In June 2026, silicon photonics company Lightmatter partnered with NVIDIA to bring its optical networking technology to NVIDIA-powered AI data centers. The goal is to replace traditional electrical connections with optical connections that move data faster and use less power.
The partnership focuses on using Lightmatter’s specialized hardware, known as co-packaged optics (CPO) and near-packaged optics (NPO) that plugs directly into NVIDIA’s existing optical and SerDes (serializer/deserializer) technologies. By adapting Lightmatter’s technology to work with NVIDIA’s setup, custom AI chips from different companies can now talk to each other seamlessly while cutting the physical fiber cables and connectors needed to link them by 50 percent.
Lightmatter’s photonic interconnect technology
Founded in 2017 by Nicholas Harris, Darius Bunandar, and Thomas Graham, Lightmatter was established as a spin-out from the Massachusetts Institute of Technology. The company was formed to address the bandwidth, power, and thermal limitations of traditional electronic computing by applying advances in nanophotonics. Its technology uses optical communication to move data between computing components, targeting the increasing bandwidth and energy requirements of AI workloads.
At the core of Lightmatter’s architecture is a photonic interconnect that replaces traditional copper wiring with optical pathways. The system combines photonic integrated circuits with laser light sources and optical modulators that encode electrical data onto light. Using Dense Wavelength Division Multiplexing, multiple data streams travel simultaneously across different wavelengths within a single optical channel, increasing bandwidth while reducing power consumption.
We previously examined similar approaches in our analyses of Celestial AI’s Photonic Fabric and Taara’s photonic communication chip, both of which use light-based technologies to address the bandwidth, latency, and power limitations of conventional electrical interconnects. The following patents provide further insight into the technologies that underpin Lightmatter’s photonic interconnect architecture.
Exploring Lightmatter’s interconnect technology
Lightmatter’s patent portfolio outlines the technical mechanisms behind its optical computing and interconnect hardware. These filings describe methods for generating, routing, and processing optical signals within integrated photonic chipsets.
Doubling fiber capacity via polarization multiplexing
Traditional systems speed up data transmission by sending multiple signals simultaneously down glass fiber cables using different “colors” (wavelengths) of light. To double this speed, engineers can use a technique called polarization multiplexing, which encodes two independent streams of data on the exact same color of light by vibrating one stream horizontally and the other vertically.
However, when using standard, cost-effective fiber cables, these light waves twist and rotate unpredictably as they travel. By the time they reach their destination, the waves are completely scrambled, making the data impossible to read.
U.S. Pat. App. Pub. No. 2026/0163644 entitled “Devices and methods for optical interconnects using polarization multiplexing” describes a specialized microchip (an optical transceiver) that solves this scrambling problem using a master “untwisting” component called a Polarization Splitter Rotator (PSR) combined with polarization controllers.
When transmitting data, the PSR rotates one light wave by 90 degrees so the two streams can travel together without interfering. On the receiving end, the PSR splits the scrambled light back into separate streams, while the polarization controllers act like active fine-tuning dials to instantly untwist and realign the waves so the chip can perfectly read the doubled data stream.
The patent filing was filed on December 9, 2025 and was published on June 11, 2026. The listed inventors are Joyce Poon, Nikhil Kumar, Brian Donovan Jones, Kuang Liu, Clifford Chao, Anthony Kopa, Geoffrey Tu, Danial Stodolsky, and Darius Bunandar.
Doubling fiber capacity via polarization multiplexing
Conventional processors consume significant power and generate substantial heat when executing the large-scale linear algebra operations required for artificial intelligence and deep neural networks.
To improve efficiency, Lightmatter’s system uses an optical encoder to convert digital data into optical signals by modulating both the amplitude and phase of light.
U.S. Patent No. 10,763,974 describes a photonic computing architecture that performs matrix-vector multiplication using light instead of electrical signals. The architecture relies on a mathematical technique called Singular Value Decomposition, which factors large data matrices into three simpler matrix operations that map directly onto physical hardware.
Inside the photonic processor, these math operations take place as light travels through a network of programmable components known as Variable Beam Splitters, often configured as Mach-Zehnder Interferometers.
As light flows through this optical mesh, adjustable phase shifters and attenuators manipulate the light beams to perform matrix operations simultaneously through optical interference. Once the light finishes its path through the processor, an optical receiver detects the resulting light waves and converts them back into digital data, completing the computation with lower latency than traditional electronic hardware.
This patent entitled “Photonic processing systems and methods” was filed on May 14, 2019, and was granted on September 1, 2020, with Darius Bunandar, Nicholas C. Harris, and Carl Ramey are listed as the inventors. Wolf Greenfield & Sacks provided the legal representation.
Together, these core patent filings illustrate how Lightmatter’s optical hardware functions at both the interconnect and processing levels. To gain a broader perspective on how the company is protecting and expanding these technical innovations worldwide, we will now look into the patenting activity of Lightmatter.
Lightmatter: Patenting Activity
Lightmatter’s global patent filings increased rapidly between 2018 and 2019, reaching an early peak in 2019. This period reflects the company establishing its foundational patent portfolio following its spinout from MIT and early venture financing. From 2020 through 2022, patent activity stabilized as engineering efforts shifted toward hardware development, including the Envise AI accelerator and the Passage photonic interconnect platform, advancing the technology toward commercial deployment in AI data centers.
Patent activity accelerated again in 2023 and reached another peak in 2024, reflecting expanded research into 3D co-packaged optics, multi-wavelength optical systems, and other next-generation photonic technologies. This renewed filing activity coincided with significant capital investments, including the company’s Series C funding round in late 2023 and Series D financing in late 2024, supporting the expansion of AI data center infrastructure.
Lightmatter: Top Technology Areas
Lightmatter’s patent filings reflect a primary focus on optical infrastructure and high-speed data transmission systems. The largest share of filings falls under optical elements and systems (G02B), followed by transmission technologies (H04B), digital data processing (G06F), specialized computational modeling systems (G06N), and multiplex communication (H04J).
Together, these core classifications account for the majority of the company’s patent activity. This concentration aligns with Lightmatter’s core technical objective of deploying silicon photonics that use multi-wavelength optical channels to replace copper interconnects, directly addressing data transfer bottlenecks in large-scale computing clusters.
The remaining portfolio is distributed across specialized component and packaging categories, including optical computing devices (G06E), optical modulators and control devices (G02F), semiconductor packaging (H10W), solid-state devices (H01L), and telecommunication switching (H04Q).
These categories support the physical manufacturing and integration of photonic hardware. The presence of patent filings in semiconductor interconnections and optical control devices corresponds to engineering efforts in 3D photonic interposers and co-packaged optics, where laser engines, modulators, and silicon dies are integrated into unified chip packages to manage thermal and density constraints in AI data centers.
Lightmatter: Top Jurisdictions
Lightmatter’s patent strategy is centered on the United States, reflecting its headquarters and primary market of North American hyperscale data centers. Japan is its largest foreign filing destination, supported by investment partnerships and its role in the East Asian semiconductor ecosystem.
Lightmatter: Top Law Firms
Lightmatter’s patent prosecution is handled primarily by Boston-based intellectual property firm Wolf, Greenfield & Sacks. This concentration reflects Lightmatter’s origins as an MIT spinout located in Boston, facilitating direct collaboration between inventors and primary legal counsel during initial patent drafting. Foreign filings in major international jurisdictions are managed by Onda Techno Intl. Patent Attys. in Japan and Withers & Rogers in Europe. Onda Techno handles national phase entries into Japan following regional investments from firms such as SIP Global Partners, while Withers & Rogers manages examination and validation procedures before the European Patent Office.
The remaining patent applications are distributed among regional law firms executing localized national phase filings and portfolio maintenance. In mainland China and Hong Kong, prosecution is divided among Beyond Attorneys at Law, Chang Tsi & Partners, and China Hong Kong Intellectual Property. Additional foreign representation across Singapore and the broader Asia-Pacific region is conducted by Spruson & Ferguson, while HGF handles supplementary European filings. Finally, Smart & Biggar oversees patent filings in Canada, supporting Lightmatter’s engineering operations at its Toronto R&D facility.