Reference Platform

M1000

3D Photonic Interposer for Exascale AI.

The M1000 is the world’s first 3D Photonic Superchip. The M1000 reference platform consists of a server with all components needed to operate the chip. By stacking SerDes link chiplets directly onto the Passage™ active optical interposer, the M1000 eliminates the “shoreline bottleneck” and enables massive escape bandwidth for next-generation AI accelerators.

Request Evaluation Kit View Specifications

Total Bi-Directional

114.6

Tbps

System Efficiency

2.3

pJ / bit

Includes laser power. SerDes efficiency: ~2.0 pJ/bit

The Shoreline Bottleneck

Traditional chips are limited by their perimeter (shoreline) which is where I/O devices for communication reside. As chips get larger, the ratio of edge-to-area drops, starving the core of bandwidth.

This fundamental geometric limit forces data to travel long, energy-draining distances across the chip to reach the edge, creating latency and power bottlenecks that stall AI scaling.

Core Innovation

Edgeless I/O Architecture

The M1000 utilizes a vertical stack where SerDes chiplets are bonded via copper pillars directly to the Passage PIC. This allows I/O placement across the entire surface of the die, rather than being restricted to the perimeter.

Edgeless I/O

We don't just use the edge. We use the entire 4,000 mm² surface for connectivity. Eliminate energy-draining network-on-chip journeys by placing I/O directly where it's needed.

Unprecedented Density

32 SerDes link chiplets providing 1,024 total lanes in a 2x4 tile matrix. Industry-leading I/O density impossible with conventional packaging.

Powered by Guide™

The Guide™ ELS supports the M1000 with 8-wavelength DWDM. Unprecedented optical power and reliability for massive scale.

Optical Circuit Switching

Integrated optical routing layer allows dynamic configuration of network topologies. All-to-All, Nearest-Neighbor, Hypertoroid, Ring patterns.

1 Million XPUs and Beyond

Passage enables unprecedented scale-out connectivity, linking millions of accelerators across the datacenter with photonic bandwidth.

Hardware

M1000 Platform

The M1000 evaluation kit demonstrates 3D photonic integration at scale, from chip-level innovation to rack-scale deployment.

Edgeless I/O

Photonic Interposer

4,000 mm² active area with 256 fiber connections

Reference Platform

Rack-Scale Integration

Detachable optical connectors for 3U servers

Production Ready

Validation Datacenter

Hundreds of Passage-enabled servers

See It In Action

2x4 Configuration

Industry-Leading Performance

The M1000 is presented in its high-density 2x4 configuration, delivering unprecedented throughput and energy efficiency for exascale AI networking.

CORE SPECIFICATIONS

Total Bandwidth

114.6 Tbps

External Optical BW

57.3 Tbps

Signaling Protocol

56 Gbps NRZ (PAM2)

Form Factor

2x4 Tile Matrix (8 Passage™ Tiles)

2.3 pJ/bit System

By utilizing light for data transport, the M1000 achieves a total energy footprint of 2.3 pJ/bit. The highly optimized SerDes layer accounts for ~2.0 pJ/bit, ensuring that the majority of power is dedicated to data movement.

2.0 Serdes
0.3 Laser
[ 0.0 ] TOTAL: 2.3 pJ/bit

Comprehensive SDK

The M1000 EVK is supported by a comprehensive SDK designed for deep performance characterization and validation.

Link Quality Analysis

Real-time Bit Error Rate (BER) monitoring at 56 Gbps NRZ for comprehensive link validation.

Advanced Telemetry

Sensing for power dissipation (up to 1,118 W), thermal gradients, and optical path health.

Liquid-Cooled Ready

Designed for integration into liquid-cooled 19" 3U server environments to manage high-density thermal loads.

Technical Data Sheet

Ref: M1000-EVK-SPEC-001

Technical Specifications

Total Bi-Directional Bandwidth

114.6 Tbps

External Optical Bandwidth

Up to 57.3 Tbps

Signaling Protocol

56 Gbps NRZ (PAM2)

System-Level Efficiency

2.3 pJ/bit (includes laser)

SerDes Efficiency

~2.0 pJ/bit

Form Factor

2x4 Tile Matrix (8 Tiles)

active Interposer Area

4,000 mm²

Optical Fibers

256 total

Power Delivery

Up to 1,118 W support

Infrastructure

Liquid-cooled 19" 3U server

Get Started with M1000

Connect with our team to discuss M1000 integration for your AI infrastructure roadmap. Evaluation kits available to qualified hyperscalers and XPU manufacturers.

Request Evaluation Kit