What will million-qubit computers look like in a few years?
At the APS March meeting, Google reiterated that it would achieve a million-qubit error-correcting quantum computer by 2029, but startup PsiQuantum set the target for the mid-2020s, around 2025. What will the quantum computer of the future look like? PsiQuantum has come up with a rough idea of what its 1 million-qubit quantum computer might look like.
Pete Shadbolt, chief scientific officer at PsiQuantum, said: "It will look like a big concrete building with a bunch of modules. In those modules, there's a bunch of silicon chips, half photonic, half electronic, and the whole device uses your The optical fibers seen today are connected together.” So PsiQuantum is betting that silicon photonics will play a central role in its data center-sized quantum computers.
contemporary data center
"When you think about quantum computing, you think of low temperatures in millikelvins, atoms flying through space, atomic-scale manufacturing, crazy materials, science fiction, stuff like that," Shadbolt said. Take advantage of what can already be made."
The system will be based on components fabricated using today's manufacturing technology and will not require large-scale refrigerators. "If you look at it as a casual observer, it looks like a large industrial facility with some steam coming out of the top," Shadbolt said.
PsiQuantum's approach is one of many quantum architectures pursued by big tech companies like IBM, Google and Microsoft, as well as startups like Rigetti and IonQ. Some quantum systems are already available in the cloud. Most companies focus on building scalable and reliable error correction systems. In a quantum computer, information is encoded in qubits. To deal with the finicky nature of qubits, error correction is required. PsiQuantum's error-correcting quantum computer specifically involves silicon photonic modulators, fiber-optic networks and other components.
"The idea is that we're going to use the same manufacturing process that we use to make transistors," Shadbolt said. "Specifically, we're going to make optical waveguides. We're going to put light in a chip. We can then use a toolbox of components to manipulate This light. At the physical level, it's actually a single photon that travels the same way as a data center."
Russian researchers have demonstrated[1] that photonic beam splitters can be used to build a universal quantum computing system.
"It's a starting point, and turning it into a computer is very complex, and it's going to be a massive, high-performance computer-like system with lots of silicon and fiber in it," Shadbolt said.
Like PsiQuantum, Google, IBM, and others are also looking to build a 1-million-qubit system, but cooling may be a limitation of such superconducting systems because they operate at one-hundredth the temperature of deep space.
"Photons don't feel heat," Shadbolt claims. We do use some cryogenic cooling systems, but not nearly as much. Our qubits experience photon loss, but they don't really feel heat, and they don't feel electromagnetic interference. "
Connectivity is also a consideration when building data center-scale quantum computers, as it is impossible to fit a million qubits on a single chip.
"You need to connect the chips together," Shadbolt said. "You can't just use Ethernet, you need a quantum interconnect that can send qubits from one chip to another. The only good way is with light, with photons. "
The startup is acquiring system components, such as single-photon sources and single-photon detectors, for its quantum systems built by chipmaker GlobalFoundries. The company says it already has the control electronics needed for qubit coherence.
PsiQuantum's Q1 chip
Shadbolt promises: “We’re building a data center-like system — we have to lay concrete, erect steel beams, and it takes time. But a safe, straightforward answer is that in the mid-2020s, we’re going to stop all manufacturing Process. Soon we will have a quantum computer."
The unique route of PSiQuantum
Unlike other teams, PsiQuantum hopes to avoid the NISQ (Noisy Intermediate Scale Quantum) era and go straight to fault-tolerant processors. By doing so, they hope to reach their fault tolerance goals faster than others. They believe that all commercially viable applications require a fault-tolerant architecture. Any activity that users do now with the new generation of NISQ machines has limited value as they cannot accomplish anything useful, and fighting bugs in these machines will distract users from creating the right algorithms to solve their problems .
Several other quantum companies have a similar view, including Microsoft, Intel, and possibly Google. While these companies may not be so absolute, it may just be saying that most commercially relevant quantum applications will require full fault tolerance, not that all applications will.
Another core belief of PsiQuantum is that creating such a computer will require the use of the best manufacturing processes in industrial semiconductor fabrication plants. In the silicon photonics technology used by PsiQuantum, precise process control is critical in order to minimize losses and noise in the circuit. For example, when a photon travels through a channel in a silicon wafer, the surface of the channel should be as smooth as possible. Any roughness on the channel surface will cause signal loss, which can be quite a problem.
Roughness affects photon detection efficiency
PsiQuantum utilizes standard semiconductor processes and does not require the use of unusual materials or atomic-scale fabrication steps that Tier 1 fabs would otherwise be reluctant to work with. PsiQuantum selected GF as its foundry partner for the most advanced process and precise process control. They noted that as they moved from a 200mm fab to GF's 300mm fab, the pattern quality improved fivefold and the intrinsic efficiency of the single-photon detector increased from 97% to 99.7%.
300mm wafer
By the mid-2020s, PsiQuantum says, all the manufacturing processes needed to build a 1 million physical qubit machine will be in place. But it doesn't mean that there is a machine available at this time. It still takes some time to build and integrate all the parts for a functioning machine. So the availability of machines that users can use to solve real-world problems will likely be in the late 20s rather than the mid-20s.
Part of the reason it might take a while is their modular architecture. The modularity of optical quantum computers, coupled with the flexibility of optical connections, enables many possible fault tolerance schemes. Their component is a self-contained module. But to provide a complete 1 million-qubit system, they will use optical fibers to interconnect many of these modules together. This technique should be easier to do than other qubit technologies. Because their technology is based on photonics, they don't need to use converters to convert qubits into something that can be sent over fiber-optic cables.
Other technologies, such as superconducting or ion trap qubits, want to send the qubits through a fiber-optic cable to another processor chip, first requiring a converter to convert the qubits to photonic qubits.
Another problem with multiprocessor implementations is the connection between modules. While technologies such as ion traps may enable full connectivity within one module, this is not the case when circuits require the use of two qubits in different modules. One can solve this problem using a circuit called a swap gate, which swaps the states between two qubits. However, swapping gates is problematic if you use lower fidelity gates. If there is a chance of error every time a qubit passes through a gate, you will end up with large errors if your implementation needs to use too many qubits.
But with an error-correcting architecture like PsiQuantum, swap gates are no longer an issue, and you can use them to solve inter-module connectivity issues. There may be other, more sophisticated ways to solve this problem that may be better than the linear overhead of swapping a chain of gates, but these can still be problematic if the gates don't have a zero error rate.
The result will be that PsiQuantum's quantum processors are physically quite large. The system will have a large number of modules and will look more like a cluster of classical computing data centers. It will be placed in a room the size of a large data center.
Unique business model
You might think that since PsiQuantum's hardware is not yet available, they would have limited exposure to customers. But that's not the case either. They deal with customers, but not in the same way as most other hardware vendors. PsiQuantum is working with clients on detailed thesis analyses of the problems the clients want to solve with quantum computers. They are calculating the exact number of qubits and gates needed, the time to solve the problem, and the physical error rate required before error correction to run a program that could provide a solution to a real-world customer problem.
By contrast, most customers today work with other hardware vendors for training or experimentation purposes, using what is sometimes called a "toy model." These certainly have value, but it is uncertain whether customers can scale their solutions from "toy models" to larger models with commercial value. Things change as you scale up and encounter new, unanticipated problems, which you'll catch earlier if you did a more detailed thesis analysis.
Optical quantum computing stack
PsiQuantum has customers and partners in the financial, pharmaceutical, energy and automotive sectors to help commercialize its quantum initiatives, Shadbolt said. But the high-risk, high-reward nature of quantum computing depends on getting a commercial product to market quickly.
To achieve this, they still have a lot of work to do. But PsiQuantum is very clear about how they intend to make their quantum machine a reality, and the necessary steps needed to make it happen. They are also very well-funded, raising a total of $665 million, and have assembled a large, talented team to tackle the remaining engineering problems.
On the other hand, other hardware companies are rapidly advancing their technology. So it's still a race to see how PsiQuantum's approach stacks up against other quantum computer implementations, and we'll have to wait another 5-10 years to see how it turns out.
Link:
[1] https://www.nature.com/articles/s41598-021-89838-5
[2] https://www.theregister.com/2022/03/08/psiquantum_data_center_quantum/
[3] https://quantumcomputingreport.com/a-closer-look-at-psiquantum/
