Synopsys has unveiled AgentEngineer, a suite of AI agents built on the Autopilot platform designed for long-term chip development tasks. The system covers six key areas, including chip and system verification, physical implementation, analog design, simulation, and analysis.
The architecture consists of three tiers: super-agents for coordination, executor agents for specific tasks, and tool-level engines for execution. The platform supports cloud-based, on-premises, or local deployment, allowing the use of commercial, open-source, or fine-tuned AI models.
During testing with Nvidia, chip verification speed increased by up to 50 times, with a 20% improvement in verification coverage. Fujitsu reported a 30% boost in engineering productivity for RTL code generation, while another client reduced AI token consumption by half.
The system is classified as L5 autonomy, the highest level in the Synopsys framework. While agents operate autonomously, human engineers retain control, setting checkpoints to review decisions and adjust intervention levels as needed. The product is currently used by over 50 clients.
