing

Today's von Neumann-based computer architectures have enormously high power consumption, so that a massive extension of computing to the edge would not make sense.

The researchers now aim to develop neuromorphic accelerators with orders of magnitude lower power consumption and optimize them for AI algorithms as part of the project. At a later stage, two use cases will be implemented based on this hardware: (I) speech recognition (keyword spotter + audio event detector) and (II) image recognition (automotive or autonomous driving). Here, the machine learning is to take place in the distributed systems without the basic data having to be given to the central cloud. In this way, sensitive data can remain in the local systems and data protection is better ensured.

The work is funded in-house as a Fraunhofer lead project. 


You might also be interested in:

Research Area: Neuromorphic Computing

IC-design and layout

Circuit Design for Neuromorphic Computing

Project: Energy Efficient Neuromorphic Computing