PreFab provides a data‑driven design platform for photonic devices that predicts post‑fabrication geometry and incorporates those predictions directly into optical simulations.
Funding
Funding not disclosed
Founders
Product
Problem
Photonic designers often rely on idealized simulations that ignore systematic and random variations introduced during chip fabrication, leading to performance mismatches, costly re‑fabrication cycles, and uncertain yield.
Solution
PreFab offers a data‑driven design platform that predicts the post‑fabrication geometry of photonic structures using deep‑learning models trained on real process data. The predicted geometry can be fed directly into optical solvers, allowing engineers to evaluate realistic device performance before tape‑out. An automatic corrector model pre‑distorts the input layout to counter known fabrication biases, so the manufactured device matches the target specification. The platform also quantifies device‑to‑device variability, highlighting regions of uncertainty to inform yield and robustness assessments. All functionality is exposed through a Python‑first API that integrates with common photonic layout and simulation tools, and results are delivered via cloud acceleration in seconds.
Target Audience
Primary users are photonic engineers and researchers in academia and industry who develop integrated photonic circuits and require accurate performance prediction and yield estimation before committing to fabrication.
Features
- Deep‑learning model that predicts post‑fabrication geometry and associated uncertainty for arbitrary photonic designs
- Automatic pre‑compensation of layouts to offset systematic process biases, enabling “design‑for‑fabrication” without manual correction tables
- Seamless integration with optical simulation frameworks by supplying predicted geometry for realistic performance analysis
- Fabrication‑aware inverse design (FAID) that incorporates true manufacturing constraints into optimization loops
- Python‑native API with a concise, intuitive interface that works with standard photonic design and layout tools
- Cloud‑based inference engine delivering predictions and corrections within seconds
- Modular architecture supporting multiple foundries and device types, allowing easy extension to new fabrication processes