SEC-Learn Forum on Neuromorphic Computing

SEC-Learn addresses highly topical and highly relevant issues of future computing at a time when the replacement of today's von Neumann-based processor architectures by new architecture concepts is foreseeable and the increasing shift of computing from the cloud to the edge requires the development of new, resource-conserving and energy-efficient platforms.

In order to be able to realise computationally intensive special tasks even in resource-limited edge applications, for example in mobile devices or vehicles, novel approaches from next-generation computing are being pursued in the SEC-Learn (Sensor Edge Cloud for federated learning) project. The project addresses the need by developing an innovative platform with neuromorphic accelerators - Spiking Neural Network (SNN) cores for energy-efficient AI applications and especially for federated (distributed) learning. Developed hardware components and the algorithms for federated learning will come together on a common platform to enable efficient implementation of inference and training.

The neuromorphic accelerators to be developed in this project have orders of magnitude lower power consumption and can be optimised for AI algorithms, enabling a shift of computing to edge devices. The scalability of the development results is demonstrated in two applications:

I. Speech recognition and acoustic event detection for voice assistants.

II. Image Recognition (automotive or autonomous driving)

The learning of the neuromorphic platform should take place in the distributed systems without having to process customer-sensitive data in the central cloud.