Four Institutions for Edge AI Excellence
The consortium was constituted to provide the necessary knowledge, as well as the infrastructure for project realization. The advanced partners have multidisciplinary expertise, high scientific quality and a strong background in specific areas of EdgeAI and are well known in their field of activity. They also have long-term collaboration with numerous partners from academia, research institutions and industry, and significant results in knowledge transfer, especially towards young researchers. Consortium involves highly qualified female researchers in research and management activities, respecting flexible working hours and family policies, thus fully complying with aspects of gender equality.
Faculty of Electronic Engineering in Niš (FEEUNI)
The coordination institution Faculty of Electronic Engineering in Niš (FEEUNI) operates as a public educational and research entity within the University of Niš (ranked among the three top universities in Serbia). FEEUNI has a tradition of 60 years in teaching at all study levels (B.Sc., M.Sc., Ph.D.), with more than 10000 graduated students and more than 300 PhDs. It also has experience in networking and cooperation with many scientific and business institutions. Through networking, FEEUNI published dozens of scientific papers in international journals every year. In cooperation with other institutions, FEEUNI also organizes several international conferences: MIEL, TELSIKS, PES, Student projects conference IEEESTEC. FEEUNI has participated in several FP7 and H2020 projects, in numerous projects within COST, ERASMUS, TEMPUS and NORBAS programs, as well as in many bilateral projects (with EU countries). In this project FEEUNI brings expertise in compression of ANNs using quantization, block-wise normalization and pruning, based on the statistical modeling of data and adaptation of quantizers themselves to the statistical characteristics of the input data.
Leibniz Institute for High Performance Microelectronics (IHP)
The Leibniz Institute for High Performance Microelectronics (IHP) has around 350 R&D professionals with core competence in material science, process technology, circuit design, wireless systems and system architectures. The research is focused on socially relevant topics such as communication, mobility, health, environment, industry, agriculture, and security. The IHP’s key differentiators are its cutting-edge SiGe BiCMOS technology for chip fabrication and RRAM manufacturing facility. IHP is active in research commercialization, and has initiated two start-ups: Silicon Radar and Lesswire. In addition, IHP has established a technology transfer company IHP Solutions focused on commercialization of IHP’s research results. IHP has strong collaboration with both German and European universities, providing excellent opportunities for training of young researchers. One of the main strengths of the IHP team is its fault tolerance design competence. IHP’s fault tolerant chips are available on the market, and two IHP’s chips are onboard ESA’s Jupiter Icy Moon Explorer, launched in April 2023. In this project, IHP participates with a team composed of researchers from System Architectures, which conducts the research on digital ASIC design, fault tolerance computing, modeling and design of AI systems (including RRAM).
University of Manchester (UoM)
The University of Manchester (UoM) is a public research university with a traditionally strong research, education and administration processes organization, ranked #32 in the QS World University Rankings for 2024, and ranked #8 in Europe. The School of Computer Science is one of the longest established schools of Computer Science in the UK and of the largest, with a very strong research history. Both the world’s first stored-program computer (the 1948 Manchester Baby) and the ground-breaking Atlas computer (the world’s most powerful computer at that time; 1963) were developed at University of Manchester. More recently, another historical milestone has been achieved by the Advanced Processor Technologies group (taking part in this project) through the SpiNNaker framework, delivering the world’s largest neuromorphic computing platform, which is currently going through its second generation of technology, implementation and architecture. Leveraging his strong background in brain-inspired computing, UoM is currently extending his focus to edge artificial intelligence through new hardware/software neuromorphic solutions, including analog mixed-signal designs, memristive accelerators, ultra-low power AER routing and hybrid SNN-ANN models. Therefore, UoM brings SNN expertise and silicon implementation knowledge to research and develop neuromorphic architectures that will communicate directly with an event-driven adaptive vision system, filtering noise and delimiting regions of interest. This will help limit the downstream processing required, reducing overall energy requirements whilst regulating the asynchronous stream of sensor events to find the optimal energy efficiency and accuracy trade-off.
University of Ferrara (UNIFE)
The University of Ferrara (UNIFE), founded in 1391, as one of the oldest in Italy, with more than 25.000 students and 653 permanent academic and 490 administrative and technical staff, has participated in 42 international research projects – FP7, 4 as Coordinator, 46 projects – Horizon 2020 (1 in Grant Preparation, 4 as Coordinator and 8 Individual Fellowship Marie Skłodowska-Curie), and several projects funded by other European research programs; and has around 1150 international cooperation agreements. UNIFE is a member of several International Networks, among them the ECSEL-ARTEMIS Industry Association and BBI Bio-Based Industries, UNIADRION, EFSA – European Food Safety Authority. Two research groups from UNIFE are involved in the AIDA4Edge project: Artificial Intelligence group which is internationally recognized as a pioneer and expert in the field of machine learning and Multiprocessor System-on-Chip (MPSoC) group that has a broad research scope on digital circuit design and embedded computing architectures. UNIFE excels in reducing AI model complexity through algorithmic optimization, particularly through hyperparameter optimization and dynamic adaptation of ANN models, considering confidence estimation.