Project description
Wireless solutions for chiplet challenges
The rise of domain-specific applications, such as AI, and surging chip manufacturing costs have propelled a shift towards chiplet-based computing system designs. Yet, current network-on-package (NoP) infrastructures struggle to handle the growing number and communication needs of diverse chiplets, such as CPUs and AI accelerators. This bottleneck hampers computing progress. The ERC-funded EWiC project will tackle this challenge by pioneering wireless communication within chiplet-based computing systems. Building on promising simulations, EWiC aims to experimentally validate wireless links combined with high-bandwidth wired connections, enabling >5× speed-ups. The project could revolutionise many industries, from accelerating drug design to boosting EU competitiveness in advanced computing. EWiC paves the way for scalable, high-speed computing solutions with transformative societal and economic benefits.
Objective
Domain specialisation (e.g. AI) coupled with skyrocketing manufacturing costs of chips led to a paradigm shift towards chiplet-based computing. However, the current Network-on-Package (NoP) is not suited for the increasing chiplet (CPUs, AI accelerators, etc) count and diverse communication needs. As a result, communication has become the main bottleneck to computing advances.
Wireless transmission can enable one-hop communication with low latency parallelism. The wireless links can be reconfigured dynamically for demand-specific services. Finally, the native broadcast capability of wireless transmission will provide natural support for scalability. However wireless communication is limited by the bandwidth. Hence, the ERC Starting Grant WINC explored via simulation combining wireless communication with the high bandwidth of wired connections, proving potential for >5 speedups.
In EWiC we aim, for the first time, to emulate and demonstrate wireless communication among chiplets inside a computing system. The goal is to experimentally validate our simulation results, which have shown high speedups in multi-chiplet systems. Successful results will generate interest in the industry for further research and application in various domains, such speeding up computation in pharma for new drug design. EWiC will thus help fully realise the trillion-euro potential of advanced computing, alongside immense social benefits, and establishing EU leadership in this field.
The EWiC project will thus assess the commercialization potential of our technology, engaging with key stakeholders such as chipmakers and end-users. Exploitation options including licensing, startup creation, or joint ventures, will be explored based on prototyping results and market entry considerations. Our IPR strategy will involve a thorough analysis to identify and protect the novel results of the project. We will additionally carry out a Freedom to Operate (FTO) analysis to ensure exploitability.
Programme(s)
- HORIZON.1.1 - European Research Council (ERC) Main Programme
Funding Scheme
HORIZON-ERC-POC - HORIZON ERC Proof of Concept GrantsHost institution
08034 Barcelona
Spain