Few challenges are as formidable as building a quantum computer. It’s not just about wiring its components but also making them work together to produce accurate computation results despite the presence of noise that can introduce errors in quantum computations.

While you can write down a quantum computation as an abstract mathematical model, implementing it practically still means adjusting certain parameters, such as microwave pulses or lasers, which you can only do with limited precision. Thus, there would inevitably be a gap between the model you intended to implement and the actual outcome.

Qruise leverages machine learning to narrow this gap. By building a physical model from a quantum computer’s experimental data and comparing it to the intended behavior, Qruise helps physicists and engineers to improve their quantum computers. This also works for other quantum devices, such as quantum sensors, and other fields like photonics.

Founded as a spinoff from Forschungszentrum J&uum1lich in late 2021 by Shai Machnes, Frank Wilhelm-Mauch, Tommaso Calarco, and Simone Montangero, Qruise raised funding from Constructor Capital and went through the Creative Destruction Lab startup program.

Learn more about the future of machine learning for quantum computing and beyond from our interview with the co-founder and CEO, Shai Machnes:

Why Did You Start Qruise?

Even before Qruise, my co-founders and I were researching how to control quantum systems. At some point, more academic groups than we could handle as researchers wanted to collaborate with us, so we decided to found a company in late 2021.

Our initial focus was just quantum control, but we soon realized that the very same tools could be applied to quantum sensing or even other domains, like photonics.

When I think of Qruise today, we’re actually building a “machine learning physicist”—a system that can predict and control all kinds of physical processes. I already had this vision more than 15 years ago as a researcher, but I wasn’t able to realize it. With recent advances in computing power and data availability, machine learning has greatly improved, making it possible for us to pursue this vision.

How Do You Control a Quantum Computer with Machine Learning?

Quantum physics governs how nature behaves at a very small length scale. These behaviors usually don’t make sense to our human brain—which is used to much larger length scales—but when you go very cold and very small, the laws of physics look a lot different from what we’re used to, and we could leverage them for computing. (You can watch Shai’s TEDx talk on quantum computing to get the gist of this.)

If you don’t measure where a particle is, it can be in more than one place at the same time! If you label two boxes with “0” and “1” and you don’t look, an electron could be in both boxes simultaneously. That’s why quantum physics is weird; it doesn’t make sense intuitively. But it’s our imagination that’s limited rather than nature being wrong. Understanding something means translating it into terms you’re already used to—and no one is used to quantum phenomena. So don’t worry if you find quantum physics confusing.

A quantum system with two states, “0” and “1”, is a quantum bit, or in short, qubit. You can visualize it as a point on a sphere, where 0 is at the south pole, and 1 is at the north pole. So, in general, a qubit is a combination of both states, where the latitude, how close you are to the south or north pole, measures the probability that you’ll get the respective state 0 or 1 when you measure the qubit’s state.

The magic of quantum computing comes about when you entangle different qubits so they share a joint state—a single qubit encodes a combination of two states, but two entangled qubits encode a total of four states, and three qubits encode eight states, and so on. You see that it grows exponentially as the number of qubits increases. Ultimately, 200 qubits can be at the same time in more states than there are atoms in the universe.

This has an important consequence for computing: even if every atom were turned into a bit, we could not write down the state of a 200-qubit quantum computer. It is impossible to simulate such a quantum computer on a classical computer. It’s like comparing a car to a spaceship—the spaceship can go where a car can’t, though you don’t need a spaceship to go grocery shopping. For most things like browsing the internet, you won’t need a quantum computer, but you need them to solve very specific and tough computational problems such as designing new materials.

No matter how perfect you are at building and operating a quantum computer, there will always be a gap between what you intend to implement as quantum computation and what you actually achieve. This is where R&D happens. At Qruise, we’re giving those developing quantum computers, or more broadly, quantum technologies, the tools they need to understand why their quantum device is not perfect and the reasons for the imperfections.

Simulating and building an electric motor is different from building a quantum computer since we understand how things work in the former, and if we design and simulate it thoroughly, it will work as expected. But quantum computing is hard to understand and get right, as different sources of noise can introduce errors and spoil your computation results.

We help our customers, specifically the engineers and physicists working in the lab, to improve their quantum devices iteratively. We give them the tools to reverse-engineer their hardware so they can learn about them and build the next generation better than the previous one. Machine learning helps to extract a model of a hardware device intelligently from experimental data, quantify the gap between experimental data and the model, and then improve the model to close that gap.

How Did You Evaluate Your Startup Idea?

When we started, we already knew from our time as researchers that there were many other research labs out there interested in what we were building for quantum control, so we focused on that.

We estimated that if you could make a team of three as effective as a team of five and multiply that by the number of labs, there would be decent value created and enough for us to capture. While quantum technologies is the starting point, we will move even beyond that—the market for simulation software is a $20 billion per year market, and by leveraging machine learning, we are going to make a dent in that market as well.

What Advice Would You Give Fellow Deep Tech Founders?

Building a startup is a lot of stress and work, and it’s psychologically so much easier if you don’t feel like you’re doing it alone. If you are a solo founder, you feel the weight of responsibility on your shoulders.

Therefore, I suggest starting a company with a friend or two that you've known and worked with for several years. It’s like getting married, so it might be a good idea to go on a long vacation with that person and see how you handle things together. Put yourself in stressful situations—only then will you know you work together well as a founding team.

Who Should Contact You?

We’d love to speak to fellow quantum and ML enthusiasts and look for bright people to join our team. Please check out our hiring page.