TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

chips infrastructure

The Chip Agent Worked Better Above RTL

A two-stage high-level synthesis workflow beat direct RTL design on eleven FPGA tasks.

Published Updated Story ID: mp-2026-09-21-003
Read the complete editionStory JSON

Summary

A two-stage high-level synthesis workflow beat direct RTL design on eleven FPGA tasks.

Researchers compared several ways to let language-model agents design FPGA circuits. Their combined method first designed in high-level synthesis, then refined the generated RTL. Across eleven benchmark tasks, it achieved a 2.6-fold geometric-mean speedup over direct RTL design. The authors attribute the gain to reusable high-level abstractions followed by low-level optimization. This is an FPGA benchmark, not evidence of fabricated production chips.

Why it matters

A two-stage high-level synthesis workflow beat direct RTL design on eleven FPGA tasks.

Limits and context

  • This is an FPGA benchmark, not evidence of fabricated production chips.

Key claims

  1. A two-stage high-level synthesis workflow beat direct RTL design on eleven FPGA tasks.

    Qualification: This is an FPGA benchmark, not evidence of fabricated production chips.

    Evidence: source-2026-09-21-003

Sources

  1. arXiv preprint 2609.21157arXiv · primary research

Corrections

No corrections have been recorded for this story.