asicdesign.ai

ASIC workflows, VLSI learning, and chip design tooling. Powered by AI

Why this hub exists

Verification is still the schedule sink. Public LLM-for-EDA progress is real, but the stack is fragmented.

The case for this hub is practical, not fashionable. Recent industry and academic work points in the same direction: verification still consumes a huge share of chip schedules, public LLM-for-EDA results are now meaningful, and the usable public stack is still spread across narrow point solutions instead of one mature, reviewable operating model.

  • Verification pressure is measurable: the 2024 Siemens EDA / Wilson Research functional verification report puts first-silicon success at just 14%, and the 2026 ASPDAC verification survey describes verification as consuming up to 70% of the development cycle.
  • Public progress is real: Ask-EDA reports large recall gains from hybrid RAG, ChipNeMo shows that domain adaptation materially improves hardware-assistant performance, and the VerilogEval revisit pushed the best public spec-to-RTL pass rate to 63%.
  • The public research shelf is wider than it looks: recent surveys now catalog named systems for assertion generation, testbench and stimulus generation, RTL debugging, and collaborative verification.
  • The missing layer is integration: engineers and students still need a place that turns papers, repos, and tool demos into reviewable workflows, grounded learning, and reproducible practice.
Research snapshot

Public LLM-for-EDA work is broadening, but mostly as focused sub-systems rather than one stable workflow layer.

Industry pressure 14%

First-silicon success in the 2024 Siemens EDA / Wilson Research verification report.

Benchmark progress 63%

Best reported public spec-to-RTL pass rate in the 2025 VerilogEval revisit.

Representative public verification systems named in the 2026 survey
Collaborative verification 11 systems
Assertion generation 10 systems
RTL debugging 6 systems
Testbench and stimuli 5 systems

Counts reflect the named systems in the taxonomy figures of the ASPDAC 2026 LLM-assisted circuit verification survey. They show breadth across sub-problems, not a census of all hardware-AI work.

Ecosystem

Learning makes AI useful. Guardrails keep it safe. Open-source tools make it real.

01

Learn the hardware ground truth

Start with RTL, timing, verification, architecture, and physical design so the model has a human-reviewed baseline to work against.

02

Use AI with controlled boundaries

Ask the model to draft, summarize, and transform, then keep the result inside diffs, scripts, tests, and review gates.

03

Hand validated work to the open stack

Move the result into reproducible open-source tooling, where Yosys, Verilator, GTKWave, and the rest of the ecosystem can exercise it.

Verilator Yosys GTKWave OpenROAD OSS CAD Suite Icarus Verilog