The 2AI (Algorithm Architecture Interactions) team designs original architectures for the integration of specific algorithms in information processing and contributes to the definition of these algorithms by early integrating of the architecture (whether hardware architectures or deep neural networks) constraint. It interacts and complements the other teams of the T2I3 department with the integration dimension and efficient implementations. The targeted approach is based on the combination of skills in algorithmic and architecture design.
Two fields are considered: digital communications, in close interaction with the other teams of T2I3, and artificial intelligence.
Similarly, two classes of architectures are explored depending on the application considered, the constraints imposed and the targeted integration technology. The first concerns the proposal of highly optimised architectures in terms of throughput, energy or area, while the second explores new architectural models and design methodologies to meet the challenges of flexibility, dynamic adaptability and the processing of complex or irregular signals.
In this context, the 2AI team has a complete development environment with all the necessary expertise, tools, platforms and libraries for (a) algorithm definition, simplification and optimisation, (b) architecture exploration, (c) hardware/software implementation and (d) validation and demonstration on hardware platforms.
These expertises, tools, platforms and libraries concern ASIC and FPGA technologies as well as several architecture models and associated design flows: multiprocessor, GPU, ASIP, NoC, HLS. They allow the team to develop and implement proofs-of-concept, often through embedded and heterogeneous systems. Finally, the algorithm-architecture interaction approach takes on its full meaning thanks to a real and internationally recognised dual algorithm/architecture competence available within the 2AI team and the T2I3 department.
Keywords: algorithm-architecture interaction, digital communications, AI, deep learning, hardware/software architectures, application-specific architectures.