Toward Topology-Optimized Foundry PDKs: A Seeded Design Framework for Multimode Interferometers
This paper presents a foundry-compliant, end-to-end design framework that combines parameter optimization with seeded topology optimization to significantly enhance the performance and fabrication robustness of multimode interferometer-based photonic devices, as validated through both simulation and experimental fabrication on a commercial process.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of modern electronics, the silicon chip has long been the engine of progress, but a new frontier is opening where light, rather than electricity, carries information. This field, known as silicon photonics, aims to build circuits that guide light through microscopic channels etched into silicon, much like wires guide electrons. These light-based circuits promise to move data faster and with less heat than traditional electronics, a crucial advantage as our digital demands grow. However, building these circuits is not as simple as drawing a line; the light must be split, combined, or redirected with extreme precision. If the path is even slightly off, the signal fades or scatters. For decades, engineers have relied on established mathematical rules to design these light guides, creating shapes that are reliable and easy to manufacture. Yet, these traditional shapes have limits, often leaving performance on the table because they are bound by the simplicity of their original formulas. The challenge has been to find a way to push performance higher without making the devices so complex that factories cannot build them.
A team of researchers at the Georgia Institute of Technology has developed a new method to bridge this gap, creating a design process that starts with the known and evolves toward the optimal. They focused on a specific type of light-splitting component called a multimode interferometer, which acts as a junction where light waves overlap and interfere to create a desired output. Traditionally, these devices are designed using analytical models—mathematical descriptions of how light behaves in simple, predictable shapes. While these models are robust and easy to manufacture, they cannot always find the absolute best performance. On the other hand, a powerful computer technique called topology optimization can theoretically find the perfect shape by testing millions of variations, but it often produces designs that are too fragile or complex for real-world factories to build. The researchers' solution was to combine the two: they began with a solid, factory-approved design and then used the powerful optimization tool to refine it, ensuring the final result was both high-performing and ready for mass production.
The team tested this approach on three different types of light-handling devices: a simple splitter that divides light into two paths, a multiplexer that combines different light modes, and a splitter that separates light based on its polarization. For each device, they first created a standard design using the traditional mathematical rules. This initial design served as a "seed," a functional starting point that already worked well. They then fed this seed into their optimization process. Instead of letting the computer start from scratch with random shapes, which often leads to designs that violate manufacturing rules, the computer began with the seed and made small, iterative adjustments. It smoothed out sharp corners, tweaked the width of the light channels, and adjusted the spacing between parts, all while constantly checking that the design remained within the strict limits of what a commercial factory could produce. This process allowed the computer to explore a much larger space of possibilities than traditional methods, finding subtle improvements that the original formulas had missed.
The results of this method were measured on actual chips fabricated at a commercial foundry. For the simple splitter, the optimized version reduced the loss of light as it passed through the device. While the original design lost 0.20 decibels of signal, the new design lost only 0.14 decibels. In the world of light transmission, where every fraction of a decibel counts, this is a significant gain. The team also tested a more complex device that combines two different types of light modes. Here, the improvement was even more dramatic. The original design allowed only a fraction of the intended light to pass through, but the optimized version let significantly more light through, improving transmission from a loss of 2.79 decibels down to 1.01 decibels. Crucially, these improvements were not just theoretical; they were confirmed by physical measurements on real chips, proving that the designs were robust enough to survive the manufacturing process.
Beyond creating new devices, the researchers applied their method to improve existing designs that are already part of standard commercial toolkits used by engineers. They took a standard splitter and a more complex four-port splitter provided by a major foundry and ran them through their optimization pipeline. The results showed that even these established, reliable designs could be made better. The optimized four-port splitter, for instance, divided the light more evenly between its outputs, moving from a ratio of 0.524 to a near-perfect 0.506. Furthermore, the team simulated what would happen if the manufacturing process varied slightly, such as if the etching of the silicon was a tiny bit too deep or too shallow. The optimized designs held up better than the original ones, maintaining their performance even under these imperfect conditions. This suggests that the new method does not just squeeze out extra performance but also makes the devices more forgiving of the inevitable small errors that occur in any factory.
The work demonstrates that the best path forward for advanced photonic design is not to abandon the reliable, well-understood methods of the past, but to use them as a foundation for more sophisticated optimization. By starting with a design that is known to work and then refining it, the researchers avoided the pitfalls of trying to invent entirely new shapes from scratch, which often leads to designs that are impossible to build. Their approach offers a practical route to incorporating the benefits of advanced computer-aided design into the commercial chips that will power future technologies. As the team continues to refine these methods, they are looking toward even larger and more complex devices, aiming to bring the full power of inverse design to the standard components that engineers rely on every day. The goal is a future where the most efficient light-guiding structures are not just theoretical possibilities, but standard parts available in the toolkits of chip designers everywhere.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.