Project Objective
The objective is to develop for commercialization an intellectual property (IP) framework aimed at the acceleration of artificial intelligence (AI), which will serve as the basis for the evolution towards a general-purpose AI chipset accelerator.
This development pursues a double purpose: 1) to consolidate an advanced technical solution and 2) to facilitate its initial adoption in the market through a two-phase introduction strategy.
The proposed IP framework will broadly include:
- Optimized and validated arithmetic units (AU) for the efficient execution of specific AI operations.
- Integration interfaces and stand-alone mechanisms, compatible with RISC-V-based architectures, including custom ISA instruction extensions.
- Software tool for model assignment and optimization, aimed at facilitating the efficient deployment of neural networks and AI algorithms on accelerator hardware.
As an evolution of this IP framework, the general-purpose AI accelerator will consist of at least the following hardware IP blocks, in addition to the peripherals required for integration into a System-on-Chip (SoC):
- Workload mapper module.
- Custom RISC-V instructions for AI applications.
- Internal data bus optimized for high efficiency.
- Floating point computing unit adapted to the requirements of current models.
The planned positioning for the product is focused on the segment of very low power accelerators, oriented to edge devices, with an architecture based on RISC-V whenever appropriate. The system will be optimized not only for AI inference tasks, but also for data acquisition, preprocessing and management functionalities in Internet of Things (IoT) applications.
Type of project: Funded by the European Union under the HORIZON-JU- Chips 2024 Innovation Actions scheme.
Project Name: Shift2SDV
Type of project: Funded by the European Union under the HORIZON-JU- Chips 2024 Innovation Actions scheme.
Role of S&C: S&C is responsible for the development of a decision support system for anomaly detection, integrated into the SDV software. This system will generate decisions and send them to the software components.
File number: –
