IKERLAN leads the European NimbleAI project to develop "the eyes" of IoT devices
NimbleAI will seek to replicate the energy efficiency of the human vision system, the result of millions of years of evolution, and create neuromorphic technology with high impact potential.
- The project will accelerate the design and commercialisation of new generations of neuromorphic chips, which can be applied in sectors such as IoT (Internet of Things), robotics, autonomous mobility, Industry 4.0, healthcare and virtual reality.
- The NimbleAI neuromorphic chip will consume almost a thousand times less energy than an LED bulb and will occupy an area of silicon about a quarter of a penny.
- With a budget of €10 million for the period 2022-2025, NimbleAI will help strengthen European semiconductor sovereignty and reduce the dependence of today's IoT devices on the cloud.
- NimbleAI will boost the competitiveness of Basque industry through the use of pioneering neuromorphic technology and will support the creation of intellectual property in the Basque Country, based on the neuroscience knowledge of the Basque Excellence Research Centres (BERCs).
Can artificial intelligence (AI) learn to 'see' like humans and do so by consuming extremely little energy? The IKERLAN technology centre thinks so, and that is why it leads the Horizon Europe NimbleAI research project, which aims to create 'the eyes' of Internet of Things (IoT) devices. Autonomous vehicles, Industry 4.0, healthcare, robotics and virtual reality are the main target market for NimbleAI technology, as they are established applications in which vision plays a key role.
The "eyes" that IKERLAN will be working on, together with other technology centres, universities and leading companies in the sector, will replicate on a silicon chip the functioning of the retina and the neurocognitive processes that enable human vision. These systems, the result of millions of years of evolution, are characterised by investing energy only in capturing and processing visual data with high information content and allowing to carry out tasks of great complexity in a safe way, such as driving vehicles. In the retina, photoreceptor cells react only to significant changes in light, and neurons in the brain that process visual information consume energy only when stimulated by photoreceptor cells in the retina or by other active neurons NimbleAI will adopt this biological operation, based 'on events' (activations of photoreceptors and neurons), which is completely different from that used by traditional processors. These process data continuously, regardless of the value of the information provided by the processed data, and therefore are inefficient.
NimbleAI will also use the latest advances in 3D chip manufacturing to integrate as many photoreceptors and artificial neurons as possible into a single volume of silicon, emulating the 3D structure of a biological brain.
Moreover, the chip designed at NimbleAI is expected to consume just a few tens of mW - almost a thousand times less than an LED bulb - and use a silicon area of just 50 mm2, about a quarter of the area of a penny coin. Xabier Iturbe, NimbleAI project coordinator and senior researcher at IKERLAN, says, "NimbleAI aims to design a neuromorphic chip that improves AI processing power efficiency and latency over current commercial technology by at least 100 and 50 times, respectively.
Impact on industry and research in Europe and the Basque Country
NimbleAI will contribute to strengthening European semiconductor sovereignty and position Europe at the forefront of the current revolution in neuromorphic technology, as a more efficient alternative to traditional processors. It will also help reduce the dependence of today's IoT devices on the cloud by enabling them to carry out the most advanced (and demanding) data processing locally. Avoiding transmitting data to the cloud will minimise security risks and reduce response times - a critical issue in applications such as autonomous driving.
Beyond the transfer of the results to the companies participating in the project in the short term, giving rise to new generations of commercial neuromorphic chips, in the long term, NimbleAI constitutes an opportunity for Basque industry and for research in the Basque Country.
NimbleAI has a budget of 10 M€ for the period 2022-2025 provided by the Horizon Europe programme of the European Union and the Government of the United Kingdom. Together with IKERLAN, which leads the project, the consortium includes reference research centres in the technological sectors relevant to the project (CEA, IMEC, CSIC, BSC), a company that produces software tools for 3D chip design (MZ Technologies), companies that license intellectual property for integration in chips (CODASIP, MENTA), a company that designs and markets neuromorphic processors (GrAI Matter Labs), a company that designs plenoptic cameras (Raytrix), prestigious universities (TU Wien, Polytechnic University of Valencia, Politecnico di Milano, University of Manchester, Queen Mary University of London, Leiden University) and final equipment manufacturers (ULMA Medical, AVL and ViewPointSystem).
The consortium is supported by a committee of advisors that includes the most important company in the design and commercialization of neuromorphic imaging sensors, Prophesee, as well as companies that cover the entire value chain of the semiconductor sector, including manufacturers of lithography equipment for production of chips and the chip producers themselves ('foundries'). The goal is to ensure that the intellectual property and technology generated in NimbleAI is manufacturable.
IKERLAN is a leading centre in technology transfer and in the contribution of competitive value to industry. It offers integral solutions combining different technological domains in two main areas: Electronics, Information and Communication Technologies (EICT), and Energy and Mechatronics. The technology centre's team is currently made up of 400 people. As a co-operative member of the MONDRAGON Corporation and the Basque Research and Technology Alliance (BRTA), IKERLAN maintains a sustainable, competitive business model in permanent transformation. IKERLAN currently has the following accreditations: ISO 9001:2015 Quality Management System, UNE 16600:2014 R&D&I Management System and ISO 27001:2013 Information Security Management System. This is recognition of the excellence of the centre in the field of R&D&I.