DE-TW-FeEdge
Novel Compute-in-Memory Modules for Energy-Efficient Edge AI
DE-TW-FeEdge develops novel Compute-in-Memory modules for energy-efficient edge AI applications in a German-Taiwanese partnership. Compute-in-Memory concepts reduce data transport and thereby significantly lower the energy consumption of AI inference.
Objectives and Approach
The goal is to develop a computing accelerator for AI-based keyword recognition that performs computations energy-efficiently in memory. Ferroelectric field-effect transistors store neural network parameters non-volatilely for on-demand activation. Recurrent spiking neural networks enhance energy efficiency for event-based data processing.
Innovations and Perspectives
The targeted AI computing accelerator based on ferroelectric field-effect transistors represents the core innovation. The project strengthens German-Taiwanese collaboration in microelectronics, expands chip design expertise, and promotes the training of young professionals.
Project Coordinator
- Fraunhofer-Institut für Photonische Mikrosysteme IPMS, Dresden
Project Partners
- National Cheng Kung University, Tainan
This project is funded by the German Federal Ministry for Research, Technology and Space (BMFTR) as part of the Design Initiative Microelectronics.
