AI

SGAnalog: An End-to-End Circuit Benchmark from Open-Source Silicon Tapeouts

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 04:18:24 UTC.

Briefings

This article presents the SGAnalog benchmark, a collection of 273 open-source analog circuits from Tiny Tapeout shuttles. The benchmark evaluates schematic-to-netlist transcription and device sizing across seven models. Results show the strongest model achieves 56.1% exact graph isomorphism on 66 tasks, while the leading model reaches 91.2% on 17 sizing tasks. A key finding is that removing author labels improves structural F1 but reduces exact matches, indicating labels aid connectivity tracing. The benchmark exposes distinct visual and design capabilities among models, revealing a policy limit in one circuit family rather than pure capability.

Key points

• The SGAnalog benchmark consists of 273 topologically distinct analog designs from Tiny Tapeout shuttles, with each circuit retrieved at its revision for processing.
• On 17 sizing tasks, the leading model achieved 91.2 out of 100 accuracy against human reference, while two newer models refused 4 and 11 prompts without objection.

Context and limitations

The benchmark covers 273 circuits across seven models on 66 transcription and 17 sizing tasks. Results are limited to the specific revision dates of the source files and the author's testbenches. The study does not include human-designed circuits or external process conditions, and the 56.1% figure represents exact graph isomorphism rather than overall performance.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 06/10/2026, 05:14:54 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiaw3nzhjl6zeiwufocxdifljqe6p4gbn74eba3bejsgcnyomkwbc4