Apr 1 2020
A premium race car engine will deliver excellent performance only when all its components are adjusted and are accurately working together.
Quantum Computers" />
This artist’s conception shows how the research team used artificial intelligence (AI) and other computational techniques to tune a quantum dot device for use as a qubit. The dot's electrons are corralled by electrical gates, whose adjustable voltages raise and lower the “peaks” and “valleys” in the large circles. As the gates push the electrons around, sensitive measurement of the moving electrons creates telltale lines in the black and white images, which the AI uses to judge the state of the dot and then make successive adjustments to the gate voltages. Eventually, the AI converts a single dot (leftmost large circle) to a double dot (rightmost), a process that takes tedious hours for a human operator. Image Credit: B. Hayes/National Institute of Standards and Technology.
The same concept applies to the processor integrated into a quantum computer, whose fragile bits should be tuned optimally before it can execute a calculation. But who would be the right mechanic to perform this quantum tune-up task?
According to a group that comprises researchers from the National Institute of Standards and Technology (NIST), the quantum tune-up job can be performed by artificial intelligence (AI).
Published in the Physical Review Applied journal, the researchers’ paper shows how an AI can be trained to make an interconnected set of modifications to minute quantum dots. These quantum dots are among the numerous potential devices used for developing the quantum bits, also known as qubits,” that would create the switches in the processor of a quantum computer.
Accurate adjustment of the quantum dots is important to convert them into accurately functioning qubits, and, to date, this task had to be performed meticulously by human operators and involves extensive work to develop just a few qubits for one calculation.
A viable quantum computer containing several interacting qubits would need much more dots—and modifications—than a person could handle; this means, the researchers’ achievement may soon bring quantum dot-based processing from the world of theory to engineered reality.
Quantum computer theorists imagine what they could do with hundreds or thousands of qubits, but the elephant in the room is that we can actually make only a handful of them work at a time. Now we have a path forward to making this real.
Justyna Zwolak, Mathematician, National Institute of Standards and Technology
Usually, a quantum dot includes electrons that are limited to a constricted box-like space within a semiconductor material. The walls of the box are formed by many metallic electrodes (the so-called gates) above the surface of the semiconductor material and to which electric voltage is applied. This set-up influences the number and position of electrons in the quantum dot. Based on their position in relation to the dot, the gates regulate the electrons in various ways.
Individuals who wish to make the dots perform as they want—that is, make the dots behave as a kind of qubit logic switch or another, for instance—the gate voltages should be adjusted to just the exact values. This adjustment is manually performed by quantifying the currents that flow via the quantum dot system, then slightly altering the gate voltages, and finally rechecking the current.
If individuals involve more dots (and gates), they would find it harder to adjust all the dots at the same time, so that they get qubits that accurately function together. To sum up, human mechanics would not feel bad about losing to a machine because this is not a gig.
It’s usually a job done by a graduate student. I could tune one dot in a few hours, and two might take a day of twiddling knobs. I could do four, but not if I need to go home and sleep. As this field grows, we can’t spend weeks getting the system ready—we need to take the human out of the picture.
Tom McJunkin, Study Co-Author and Graduate Student, Department of Physics, University of Wisconsin-Madison
However, McJunkin used only pictures to look at, while making adjustments to the dots: The data McJunkin worked with was in the form of visual images, which the researchers realized that AI can detect very well.
For automated image classification, AI algorithms, known as convolutional neural networks, have become the sought-after method, provided they are subjected to plenty of examples of what they had to identify.
Hence the group’s Sandesh Kalantre, under the guidance of Jake Taylor from the Joint Quantum Institute, developed a simulator that is capable of producing an unlimited number of images of quantum dot measurements that could be fed to the AI as a kind of training exercise.
We simulate the qubit setup we want and run it overnight, and in the morning we have all the data we need to train the AI to tune the system automatically. And we designed it to be usable on any quantum dot-based system, not just our own.
Justyna Zwolak, Mathematician, National Institute of Standards and Technology
Using a setup of two quantum dots, the researchers started in a small way and they confirmed that within specific limitations, their trained AI could automatically adjust the system to their required setup. While it was not perfect, the researchers detected many areas that they need to further work on to enhance the reliability of the method—and they simply cannot utilize it to adjust an infinite number of interconnected quantum dots as yet.
However, even at this preliminary stage, the practical power of the technique cannot be denied, enabling an expert scientist to devote valuable time elsewhere.
“It’s a way to use machine learning to save labor, and—eventually—to do something that human beings aren’t good at doing. We can all recognize a three-dimensional cat, and that’s basically what a single dot with a few properly-tuned gates is. Lots of dots and gates are like a 10-dimensional cat. A human can’t even see a 10D cat. But we can train an AI to recognize one,” Zwolak concluded.