by Giovanni Grandi
Codasip participates in multiple RISC-V research & development projects co financed by the EU. This blog post will do a deep dive into one of these projects: NEUROKIT2E. This German consortium comprises a total of 25 partners around Europe. The project has received funding from the European Union HE Research and Innovation programme (via CHIPS JU) and as a German participant in this project, Codasip is supported by the Federal Ministry of Education and Research.
A novel open source deep learning platform
NEUROKIT2E is a 3-year, multi million Euros, research project collecting partners within commercial organizations as well as academia across several EU countries such as Austria, France, Germany, Italy and Netherlands.
The project is funded by the EU and aims to develop a novel Deep Learning Platform for Embedded Hardware. Most notable objective of this framework developed under the NEUROKIT2E project coordination CEA is the aim to provide an open source framework that could enable the utilization of synchronous coding (tensors) and event-driven coding (spikes) into a single end-to-end development chain. This framework will contribute to position EUROPE as a market leader in the embedded AI domain.
In order to achieve its goals the NEUROKIT2E project is based on a use-case driven approach, that is to say that the project outlined a total of seven major use cases that aim to demonstrate the capabilities and effectiveness of the proposed framework.
Codasip together with the other German partners (Infineon, Fraunhofer and University of Rostock) is focusing on the use case titled “Occupancy monitoring for efficient building control”. In this use case the German partners will implement an occupancy monitoring system based on radar sensor nodes. The node will include besides the radar sensor a Codasip RISC-V hardware for the processing of the radar data on the node itself. The CNN, DNN and SNN architectures, algorithms and corresponding processing pipelines for the RISC-V hardware will be explored and developed with the NEUROKIT2E project.
Codasip Labs is our esteemed innovation hub dedicated to pioneering new technologies and advancing future developments. In light of its forward-looking orientation, Codasip Labs is the primary department driving Codasip contributions to the NEUROKIT2E project as this is the naturally fitting endeavor for our team vision.
Custom Compute for Neuromorphic applications
Codasip’s role within the NEUROKIT2E project and specifically as contributors to use case 6 is to research, architect, design, and optimize custom RISC-V ISA (Instruction Set Architecture) extensions and other digital IP blocks. This exploration and prototyping work will be carried out using our Codasip Studio EDA tool and CodAL.
Codasip Studio is a unique collection of tools for fast and easy designing or modification of processors. It works with CodAL architecture description language, which is used to describe both the ISA and the microarchitecture.
The main idea is to select a low-power embedded RISC-V core that provides maximum energy efficiency with optional signal processing capability and enrich it further with novel custom instructions specifically tailored for the chosen application. Applying custom compute to our baseline cores will ensure that the execution of Spiking Neural Networks (SNNs) and/or a mixture of different types of Deep Neural Networks (DNNs) algorithms will fulfill the requirements of use case 6.
Beside contributing with processor and RISC-V specifically expertise, Codasip aims to leverage the research performed by academic partners in the consortium in order to promote knowledge sharing and boost cooperation needed to achieve the objectives agreed in the scope of NEUROKIT2E project proposal.
