Austin, Texas, USA - In a quantum leap toward transforming the future of data centers, Neurophos, the brainchild of Duke University and Metacept, announced on December 14 that it has successfully raised a substantial $7.2 million in seed funding. This infusion, led by Gates Frontier and backed by MetaVC, Mana Ventures and other visionaries, marks the beginning of a revolutionary step forward in the development of metamaterials and optical AI inference chips.
Neurophos joins forces with Silicon Catalyst
The strategic alliance with Silicon Catalyst's incubator program accelerates Neurophos' path to product development. Located in the thriving tech hub of Austin, Texas, the company is poised to recruit a team of engineers to advance its vision of metamaterial-based optical AI chips.
The Optical vs. Digital Saga: Neurophos' Revolutionary Approach
In the realm of AI acceleration, GPUs have been champions, but their voracious power consumption has been a bottleneck. Neurophos departs from the digital norm by exploring the potential of optical computing. Photonics proponents herald it as the key to unlocking unprecedented computing speeds while minimizing power consumption.
Metamaterials: Unleashing creativity to control light
One hurdle in the optical computing landscape has been the bulkiness of existing devices. Enter metamaterials, the architects of a new optical paradigm. These materials, with their ability to control the flow of light, have unleashed a creative explosion that has produced innovations such as invisibility cloaks and negative refractive index materials.
Neurophos' Energy Efficient Metasurfaces
Neurophos' optical metasurfaces are designed for data centers and are already setting records for computational energy efficiency. The company's strategy is to use high-speed silicon photonics to power an in-memory metasurface processor, promising fast and efficient AI computations.
As the exponential growth of AI inference workloads threatens to overwhelm global electricity demand, Neurophos is stepping in as a champion of sustainability. The technology promises improved compute efficiency, delivering more computing power per dollar while maintaining energy consumption levels. This, in turn, aims to reduce the total cost of ownership for AI accelerator chips and data centers.
Patrick Bowen's Vision: Scaling the Optical Processor Landscape
Patrick Bowen, CEO of Neurophos, emphasizes the critical role of scaling in optical processors. The larger the processor, the more exponentially faster and more energy efficient it becomes on a per-operation basis. Using metamaterials in a standard CMOS process, Neurophos has achieved an astounding 8000x reduction in optical processor size, promising an order of magnitude improvement over current GPUs.
Alexander Hayes, co-founder of Metacept, talks about Breaking Bottlenecks in Speed and Energy, the importance of moving away from Von Neumann architecture, and sees Neurophos' metamaterial and photonic applications as groundbreaking in overcoming speed and power bottlenecks.reported energy consumption data from large data center operators.
MetaVC partners: Architects of Neurophos' Genesis
In addition to providing initial funding, MetaVC Partners granted an exclusive license to its metamaterials IP portfolio for optical computing. Neurophos, which emerged from Metacept, is led by David R. Smith, a luminary in electrical and computer engineering at Duke University.
Duke University Validation, Prof. David Smith's Insight
Professor David Smith of Duke University praises Neurophos for addressing the fundamental challenges of analog inference processing. The metamaterial breakthrough, he notes, is a fundamental innovation that paves the way for an extraordinarily dense computing chip for next-generation AI applications.
Silicon Catalyst, accelerating semiconductor solutions
By joining forces with Silicon Catalyst, the world's only incubator + accelerator focused on semiconductor solutions, Neurophos gains privileged access to a robust ecosystem. This strategic move positions Neurophos on an accelerated path from idea to prototype with unparalleled support and access to industry experts.
Paul Pickering, Managing Partner at Silicon Catalyst, expressed his confidence in Neurophos' role in advancing analog optical computing. He believes that the combination of silicon photonics and CMOS foundries will position Neurophos as a leader in the next generation of AI hardware.
Silicon Photonic Optical Chips
Delving into the intricacies of Neurophos' advancements reveals a meticulous plan to shrink the size and power requirements of silicon photonic optical chips. The metamaterial-based optical modulators, which are over 1000 times smaller than standard counterparts, pave the way for a technology roadmap that promises over 1 million TOPS, dwarfing current DNN performance benchmarks.
Optical chips, catalyzing the evolution of AI
Neurophos orchestrates a dual breakthrough by fusing an optical metasurface with a compute-in-memory (CIM) processor architecture. This fusion, powered by high-speed silicon photonics, revolutionizes matrix-matrix multiplication, the backbone of neural net operations. The result is a processor that is hundreds of times more energy efficient than alternatives, placing Neurophos at the forefront of optical computing innovation.
Source: IEA analysis based on Masanet et al. (2020), Malmodin (2020), Hintemann & Hinterholzer (2022), and reported energy consumption data from major data center operators.
