Bridging the Sim-to-Real Gap in Semiconductor Visual Program Synthesis via Input Binarization
Yusuke Ohtsubo 1 , Kota Dohi 1 , Koichiro Yawata 1 , Koki Takeshita 1 , Tatsuya Sasaki 1 · 2026
Read Bridging the Sim-to-Real Gap in Semiconductor Visual Program Synthesis via Input Binarization by Yusuke Ohtsubo 1 , Kota Dohi 1 , Koichiro Yawata 1 , Koki Takeshita 1 , Tatsuya Sasaki 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02434.txt.
Description
Ohtsubo and co-authors tackle a data shortage in chip inspection. Precise control of circuit shape matters for checking semiconductors. Real training images cost a great deal to gather. Generative models can add data, but not with nanometer-level accuracy. The team turns inspection images into editable code instead.
Their framework uses a vision-language model. It reads an inspection image and writes domain-specific language code. The code describes the circuit geometry in full. Parameters can then be changed by hand for exact new samples. A snag remains, though. The model trains only on synthetic renders. Real electron microscope images differ in texture and noise. A binarization step strips that texture out. On the MIIC dataset, the mean Dice score rises from 0.4393 to 0.5256 over the raw-input baseline. A simple step closes much of the sim-to-real gap.
Catalog record
| CREATOR | Yusuke Ohtsubo 1 , Kota Dohi 1 , Koichiro Yawata 1 , Koki Takeshita 1 , Tatsuya Sasaki 1 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02434