NVIDIA HORIZON: A Hands-Free Agent that Evolves Git Worktrees and Hits 100% RTL Benchmark Completion
NVIDIA introduces HORIZON, an AI agent that autonomously solves RTL verification tasks by managing Git worktrees and achieving 100% benchmark completion.
- HORIZON is the first AI agent to achieve 100% completion in RTL verification benchmarks by evolving Git worktrees autonomously.
- The framework uses Git's version control to manage incremental solution paths, enabling traceable and reversible design iterations.
- NVIDIA's entry into RTL verification automation signals a shift toward AI-driven EDA tooling in semiconductor development.
- The system targets hardware design bottlenecks, potentially reducing manual oversight and accelerating chip development cycles.
NVIDIA has unveiled HORIZON, an AI agent framework designed to automate register-transfer level (RTL) verification tasks. The system treats each verification problem as a versioned Git repository, allowing the agent to evolve solutions incrementally while maintaining traceability. In benchmark tests, HORIZON achieved 100% completion across standard RTL verification tasks, a milestone previously unattained by automated systems.
The framework leverages Git's branching and merging capabilities to manage multiple solution paths simultaneously, enabling the agent to backtrack, refine, and optimize designs without manual intervention. This approach contrasts with traditional RTL verification methods, which often require extensive human oversight and iterative debugging. NVIDIA positions HORIZON as a breakthrough for hardware design workflows, particularly in complex chip development where verification bottlenecks are common.
The announcement highlights NVIDIA's expanding role beyond graphics and AI accelerators into core EDA (Electronic Design Automation) tooling. By integrating AI-driven automation into RTL verification, the company aims to reduce development cycles and improve reliability in semiconductor design.
Source: NVIDIA HORIZON: A Hands-Free Agent that Evolves Git Worktrees and Hits 100% RTL Benchmark Completion. Read the full piece at the source.
Provides an AI-driven tool for automating RTL verification, reducing manual effort in hardware design workflows.
Offers a competitive edge in semiconductor development by accelerating verification and improving reliability.
Highlights NVIDIA's expansion into EDA tooling, a high-value segment adjacent to its core AI and GPU businesses.
Demonstrates how AI agents can tackle complex, traditionally human-intensive tasks in engineering disciplines.
- RTL
- Register-Transfer Level, a design abstraction used in hardware description languages to model digital circuits.
- EDA
- Electronic Design Automation, software tools used to design and verify integrated circuits.
- Git worktrees
- A Git feature allowing multiple branches to be checked out simultaneously in separate working directories.
AI’s Volatile Power Use Quietly Tests Grid Limits
Anthropic is discussing a new custom chip with Samsung
Anthropic reportedly explores custom chip manufacturing with Samsung while insisting Nvidia still matters
