Research Area:
Responsible:
Research Lines:
Project abstract
The OCTOPUS project, funded by the Italian Fund for Applied Sciences, proposes a paradigm shift in the design of analogue integrated circuits, a field historically characterized by manual, iterative, and highly time-intensive processes.
The main objective is the synergistic integration of advanced Machine Learning (ML) techniques into the EDA (Electronic Design Automation) design flow, in order to create an Artificial Intelligence-based virtual assistant capable of actively supporting microelectronics designers.
The project is structured around three methodological pillars aimed at significantly optimizing the time and cost required to develop analogue chips, while standardizing processes that have so far relied on the designer’s intuition.
- AI-driven automatic layout: Development of deep learning algorithms capable of geometrically generating the circuit on silicon, optimizing component placement and routing, and minimizing parasitic effects.
- Simulation acceleration: Overcoming computational bottlenecks by replacing traditional, costly SPICE simulations with performance estimators based on predictive ML models. These fast surrogates will enable near-instantaneous evaluation of circuit behaviour, accelerating the optimization phases.
- Technology retargeting: Definition of an innovative methodology based on Transfer Learning to automatically migrate an existing circuit, originally implemented in technology A, to a new technology B, while preserving its functional specifications.
