PRIN 2022 PNRR - OPERAND

The OPERAND project addresses several themes belonging to the Strategic Emerging Topic: CIRCULAR ECONOMY: OPERAND’s very low-power AI and ML algorithmic acceleration will enable ubiquitous AI edge computing. The envisioned HW/SW ecosystem will leverage the Reduced Instruction Set Computer (RISC-V) open standard, with the very efficient and customizable Coarse-Grained Reconfigurable Array (CGRA) technology.

The six key revolutionary elements of OPERAND are:
 

1) Flexibility. The entire OPERAND HW structure will be developed in tech-independent Hardware Description Language (HDL), e.g.,
SystemVerilog or VHDL, widely adopted by the research community, enhancing open collaboration, and allowing customizability and interoperability at any level. OPERAND HW can be integrated into different devices, providing the flexibility to select the most suitable technology for the specific need and application.
 

2) Modularity. OPERAND will be built as an open ecosystem, relying on an open architecture and open interfaces, integrating different open IP cores, enabling interchange and interoperability of components. It will exploit the modular nature of CGRA fabric and its reconfigurability: it will be possible to reduce or increase the resources allocated (e.g., number of PE, memory available) for
each running application, according to the operating requirements, with very low programming effort.

3) Customizability. OPERAND long-term objective is an open-source but well-defined development framework where people can contribute (as done for RISC-V), completing the flexibility of CGRA with a Quantization framework and a dedicated task scheduler. Inrecent years, the frameworks for AI applications have simplified the development of these algorithms, making the developers’ life
very simple with a consequent reduction of the developing time and time-to-market.

4) Suitability for on-the-edge applications. OPERAND, thanks to its CGRA architecture, will facilitate the migration of computation traditionally done in the Cloud to the edge, allowing high parallelism, high performance, low power consumption and suitability for harsh environments.

5) Open HW/SW Co-Design Programming Framework. OPERAND's long-term goal is to build up an openly available HW/SW framework, which will be extensible to other application domains and algorithms, leveraging the synergy of the Open-Source approach and the successful strategy of RISC-V International, becoming the first complete open HW/SW ecosystem to efficiently accelerate AI models for all kinds of applications.

6) EU Technology Independence. OPERAND will provide a complete HW/SW system to accelerate AI algorithms of all kinds with completely European technology. The recent socio-economic situation (pandemic, chip crisis, Ukraine war) has highlighted how much Europe is dependent on non-EU technology. AI/ML HW available on the market has been developed by private companies, usually located in the USA with manufacturers in Asia. EU AI technology independence will be of strategic importance to overcome the limitations that can occur in the future with other pandemics, wars, and, in general, all situations where trade and transportation are adversely affected[6]. OPERAND will be completely developed in Europe and with European technology, including its FPGA overlay, getting rid of dependence on non-EU private companies.

ENTE PROMOTORE: MUR
NOME PROGETTO: OPERAND
PERIODO E DURATA: 24 mesi, 30/11/2023 – 29/11/2025
FINANZIAMENTO: € 110.400
COORDINATORI: Università di Pisa
REFERENTI SSSA: Alessandro Biondi