HighTide: An Agent-Curated Open-Source VLSI Benchmark Suite

2026-03-09 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Expert, extended

Summary

HighTide is a new, AI-assisted open-source VLSI benchmark suite designed to address the limitations of existing static benchmarks. It offers a diverse collection of designs spanning multiple hardware description languages (Verilog, SystemVerilog, Chisel, Python-based generators) and three technology platforms: ASAP7 (7 nm), NanGate45 (45 nm), and SkyWater 130nm (130 nm). The suite features Bazel-based incremental RTL-to-GDS compilation with remote caching and incorporates twelve Claude Code agent skills for design curation, flow optimization, tool reference, and meta-maintenance. These agents utilize per-design decision logs as long-term memory, ensuring tuning rationale propagates across the suite. HighTide significantly reduces the RISC-V CPU share from approximately 59% in older suites to 30%, allocating more to ML accelerators, NoCs, and GPGPUs, with cell counts ranging from under 20k to over 1.5M. Its infrastructure includes RTL compilation verification for stable releases, ensuring it evolves with the open-source hardware ecosystem.

Key takeaway

For ML engineers and EDA researchers developing or benchmarking VLSI tools, HighTide offers a critical, continuously updated dataset. You should integrate HighTide into your evaluation workflows to train models on diverse, contemporary designs, avoiding overfitting on stale data. Its AI-assisted curation and broad coverage across HDLs and technology nodes (7 nm, 45 nm, 130 nm) ensure more generalizable results. This approach will lead to more reproducible outcomes for your next-generation hardware designs.

Key insights

HighTide uses AI agents and a dynamic infrastructure to maintain a diverse, evolving VLSI benchmark suite.

Principles

• Benchmarks must evolve with hardware.
• AI agents can automate design curation.
• Cross-design memory improves tuning.

Method

HighTide employs Claude Code agents with twelve skills for design discovery, integration, cross-platform porting, upstream tracking, debugging, PPA optimization, and meta-maintenance, backed by per-design decision logs.

In practice

• Use Git submodules for versioned snapshots.
• Implement remote caching for build artifacts.
• Track upstream tool bugs with workarounds.

Topics

• VLSI Benchmarking
• AI Agents
• Open-Source Hardware
• RTL-to-GDS Flow
• EDA Tools
• SoC Design

Code references

• VLSIDA/HighTide
• ABKGroup/FakeRAM2.0
• chipsalliance/rocket-chip
• bespoke-silicon-group/bsg_fakeram
• antonblanchard/microwatt

Best for: AI Scientist, Research Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• First Take on CadenceLive and Its AI Agent Stacks for EDA — Cadence's agentic AI EDA stack uses a hierarchical orchestrator and super agents to accelerate system design.
• NVIDIA Levels Up Local AI Agents Across RTX PCs and DGX Spark — NVIDIA is driving secure, high-performance local AI agent adoption across its RTX and DGX hardware and software platforms.
• Accelerating on-device AI: A look at Arm and Google AI Edge optimization — Arm SME2 and Google AI Edge enable efficient, high-quality on-device generative AI inference on CPUs.
• The foundation for a governed agent workforce: DataRobot and NVIDIA RTX PRO 4500 — The NVIDIA RTX PRO 4500 provides a balanced hardware foundation for enterprise AI agent inference.
• Custom Kernels for All from Codex and Claude — Coding agents can autonomously generate optimized CUDA kernels and integrate them into complex ML frameworks.
• DeepSeek-V4: The Most Powerful Open-Source Model Ever — DeepSeek-V4 commoditizes high-reasoning AI with open-source models, massive context, and disruptive pricing.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.