Breker Hosts an Energetic Panel on Spec-Driven Verification

Key Takeaways

• AI can extract spec changes with 90‑95% accuracy.
• Multi‑agent critique still experimental, not production ready.
• Verification engineers become “verification architects” with AI assistance.
• Trust built via human‑checkable checkpoints, not end‑to‑end automation.
• Breker‑Moores Lab partnership launches first commercial AI verification tool.

Summary

At DVCon 2026, Breker hosted a panel on AI‑driven, spec‑driven SoC verification featuring leaders from Moores Lab, Breker, Arm, and Intel. Participants highlighted that AI can translate raw specifications into intermediate representations with roughly 90‑95% accuracy, but emphasized the need for iterative validation to avoid hallucinations. The discussion underscored that AI will augment, not replace, verification engineers, shifting roles toward verification architects. Breker and Moores Lab announced a commercial partnership to deliver the first AI‑powered verification solution.

Pulse Analysis

Spec‑driven verification has long been a bottleneck in semiconductor design, as specifications evolve throughout the product lifecycle. Traditional methods rely on manual diffing and extensive review cycles, which are error‑prone and time‑consuming. AI offers a way to automatically parse and highlight relevant changes, enabling designers to shift left earlier and maintain alignment with customer requirements. By converting raw specifications into structured artifacts such as opcode tables or flow diagrams, AI reduces the cognitive load on engineers and accelerates the creation of test benches.

The DVCon panel revealed that current AI models achieve 90‑95% correctness on first‑pass translations, but they still suffer from hallucinations and over‑confidence. Practitioners recommend a multi‑step verification loop: repeat queries, compare multiple agent outputs, and apply human‑in‑the‑loop feedback to converge on accurate results. Trust is established through granular checkpoints—outputs that engineers can readily inspect, like property specification sheets—rather than end‑to‑end code generation. While multi‑agent critique systems show promise, the technology is not yet mature for full production deployment, prompting vendors to focus on hybrid workflows that blend AI speed with human expertise.

The broader market impact is significant. Shorter design cycles driven by AI can lower time‑to‑market and cost, addressing the relentless demand for new semiconductor devices. As AI handles routine verification tasks, engineers are expected to evolve into verification architects or scientists, focusing on strategy, model training, and metric definition. Breker’s partnership with Moores Lab marks the first commercial rollout of such a solution, signaling industry confidence and setting a benchmark for future AI‑enabled verification tools. Early adopters who integrate systematic performance metrics and continuous human oversight will likely gain a competitive edge in the fast‑moving chip ecosystem.