QuantrolOx: Shaping the Future of Scaling Quantum Computers through Automation

QuantrolOx: Shaping the Future of Scaling Quantum Computers through Automation

Quantum computers are poised to revolutionize computation by harnessing the power of quantum effects. However, these systems require continuous and meticulous calibration and delicate handling by quantum scientists to maintain their quantum properties and execute complex computations. 

Currently, a single quantum computer can keep 2-3 physics PhDs busy, which is a serious bottleneck when scaling to more qubits and more quantum computers. The remedy is a solution for the end-to-end control of quantum computers using machine learning, paving the way for a new era of accelerating the scaling of quantum computers.

In 2021, QuantrolOx was spun out of the University of Oxford to pioneer the automated control of quantum computers. Founded by Andrew BriggsVishal ChatrathNatalia Ares, and Dominic Lennon, it raised pre-seed and seed funding rounds totaling €5.5M. Its latest round in November 2022 was led by Voima Ventures with 2xNSerendipity Capital, and Oxford Science Enterprises, with existing investors Nielsen VenturesHoxton Ventures, and Hermann Hauser joining the round. 

Learn more about the future of quantum control with machine learning from our interview with the co-founder and CEO, Vishal Chatrath

Why Did You Start QuantrolOx?

For years, I have been working in early-stage science and technology development, where my sweet spot is being involved in the science to develop technology and then using the technology to create products and build a company. 

I like starting with the science since the levels of unknowns are huge, and you need to do everything yourself—develop the technology, develop a product, create a market, and educate people about why they need it. It’s a much bigger challenge than building a product for an existing market. The higher the level of uncertainty, the more things I need to figure out, and the more I learn, the happier I am. If everything were clear already, why would it be interesting? 

I have been working in deep tech since 2014, previously with two machine learning startups of which I was the founding CEO. Then I got the opportunity to join Andrew Briggs to work on quantum computing. You can’t get deeper than quantum computing in deep tech, as you are literally manipulating subatomic particles. Quantum computing is one of the most exciting fields to be in, and every day, I learn new things. 

How Does Quantum Control Work?

During the Industrial Revolution, people had to shovel coal to keep trains going. If you stopped shoveling coal, the train stopped. That’s about where quantum computing is now.

Because quantum computers are very unstable, operating one requires 2-3 scientists to constantly babysit it to keep it in the correct quantum state. If you stop babysitting it, the quantum computer stops working. Unfortunately, the quantum computing industry has a serious talent shortage, and we cannot afford to have scientists with ten years of training babysit quantum computers. Automation frees experts for more challenging tasks.

Research on how to tune qubits automatically started eight years ago, when Natalia Ares joined the University of Oxford as a postdoc under Prof. Andrew Briggs and took another leap forward when, five years ago, Dominic Lennon joined Andrew and Natalia as a Ph.D. student. Dominic’s research during his PhD provided the foundational thinking about our products.

Currently, we mainly work with condensed matter-based quantum computers, i.e., superconducting qubits, tiny superconducting electronic circuits whose quantum state gets tuned by microwave pulses and silicon-spin qubits. Our software connects to the control electronics, and as we gather data about the quantum computer’s state, we can control and tune the qubit’s quantum state increasingly better by adjusting the timing and shape of the microwave pulses sent to them, among other things. In total, we control roughly a 40-parameter state space. 

This gets exponentially harder the more qubits there are, and beyond a few tens of qubits, it’s practically impossible to do it by hand as it simply takes humans too long. And it’s not just a challenge when scaling to large qubit numbers, but also when operating more quantum computers. The number of physics Ph.D. available is a major bottleneck for the development and maturing of the quantum industry. So if we can reduce this from three to two to one and eventually to zero per quantum computer, this would be massive for scaling quantum computing.

A third factor is the uptime of a quantum computer right now: it’s around 5 – 20%, depending on the quantum hardware, but with increasing automation, we want to get above 90%. Since our software is based on machine learning, mainly Bayesian optimization, we will be able to address every major quantum hardware platform and tune it for maximum performance. Having already tested our software with two very different kinds of qubits, namely silicon-spin and superconducting qubits, we are confident that we can also adapt our software to other quantum modalities.

We are going to release the first product in August 2023 for superconducting qubits since it’s the largest market and, for now, the dominant quantum hardware platform by volume. In 2024, we plan to expand to silicon spin qubits, which involves much of the same infrastructure. From there, we could go, e.g., to ion trap-based quantum computers or neutral atoms, depending on where the market volumes are.

The fundamental challenge remains the same: tuning an instrument to control the state of a quantum computer and, depending on the quantum algorithm driving the control electronics in a way that the desired quantum state is reached, encoding the computation result. 

It is very hard to predict timelines in quantum computing, as there can be decades where not much happens and then days where decades happen. With improvements in fabrication and getting good qubits reproducibly and reliably, the technology will mature, and quantum computers will become increasingly useful. Yet, regardless of which technology platform will make the race, the automation problem has to be solved anyway. 

Many people overemphasize error correction. We have a different point of view: if the qubits were better, then there would be a lot less overhead for error correction software to deal with – which can only be a good thing. It’s like preparing drinking water: if the water is initially very dirty, you need many filters to make it drinkable. One filter will simply not cut it.

How Did You Evaluate Your Startup Idea?

One thing that is really important for company building is the environment and conditions you’re building it in. For example, will it be a high-volume or low-volume business? If you’re selling chocolate to humans, you can scale up quickly, as everyone knows chocolate, and many people like chocolate.

Quantum computing is still a low-volume industry, and we don’t know yet how long it will take till it transforms into a high-volume industry. But what is clear is that to move on from low to high volumes, we’ll need to build supply chains for all the critical components. 

In the early days of computers in the 1960s and 1970s, IBM had to build every component in-house as there was no ecosystem of component manufacturers. Eventually, around the 1980s, supply chains got established, and companies started to focus just on specific components. Intel focused just on chips and IBM just on assembling computers.

We’re now reaching the phase of componentization for quantum computers. While blue-chip giants like IBM and Google still build everything in-house, an ecosystem for individual quantum components has started to develop – and that’s why it’s now a good time to start a company like QuantrolOx, focusing on quantum control as one essential component. 

We have to build an efficient business to survive long enough on low volumes, but that’s true for almost any company working on genuine deep tech: we need to be financially self-sufficient and still build a company in a way suited for venture capitalists. Finding product-market fit is now our clear focus, so as the market scales, we’ll scale. 

What Advice Would You Give Fellow Deep Tech Founders?

Whenever you’re building a company, do a roadmap and plan for how much capital you will need – things will always take twice as long, and cash will last for half as long. 

Further Reading

Faster, Cheaper, Better: A Case for Open Architectures – Read more on the QuantrolOx blog about why open quantum computer architectures will prevail in the quantum industry

Quantum computers take hours to ‘tune’ — this startup wants to solve that with AI – Sifted article about QuantrolOx 

Finnish deep tech startup raises new round to bring unique qubit tuning software to market – Press release about QuantrolOx seed round in March 2023 on articstartup.com

QuantrolOx raises €10.5m to stabilise quantum computers – Sifted press release about QuantrolOx’s raise from the European Innovation Council (EIC) 

QuantrolOx uses machine learning to control qubits – TechCrunch article about QuantrolOx’s seed round

Machine learning enables completely automatic tuning of a quantum device faster than human experts – Nature publication featuring the research underlying QuantrolOx