PACIFIC - Photonic Accelerator ChIp For InferenCe

A new photonic chip uses amplified light to cram more data streams into a single beam, accelerating the most compute-heavy artificial intelligence tasks. The problem is that AI workloads—like training large language models or running real-time image recognition—are outpacing the performance of conventional electronic processors. Moving data between memory and processing units creates a bottleneck, wasting energy and slowing computation. This chip tackles that bottleneck directly. By using low-loss optical amplification and coarse wavelength division multiplexing, it squeezes many separate data channels into one light beam without signal degradation, allowing the chip to feed data to existing processors far faster than electronic alternatives. If successful, this chip could dramatically speed up AI inference—the process where a trained model makes predictions—without requiring a complete overhaul of current hardware. That matters for systems that already rely on AI but need faster responses: medical diagnostics scanning tissue samples, navigation systems recalculating routes, or energy grids balancing supply and demand in real time. The chip is designed to scale, meaning it could keep pace as AI models grow more demanding, closing the gap between what current hardware can deliver and what the market needs.