AI Now Designs Light-Powered Computer Chips From Plain English

Think of a computer chip that uses light instead of electricity to send data. That's a silicon photonic chip, and it's key to making internet and data centers faster while using less power. The problem? Designing these chips is incredibly complex and usually takes expert engineers weeks of manual work. Now, researchers from several universities have created an AI framework called PICasso that can do much of this design work automatically—starting from a simple request written in plain English.

PICasso works like an AI architect that follows your instructions but also checks its own blueprints. You type something like "build a small, efficient light splitter," and the system turns that into a technical design, then runs physical checks and simulations to make sure it will actually work. This matters because standard AI models often produce designs that look good on paper but fail in the real world. PICasso avoids this by combining a large language model (AI that understands text) with engineering rules and virtual testing. It's like having a bridge designer who not only drafts the plans but also stress-tests the bridge before it's built.

In testing across 36 design tasks, PICasso produced structurally correct designs 92.7% of the time on complex circuits, compared to much lower scores from ordinary AI. More importantly, it reduced signal loss—a measure of how much light energy is wasted—from 4.98 dB down to 3.25 dB. That's a 35% improvement, which means chips that waste less energy and transmit data more reliably. The system also works at speeds competitive with manual design, but without needing a human expert at every step.

So why should you care? Silicon photonics is already used in data centers, telecom networks, and high-speed devices. If AI can design these chips faster and better, we could see cheaper data centers, lower electricity costs, and quicker development of future technologies like self-driving cars and advanced AI systems. It won't replace engineers overnight, but it's a big step toward machines designing the very hardware that powers our digital world.