NEUROKIT2E

This Euro­pean Project, comprised of 25 part­ners, aims to provide an open-source and sover­eign plat­form of tools for Embedded AI with several ambi­tions:

• Posi­tion EUROPE as a market leader by providing tools capable of meeting real-time, data confi­den­tiality, energy consump­tion and usability require­ments.
• Provide a plat­form that will inte­grate hard­ware models with neural network models to opti­mize the network for embedded devices and provide a single end-to-end devel­op­ment chain.
• Develop advanced compres­sion and pruning methods to reduce model size while main­taining the perfor­mance of the orig­inal network.
• Enable the utiliza­tion of synchro­nous coding (tensors) and event-driven coding (spikes) to be combined into the same network.

SAL's primary research focus within Neurokit2E is on optimizing hardware IP for AI applications, improving existing hardware operators, and developing specialized accelerators through high-level synthesis. In doing so, SAL is working on novel processing units for more compact neural network implementations, developing supporting units for efficient data management, and creating parameterizable neural network architectures for performance and efficiency enhancement. Additionally, SAL is exploring AI optimization techniques such as quantization, pruning, and compression in the context of safety-critical applications.