At its heart, the project is about breaking down one of humanity’s oldest challenges: language barriers. Current translation tools rely heavily on distant supercomputers and constant internet access. While useful, these technologies come with two significant drawbacks: privacy concerns and heavy energy consumption. Every spoken word must travel to a server, be processed, and then returned – which is neither seamless nor sustainable.
The FVLLMONTI project wants to change that. Their goal is to build ‘Edge AI’ devices, which are small, standalone gadgets capable of doing all the computational work themselves. For translation technology, this means a lightweight earpiece that could instantly convert speech from one language to another in real time, with no internet connection required.
How can such a tiny device achieve something so powerful? The answer lies in reimagining the computer hardware. Traditional chips operate in two dimensions – spreading their components across a flat surface. But speech-to-speech translation requires enormous amounts of data to move quickly between memory and processors, something that gets bogged down in flat, 2D designs.
The FVLLMONTI project’s technology utilises a new 3D computing architecture called the Neural Network Compute Cube – or N2C2 for short. By stacking processors vertically, the team can dramatically reduce the distance that data must travel, boosting speed and efficiency. Imagine skyscrapers replacing bungalows, allowing more functionality to be packed into less space. Early tests showed that arrays of these N2C2 units could process AI tasks dozens of times faster than conventional methods.
Building the N2C2 is a true European collaboration. Each step of the technology comes together across the continent. In Dresden, Jens develops the new transistor technologies, while in Toulouse, Guilhem works on vertical device fabrication. In Bordeaux, Chandak contributes to modelling, and in Lyon, Ian works on circuit design for the N2C2 units. In Vienna, Oskar focuses on 3D routing, while Giovanni in Lausanne and Jean-Luc in Bordeaux refine the hardware and software together. Coordinating all of this is Cristell in Bordeaux, making sure the pieces fit into one coherent whole.
This distributed approach means that expertise from many countries and laboratories is woven into a single device. Fabrication may happen in France, integration in Austria, software refinement in Switzerland – each step strengthening the whole.
While the project’s immediate aim is smooth, real-time translation, the implications go far beyond breaking language barriers. The innovations pioneered here, especially the 3D chip architectures, could reshape how AI runs on all kinds of small devices, from medical wearables to autonomous drones. For now, the dream of speaking into an earpiece and having your words emerge in another tongue is tantalisingly close.
