Interview: Dr. Natalia Ares, University of Oxford
Hi Natalia, we found your work on characterizing quantum dots with machine learning fascinating. Tell us why and how you started using this approach in your measurements.
We started using machine learning (ML) techniques in our laboratory because we realized that a bottleneck to the scalability of quantum devices, common to all implementations, is that each device has to be characterized and tuned, requiring the exploration of a large parameter space. Even for an experienced researcher, to manually characterize and tune a quantum device is time-consuming; for a large array of quantum devices the task quickly becomes intractable. Our work tackles this challenge through automation, which I believe will be key in the race to scale from today’s few-qubit devices to a technologically useful number.
In our first paper on this topic, we showed how a machine learning algorithm can efficiently measure quantum devices in real time, reducing measurement times up to a factor of 4. More recently, we reported on automatic tuning of quantum devices faster than human experts without human input. The parameters vary non-monotonically and not always predictably with the control signals, making device tuning an extremely complex task to automate: scientists from DeepMind and my group at Oxford found an algorithm that can dynamically tune a 'virgin' double quantum dot device to operation conditions. We believe this technology is a promising route towards tuning large quantum circuits.
To what other measurement scenarios could ML be applied? What are the benefits and potential pitfalls?
I see a great potential in ML techniques for optimal measurement and control, especially as the complexity of quantum circuits increases. The major benefit can be seen in tasks such as the exploration of a multi-dimensional parameter space. I think the main pitfall is to believe that off-the-shelf machine learning algorithms are ready to solve the characterization and tuning challenges that quantum devices present. We realized we had to innovate the ML techniques to make a significant contribution.
Can you tell us about your journey as a scientist?
I studied physics at the University of Buenos Aires, Argentina, where I am from. My master's thesis was on quantum chaos, a theoretical project focused on the effect of perturbations on quantum systems. When I graduated, I wanted to work on experimental realizations of quantum devices and moved to France for postgraduate education. I developed SiGe quantum devices for the implementation of long-lived qubits that build on integrated circuit technology. I arrived in Oxford more than six years ago as a postdoc, and shortly after that was awarded a Marie Skłodowska Curie Fellowship. Since then, I have been awarded a Templeton Independent Research Fellowship and am currently a Royal Society University Research Fellow. At the University of Oxford I first worked on radio-frequency reflectometry for spin-qubit readout and carbon nanotube electromechanics. My group now focuses on machine learning for qubit scalability and on developing quantum devices to study quantum thermodynamics.
This has been a great journey so far, and I was lucky enough to benefit from the support of amazing people. I realized that mentors and role models are essential. Especially for women in science, I think we all have to work harder at creating mechanisms for inspiration and guidance and, most important, mechanisms that counterbalance the biases and extra challenges that women face.
What are the fundamental concepts you tackle in the area of quantum thermodynamics?
I aim at studying the laws of thermodynamics in quantum devices, for which fluctuations are important and quantum effects arise. Quantum thermodynamics is a rapidly advancing field of physics, but its theoretical development is presently far ahead of experimental implementations. I want to develop an experimental platform with the necessary ingredients to explore questions such as: what is the efficiency of a quantum engine? Understanding thermodynamics in the quantum arena will be key for the construction of nanomachines and for energy harvesting. It will also improve the engineering basis of quantum technologies by facilitating fully informed choices on device design and optimization, and it may reveal entirely novel technologies.
How do the Zurich Instruments UHFLI Lock-in Amplifiers help you in your work?
We use our UHFLIs for radio-frequency reflectometry readout of semiconductor devices. We achieved a record sensitivity to the change in quantum capacitance associated with qubit states, which is key for fast and accurate qubit readout. Thanks to the UHFLI, the experiment was faster and simpler.
We also used our UHFLIs to detect coherent nanomechanical oscillations driven by single-electron tunneling in a suspended carbon nanotube. These oscillations have not been observed before because their detection required a challenging combination of coupling strength and measurement speed. Our experiment achieved both requirements, allowing us to connect the physics of back-action with that of lasers. For this experiment, we used the PID Controller option to generate a correction voltage for stabilizing the mechanics.
How big is your group, and what other quantum-information-related efforts are ongoing in Oxford?
I lead a group of 3 master students, 7 PhD students and 4 postdocs. The University of Oxford is part of the Networked Quantum Information Technologies Hub, which has now entered its second phase. There are efforts on superconducting qubits, trapped ions, NV centers, and the theory of quantum algorithms and quantum architectures to cite a few.
What do you do when you don't work?
I used to do artistic skating, but a few years ago I took up ballet. I didn't do any classical dance as a child, so it is a real challenge! I also go back to Argentina every year to have 'mate' (a traditional drink similar to tea) with my family and friends.