Our Vision

We are setting out to bridge the gap between simulation and fabrication. In research, you can tapeout hundreds of design variations, and only one needs to hit to write a paper. In industry, every device must work, every time. We believe that by combining deep learning with photonics expertise, we can create a new paradigm where designs perform as expected the first time.

We envision a future where photonic engineers focus on innovation rather than wrestling with opaque fabrication processes—where advanced design tools are accessible to everyone.

Born as a spinout from McGill University and the National Research Council of Canada (NRC), PreFab emerged from a simple observation: techniques revolutionizing fields like medical imaging and autonomous vehicles could transform photonics fabrication too. As photonics PhDs working on AI for design, we were inspired by how ideas from outside the industry could create step-changes when thoughtfully applied.

We knew that rule-of-thumb approaches and design rules didn't tell the entire story. Conservative designs still faced process variations. Aggressive designs would sometimes work out. There had to be a way to find the true capability of fabrication processes. That's what we set out to build.

Fabrication awareness

Traditional approaches—correction tables, design rules, accumulated expertise—can't fully capture the complexity of modern fabrication. Process variations are too nuanced, too context-dependent. Machine learning can learn these patterns directly from data.

We use computer vision to model how designs actually turn out when fabricated. This virtual representation lets designers incorporate real-world manufacturing effects directly into their workflow. Our deep learning system accurately predicts and compensates for fabrication variations, helping your designs work right the first time.

Our fabrication-aware design approach enables:

• Accurate prediction of how designs will manifest physically on chips
• Verification of device performance before costly fabrication runs
• Automatic correction of designs to compensate for known fabrication effects
• Fabrication-aware inverse design (FAID) that incorporates true manufacturing constraints directly into the optimization process

With virtual fabrication, designers can create manufacturable designs that approach theoretical performance limits while reducing the need for manual iteration and repeated fabrication runs.

Modern software design principles

• Python-first approach enabling seamless integration with simulation and layout tools
• Simple, intuitive API that allows predictions and corrections in just a few lines of code
• Cloud acceleration delivering results in seconds
• Modular architecture supporting various fabrication processes and device types
• User-centric design built to suit the needs of the engineers that use it

Democratizing advanced photonics design

We want photonics designers to focus on what they do best—pushing the boundaries of what's possible—not wrestling with fabrication processes or building bespoke correction systems from scratch.

PreFab is starting with fabrication-aware design, but our vision extends further. We're working toward a future where advanced photonics design is accessible to everyone, where the tools are as intuitive as they are powerful, and where the gap between idea and working device shrinks from months to days.

— The PreFab Team