High performance and high energy efficiency, designed to scale.

Weight storage and compute units interleaved in a dense layout.

Accumulation in full precision, with minimal loss of neural-network accuracy.

Technology commensurate with CMOS scaling to low lithography nodes.

Complete compiler and software stack based on RISC-V.

The AI Processing Unit (AIPU) is DARE’s AI inference chiplet, led by Axelera AI. It is designed for high performance, high energy efficiency and scalability, and builds on Axelera AI’s digital in-memory computing (D-IMC) technology.

Together with the General Purpose Processor (GPP), the AIPU forms a complete System-in-Package: AIPU chiplets handle AI inference, GPP chiplets run classical HPC workloads. A complete compiler and software stack is developed on the RISC-V architecture and Axelera AI’s infrastructure.

Speeds up AI within scientific workloads such as weather models, materials discovery, search and physics-informed neural networks (PINNs).

Built for large-scale neural networks, including generative and foundation models.

a

Alya and Bettair (air quality), large language models and SuSy (synthetic content detection) are being optimised for GPP + AI.