Lighting up the quantum computing horizon with Aurora

January 22, 2025

A peek inside Xanadu’s latest and greatest photonic quantum computer—Aurora.

By Zachary Vernon

The Xanadu team has recently announced the successful completion of Aurora, our latest photonic quantum computer. This machine follows in a line of successively more sophisticated earlier systems demonstrations, including X8 and Borealis. Both of those were ground-breaking in their own right: X8 brought us the first commercially cloud-deployed photonic quantum computer, and Borealis became one of the few machines in the world capable of demonstrating quantum computational advantage. But Aurora is very different—and much more exciting. A full description of this system was just published in the peer-reviewed journal Nature. Here we’ll summarize the main takeaways from this significant step forward in our rapidly-developing field of quantum computer hardware development.

Pictured above, Aurora is a complete prototype of a universal photonic quantum computer following a design that will one day be able to perform fault-tolerant computation. It is the very first time we—or anyone, for that matter—have combined all the subsystems necessary to implement universal and fault-tolerant quantum computation in a photonic architecture. Many earlier demonstrations of different building blocks towards this lofty goal exist, including our own X8 and Borealis systems, whose key underlying technologies are used within Aurora. However all such demonstrations lacked one or more vital features necessary for a practical machine. Aurora, on the other hand, has it all: 35 photonic chips, networked together using a combined 13 km of fiber optics, carrying out all the essential functions needed by our comprehensive blueprint for fault-tolerant quantum computation. These include qubit generation and multiplexing, synthesis of a cluster state with both temporal and spatial entanglement, logic gates, and even real-time error correction and decoding operations that are executed within a single quantum clock cycle. All of the computation in this system happens within the confines of four standard, room-temperature server racks, fully automated and capable of running for hours without any human intervention.

The construction of Aurora is the outcome of a year of dedicated effort spanning nearly the entire Xanadu hardware and architecture teams, representing a convergence of many years of our innovations in photonic chip design, packaging, electronics, and systems design and integration. As described in our paper, we put Aurora through a set of rigorous benchmarks to showcase its many novel technological features. In one trial, we ran the system continuously for two hours and monitored the entanglement present in the cluster state over 86 billion modes—the largest number ever accessed in a context like this. Quantum correlations were also used to verify that the single-clock-cycle decoding functions were working properly. This latter capability refers to the ability to make quantum measurements, then rapidly (using classical controllers, in our case FPGAs) detect errors and calculate and implement corrective quantum gates on the next quantum clock cycle. Doing this sequence of operations will be critical for all fault-tolerant quantum computers, and has never been demonstrated in a photonic machine before.

Our paper detailing these demonstrations is entitled Scaling and networking a modular photonic quantum computer. We chose these words carefully to reflect the three major and closely related aspects of our architecture that Aurora substantially de-risks: scalability, networkability, and modularity. A truly useful quantum computer will require a very large number of physical qubits, necessitating an approach that includes a clear template for scaling up. Irrespective of the hardware approach, these qubits will not be able to fit into one contiguous system. Modularity is thus crucial: there must be a straightforward way to distribute the qubits amongst discrete modules that can be mass-manufactured independently. And modularity is useless without the ability to network the modules together in a way that enables entanglement to be shared across separate chips. Aurora is a powerful demonstration of all three, and allows us to set aside any doubts about these aspects of our photonic architecture.

3D rendering of the Aurora system, showing the fiber connectivity between adjacent racks, enabling multiple chip modules to be entangled. On the right is a high-level breakdown of what’s inside a typical module for each subsystem. Aurora incorporates 24 qubit source chips, 6 multiplexer chips, and 5 quantum processing unit (QPU) chips. The only subsystem not pictured is the photon detection system, which is housed in a cryostat (the only cryogenic component in Aurora) and includes 36 detectors.

These scalability features are necessary for long-term success in quantum computing, but alone they are not sufficient. To make scaling up worthwhile, physical qubit performance must also be sufficiently high. Otherwise, error correction cannot work, and adding more qubits simply ends up making logical error rates even worse. Once component performance passes a level known as the fault tolerance threshold, this asymptotic trend flips, and adding qubits suppresses logical error rates, enabling algorithms to run for longer. This is necessary for all known high-value quantum computing applications, and ultimately required for our fledgling industry to deliver a positive return on investment.

In our photonic approach, crossing the fault tolerance threshold translates into driving optical losses down. When photons propagate through chip and fiber components, small imperfections, like material impurities or roughness in optical interfaces, can cause some of the light to be absorbed or scattered away. Lowering the degree of such losses in our components is largely all that stands in the way of a system like Aurora achieving fault tolerance. As part of this project, and as reported in our publication, we undertook a deep study of the optical loss requirements for our architecture, combining our latest optimizations to refine and update our architectural blueprints and derive precise quantitative loss budgets for all the optical paths in our system.

The results of this study give us a clear mandate for how to improve the performance of our components to reach the fault tolerance threshold. The chips used within Aurora were based on commercially available fabrication platforms that had not been specially optimized for this application. Meanwhile, we have been working with our foundry partners on customized fabrication processes that will satisfy the stringent performance demands of fault-tolerant operation. This mandate is now the sole focus of the Xanadu hardware and architecture teams, and much exciting progress has already been made at winning this war on loss. Stay tuned!

Published on January 22, 2025