Neurophos, an Austin-based photonics startup, raised $110 million to develop optical processors designed to accelerate artificial intelligence inferencing while reducing power consumption. The company, a spin-out of Duke University and Metacept, an incubator run by Duke professor David R. Smith, is leveraging advances in metamaterials to create metasurface modulators that function as tensor core processors. These processors are designed to perform matrix vector multiplication, a core mathematical operation in AI, more efficiently than traditional silicon-based GPUs and TPUs.

Neurophos claims its optical processing unit, by fitting thousands of modulators on a single chip, offers significantly faster performance compared to current silicon GPUs used for AI inferencing. The technology aims to address the growing challenge of scaling computing power for AI applications without a corresponding surge in energy usage, a critical concern for AI labs and hyperscalers.

The underlying technology stems from research into metamaterials, artificial composite materials that manipulate electromagnetic waves in unconventional ways. Smith, a Duke University professor, demonstrated the potential of these materials two decades ago by creating a rudimentary invisibility cloak, which, while limited in its capabilities, showcased the potential of metamaterials to control light. This early work laid the foundation for Neurophos's approach to optical computing.

AI inferencing, the process of applying a trained AI model to new data to make predictions or decisions, is a computationally intensive task. Currently, specialized GPUs and TPUs handle this workload, but their power consumption is a growing concern. Neurophos's optical processors offer a potential alternative by using light instead of electricity to perform calculations, potentially leading to significant energy savings.

The implications of more efficient AI inferencing extend beyond data centers. Faster and more energy-efficient AI could enable new applications in areas such as autonomous vehicles, personalized medicine, and real-time language translation. However, the increased accessibility of AI also raises societal questions about bias, privacy, and job displacement, issues that require careful consideration as the technology matures.

Neurophos is currently focused on developing and commercializing its optical processing unit. The $110 million funding round will be used to scale up production, refine the technology, and expand the company's team. The company aims to position itself as a key player in the future of AI hardware, offering a more sustainable and powerful alternative to traditional silicon-based processors.