VITFOX - Vision transformers with ferroelectric oxides

The demand in computing resources scales faster than the processors performances. Therefore, there is a strong need for new architectures able to handle large amounts of data. Neuromorphic systems mimic the processing methods of the human brain by arranging non-volatile memory in large-scale arrays and storing synaptic weights in the memory. 

To contain the volume of data exchanged between devices and cloud, the deployment of AI at the edge is necessary. This requires energy-efficient AI hardware.

Goals of the project

The ViTFOX project will develop a ferroelectric-augmented intelligent semiconductor technology to demonstrate a Vision Transformer (ViT) with superb energy efficiency of > 50 TOPS/Watt to make an impact on AI-powered edge applications.

ViTFOX proposes advancements beyond the state of the art, at a TRL 4-5, in the whole value chain from materials and devices to heterogeneous and monolithic integration as well as design and simulation of the ViT circuits and systems.

Collaboration between the EU and Korea

ViTFOX is a collaborative project between European and Korean research organizations. The project aims to strengthen the leading position of EU and Korea in Hafnia-based Si-compatible ferroelectric electronics pioneered in Europe and significantly advanced by Korean researchers. 

The multidisciplinary character of ViTFOX will allow substantial progress beyond the state-of-the-art in the field of neural network architectures, proposing a unique framework for the co-optimization of hardware and software of complex networks.

Besidesd the technological collaboration, student exchange is also envisioned. Fraunhofer IPMS will offer Korean students training on the ViTFOX topics from device to compute-in-memory demonstration and will seek to extend the collaboration by signature of MoUs including student exchange to promote the international exchange.