During the AWS annual Re:Invent conference in 2019, the public cloud provider announced Braket, a service and hardware initiative for quantum computing.
The service was introduced in 2020 with support for three machines: a superconducting gate-based machine from Rigetti; a trapped-ion machine and a quantum annealing device from D-Wave. The latest update is claimed to offer a 10x improvement in the performance of hybrid workloads.
Discussing how the company’s quantum computing strategy began, Richard Moulds, general manager at AWS Braket, says: “We launched a hardware effort, which is actually co-located at the Caltech campus in California to provide a superconducting quantum computer.” This, he adds, was accessed via Braket services.
The company’s quantum computing strategy is focused on being hardware agnostic, offering its customer access to quantum computing technology through Braket. Since quantum computing technology is evolving, no one knows for certain which quantum technology will emerge as the winner.
Organisations wishing to use quantum computing need to understand the difference between the various architectural approaches, weigh up the pros and cons, and learn how to take advantage of each type of quantum computer when developing use cases.
By 2020, Moulds says AWS had three quantum computing devices available: “Our goal was to make a wide variety of machines available to customers so they could see what was real, so they could see what claims made sense and could get a sense of the trajectory in terms of how quickly these machines are evolving.” This, he says, helps customers pan and enables them to understand if quantum computing is going to be in their business.
Different approaches to quantum computing
Moulds points out that there is no one perfect architecture for a quantum computer. Businesses need to understand the different technical approaches each architecture offers, “so customers want to experiment across different devices. Nobody wants to be locked into a single technology these days. It’s just way too early,” he adds.
“The thing that customers can do right now is experiment with today’s quantum hardware. There’s lots of different ways you can build a quantum computer. You know you can use ions, electrons or photons. But they all have really different trade-offs in terms of quality, the error rates, speed of operation, the number of qubits, the type of gates they can deliver and the connectivity between those gates. All vary enormously.”
Among the areas of quantum computing AWS has tried to address is to make it easier for customers to access a quantum computing machine to see how they are operated. According to Moulds, the companies developing quantum computing may often only have a few machines that are operational: “These things are in labs and oftentimes you only have private access and you need to know the scientists involved.
“It’s not like they have racks of servers. These are early stage devices. In some cases you might even describe them as prototype machines. So, they need some care and feeding and might require calibrating every day. The technology is not mature enough right now, unfortunately, to build quantum computers in an Amazon datacentre.”
“Nobody wants to be locked into a single technology these days. It’s just way too early”
Richard Moulds, AWS Braket
What this means is that the programs that customers want to run need to run on quantum machines that are not inside the AWS physical boundary. Mould says: “We have to do a lot of work in terms of securing that connection and we work very closely with the quantum computing hardware providers to make sure that their security infrastructure is appropriate for running our customers’ data.”
This also means that applications on AWS that want to make use of quantum computing resources through Braket require optimisation to reduce the network communications to the target quantum computing site.
“There’s always a real important interplay between classical computers and quantum computers – they work hand in hand. So, whatever you can do to make that connection efficient, will ultimately speed up the application,” says Moulds.
Industry challenges
Along with access to the hardware, Moulds says that another problem in quantum computing is that the industry is fragmented. “Each machine basically has its own developer environment, its own developer toolkits. We want to get away from that and try to make it possible for customers to design quantum algorithms then run those algorithms on as many different types of [quantum computing] machine as possible.”
According to Moulds, today a quantum computer cannot outperform a classical machine. He says: “People have come up with some synthetic problems and synthetic examples, which generally is referred to as quantum supremacy, but these are not useful problems. No one is using these devices for running production workloads right now. No one’s saving money.”
He says many businesses want to find out if there are any useful problems that can be solved by quantum computing faster or cheaper than a classical machine.
There is, however, a lot of work being done in research, where scientists try to understand how they build a better machine and investigate how to identify better algorithms for quantum computing.
Moulds says people are looking how to use classical computing techniques combined with machine learning to make quantum algorithms run better, adding:
“Today, it’s all about experimentation.”
Moulds says scientists are also looking at the potential to use quantum computing for molecular simulation. He believes this may be the first quantum computing use case that has the potential to succeed. “It sort of makes sense,” he says. “When you think about it, you’re using a quantum machine to basically simulate a quantum system.”
A chemical reaction involves atoms and molecules interacting at the quantum level – electrons and protons. The quantum computer effectively runs a digital twin of the experiment. The results from the real-world, “wet lab” experiment are then used to adjust parameters in the digital twin, until it produces the same results as the real chemical reaction.
But, as Moulds points out, digital twins that run on a classical computer are unable to model chemical reactions accurately: “You can’t model this stuff perfectly using a classical machine. It’s just too complicated.”
Petrochemicals, drugs and plastics are based on huge molecules, which simply cannot be modelled. With classical computing, he says that there are lots of compromises and lots of simplifications. A quantum computer offers scientists a potential way to run the digital twin without simplification and compromises.
From the conversation with Moulds, it seems that quantum computing devices are a long way off. Certainly, it will take some time before the technology has matured to a point where they can be deployed in the same way as datacentres infrastructure in public clouds. Nevertheless, this should not stop organisations from experimenting with the different types of quantum computing architecture to investigate potential use cases.
However, the fact that it is still early days for quantum computing means that applications that target quantum computers can be verified using classic computing architectures. In the future, as and when quantum supremacy is achieved, it will no longer be possible to run these simulations to verify the results.