Silicon photonics (SiPh), the manufacturing of integrated photonics on CMOS platform, has been a buzzword in the recent two years, given the technology's promising prospect to deliver a faster, securer and more efficient solution to data centers increasingly burdened by the ever-growing transmission demand of AI. However, the potential of silicon photonics is not confined to the realm of conventional computing and communication.
Xanadu, a quantum computing company founded in 2016 and headquartered in Toronto, Canada, has been building fault-tolerant computers based on silicon photonics chips. Using photons as qubits, Xanadu believes that silicon photonics will be the quickest path to achieve a fault-tolerant quantum computer able to operate at room temperature. Zachary Vernon, Chief Technology Officer of Xanadu responsible for hardware, talked to DIGITIMES Asia about opportunities brought by photonics to the realm of quantum computing, as he visited Taiwan in November to attend the 2023 Asia Pacific Executive Forum hosted by the Global Semiconductor Alliance.
The race to achieve fault tolerance
Currently, there are several types of quantum computers based on different principles, including superconducting qubits, quantum dots, ion traps and photonics. "At the present moment, there is certainly tough competition between all of them, and you see a lot of approaches that are distinguished by what type of hardware they use," remarked Vernon, noting that as these various approaches are still at prototyping phase, different types of quantum computers are suitable for different near-term problems, and all these near-term problems are a bit short of real business applications. Ideally, if all approaches of building a quantum computer turn out successful, they should all be equivalent and able to address the same problems, according to Vernon.
"In order to get to that point, we need to achieve fault tolerance and error correction, and we think photonics will be the first to get there and the fastest to scale," the CTO explained, emphasizing that scaling and performance are both very important to achieve fault tolerance. "You need lots of qubits to encode error correction, but you also need high performance qubits."
Photonics enables one to network different chips with optical fiber in various patterns, obtaining better connectivity than one would usually be able to access in a superconducting approach. As a result of the better connectivity, Vernon indicated, one can access better codes, especially quantum low-density parity-check (LDPC) codes.
"Photonics is really the only approach that can access it, since the other approaches are very constrained in the connectivity between their qubits, whereas photonics can leverage optical fibers to route qubits wherever you want," said Vernon. As the number of qubits - now usually measured in millions - has come to be synonymous with the global quantum race underway, the Xanadu CTO pointed out that photonics will also need millions of qubits to deliver an advantage. Due to Xanadu's ability to use better LDPC, it can access ten to one hundred times more logical qubits than competing approaches.
"We think it's very important for anyone working in the manufacturing of silicon photonics to pay attention to the quantum computing industry," emphasized Vernon, pointing to the two major advantages offered by silicon photonics: scalability and quantum computing at room temperature. "All of our actual computation happens on room-temperature devices."
Photonics will be the fastest path to scale
According to Xanadu's projection, once it reaches fault tolerance and begin to scale up to add hundreds of logical qubits per year, Xanadu alone would require hundreds of thousands of 300mm wafers per year, as a quantum computer is like a data center that literally takes thousands or millions of chips to build. "It's a very significant market opportunity that in a few years will essentially directly compare with silicon photonics wafer volume in the present day," observed Vernon. In the coming years, he said, Xanadu hopes to achieve fault-tolerance and scale up to 1000 error-corrected logical qubits. "That would look like a data center with about 10,000 racks," he said, indicating that the company's main priority now is to develop the hardware needed to deliver a cloud-deployed, fault-tolerant computer.
In the long run, the use of silicon photonics has the potential to deploy quantum computers closer to the edge. "In principle, there's no fundamental reason why a quantum computer that uses photonics can't be inside a consumer device," explained Vernon, "there are certain technologies that need to be developed for that, but fundamentally, that capability is there because they can all work at room temperature in principle." Whether there'll be a utility for that application for that, however, more time is needed to study how that will look.
In terms of immediate engagement with customers, Xanadu's software library for programming quantum computers, PennyLane, is the main product offering of the company. Xanadu partnered with Amazon Web Services in its development, in addition to cooperation with Nvidia. "PennyLane is one of the leading software APIs for developing algorithms for quantum computers," remarked Vernon, "it started out specializing in machine learning applications - quantum machine learning - but a community grew around it to take hold of quite a significant portion of the market for algorithm development." The Xanadu CTO also highlighted the hardware-agnostic characteristic of PennyLane: it's not limited only to photonic quantum computers or our hardware - one can use it on different platforms, and Xanadu has partnered with multiple hardware providers to enable that. In one example, Xanadu cooperated with multiple automakers like Volkswagen that leverage PennyLane to develop quantum algorithms for battery simulation.
The risk of missing out in a global competition
On the eve of an AI revolution, the Xanadu hardware CTO pointed out that a lot of work has been undertaken in quantum machine learning, even though it's still in its early days. "There's a lot of algorithmic development going on, and it does seem that quantum computers will be able to have the ability to address certain machine learning tasks in a very different way," said Vernon. Nevertheless, large scale fault-tolerant quantum computers are still needed before one can fully access the implication. "Once these things have scaled up to very substantial sizes, then it can address conventional machine learning basic operations - such as matrix operations - more efficiently," according to Vernon.
Instead of replacing data centers, Vernon believes that quantum computers will augment them, indicating that quantum computing doesn't address computational problems and applications that are currently being done by edge clusters. "The types of algorithms that quantum computing across all approaches address are completely different, and they are completely inaccessible by ordinary classical computers as a result of the mathematical structure of the problems to be solved," he pointed out, adding that quantum computing development is not something incremental that gains a slight edge over pre-existing technology. "A good example is the most recent cloud-deployed machine that we built, Borealis, which was able to beat the world's most powerful supercomputer - benchmarked against Fugaku - by many orders of magnitude."
Fundamentally, quantum computing tackles a completely different set of problems that can't merely be addressed by scaling up data center. "It opens up application markets that are simply out of reach and will always be out of reach with present day technology," said Vernon.
Given quantum computing's strategic significance, a global race has been underway. Regarding Canada's advantage in it, the Xanadu CTO observed that the country is punching above its weight in the ecosystem, especially in workforce development, and a number of stellar physicists and engineers coming out of Canadian universities were directly hired by Xanadu. When it comes to Taiwan's advantage, Vernon believes that Taiwan is "the Mecca of semiconductor" and therefore will also be a hub of silicon photonics one day, thus playing a critical role in Xanadu's supply chain in the future. He however stressed that the Taiwanese ecosystem has to pay attention to the customization and optimization "that'd be better to happen earlier than later."
For the photonics ecosystems in Taiwan and elsewhere, Vernon believes that silicon nitride and lithium niobates are two emerging platforms that are extremely important, and Xanadu has been working on them for quite some time. With the help of the Canadian Trade Office in Taipei, Xanadu has been building relationships with multiple foundries and OSATs based in Taiwan, especially those that have been involved in manufacturing Xanadu's devices. "We think Taiwan is perfectly positioned to be a dominant supplier, and what's needed right now is process customization and optimization to ensure the compatibility of the silicon photonics processes, both on the fabrication and packaging sides, with the requirements of photonic quantum computing," indicated Vernon.
"There's a risk of missing out on it if it's not active," he warned, noting that the US and Europe have gained a slight edge in photonics quantum computing since they have spent more time and paid more attention to the relevant requirements over the last couple of years. "The time to act is now to make sure Taiwan stays competitive, and there's no better place in the world that has the sorts of existing infrastructure to support this."
Zachary Vernon, CTO—Hardware at Xanadu.
