Lightmatter Unveils Passage M1000: World’s Fastest AI Interconnect Photonic Superchip Delivers 114 Tbps Bandwidth Breakthrough

Lightmatter has unveiled the Passage M1000, a 3D photonic superchip designed for next-generation AI infrastructure, in Mountain View, CA. The M1000 achieves a record-breaking 114 Tbps of optical bandwidth and supports die complexes exceeding 4,000 square millimeters, enabling connectivity to thousands of GPUs. It overcomes traditional electrical I/O limitations by integrating electro-optical connections across its surface, supported by a reconfigurable waveguide network and 256 fiber attachments for high-bandwidth transmission.

Collaborating with GlobalFoundries and Amkor, Lightmatter has ensured production readiness, leveraging GF’s Fotonix platform for seamless photonic-CMOS integration. The M1000 features an 8-tile interposer, programmable waveguides, 56 Gbps SerDes, WDM transmission, and a compact package with 256 fibers, delivering unprecedented performance. It will be available in summer 2025 alongside Lightmatter’s Guide light engine, advancing AI model training capabilities.

The Passage M1000 photonic superchip is designed to address the demands of next-generation AI and high-performance computing workloads. By integrating photonics with CMOS logic, Lightmatter has created a solution that enhances scalability and performance for data center infrastructure.

Key Features Of The Passage M1000

The M1000 features a 3D-stacked design that efficiently connects thousands of GPUs, eliminating traditional data bottlenecks. This innovation enables seamless communication between processors at unprecedented speeds.

The chip’s extensive waveguide network supports WDM transmission, ensuring efficient handling of high-bandwidth signals across the system. The M1000 includes an 8-tile interposer with a programmable waveguide network and 1024 Electrical SerDes operating at 56 Gbps, providing robust connectivity.

With 256 optical fibers, each delivering 448 Gbps, the M1000 significantly improves bandwidth density compared to conventional solutions. This enhancement facilitates the creation of larger die complexes and enhances GPU connectivity, crucial for high-performance computing tasks.

The chip’s power efficiency is optimized with a 1.5 kW delivery in an integrated package, ensuring high performance while managing energy consumption effectively.

Lightmatter has collaborated with industry leaders GlobalFoundries (GF) and Amkor to ensure the Passage M1000 is production-ready. This partnership leverages GF’s Fotonix platform, which integrates photonics into CMOS logic for seamless scalability in AI infrastructure.

Anticipated for release in summer 2025, the Passage M1000 will be complemented by Lightmatter’s Guide light engine, further enhancing its performance capabilities. This innovation represents a significant step forward in optimizing data center infrastructure for AI workloads.

The Passage M1000 photonic superchip introduces transformative advancements for next-generation AI and high-performance computing. Its 3D-stacked design efficiently connects thousands of GPUs, eliminating data bottlenecks and enabling unprecedented processor communication speeds.

By integrating photonics into CMOS logic using GF’s Fotonix platform, the chip enhances scalability for AI applications. The extensive waveguide network supports WDM transmission, ensuring efficient handling of high-bandwidth signals across the system.

The M1000’s design includes an 8-tile interposer with a programmable waveguide network and 1024 Electrical SerDes at 56 Gbps, providing robust connectivity. With 256 optical fibers delivering 448 Gbps each, the chip significantly improves bandwidth density compared to conventional solutions.

The Passage M1000’s power efficiency is optimized with a 1.5 kW delivery in an integrated package, ensuring high performance while effectively managing energy consumption. This innovation represents a significant advancement in optimizing data centre infrastructure for AI workloads.