The beating heart of IBM’s quantum computer is a chip no bigger than a quarter. These extravagant machines promise to solve difficult problems that stump today’s best classical computers. The chip itself is only one part of a bigger puzzle. Unlike the portable laptops that people use in everyday life, the computing infrastructure that supports the work a quantum chip does is layered like a Russian doll, with convoluted interconnections within a Rube-Goldberg-like contraption.
However, even with its complicated construction and mind-boggling design, a quantum computer is still a machine that performs operations by employing both hardware and software. Some of those actions are similar to those performed by classical computers. Curious to understand how they function? Popular Science got a look around the quantum center in IBM’s Yorktown Heights, New York campus. Take a closer look at what’s happening inside—starting with something called the qubit (more on what that is in a moment) and zooming out, bit by bit.
To exhibit quantum qualities, objects have to either be very small or very cold. For IBM, this layered chandelier-like structure, which looks like an upside-down gold steampunk wedding cake, is called a dilution refrigerator. It keeps their qubits cool and stable, and is the infrastructure that the company created for this 50-qubit chip. It contains multiple plates that get successively colder the closer they are to the ground. Each plate is a different temperature, with the very top layer sitting at room temperature.
The quantum processor is mounted to the lowest, and coldest, plate of the dilution refrigerator that gets to a temperature around 10 to 15 milli-Kelvin, which is roughly –460 degrees F. The first stage of cooling involves large copper pieces seen draping down in the top layer that are connected to cold heads as part of a closed-cycle helium cryocooler. More tubes feeding into the lower levels introduce another closed cycle of cryogenic material, made up of a mix of helium isotopes.
In the back of the housing structure are the hidden support infrastructure for the chandelier. This includes the gas handling system that supports the cryogenic infrastructure, as well as pumps and temperature monitors. And then there are the custom-built classical control electronics. When users run a program through IBM’s quantum cloud service, Qiskit, they are effectively orchestrating a set of gates and their circuits. Those get turned into microwave pulses that are appropriately sequenced, aligned, and distributed out into the system to control the qubits. And the readout pulses retrieve the states of the qubits, which are translated back into binary values and returned to the users.
Qubits and an ‘artificial atom’
Classical computers represent information using binary one-or-zero bits. In the case of quantum, information is represented through qubits, which can come in a combination of zero and one. This is a phenomenon referred to as superposition. “You have superposition all the time in the real world. Music is a superposition of frequencies, for example,” says Zaira Nazario, the technical lead of theory, algorithms, and applications at IBM Quantum. Because it’s a waveform, it provides an amplitude of zero and one. That means it comes with a phase, and like all waves, they can interfere with one another.
The superconducting qubits sit on the chip and have been packaged into something like a printed circuit board. Wires and coaxial cables for input and output signals protrude off the circuit board. In newer models of higher-qubit chips, IBM has been working towards more compact solutions involving wiring and integrated components to be more efficient with space. Having less clutter means that the components would be easier to cool. Currently, it takes about 48 hours to completely cool down a quantum computer to the desired temperatures.
For the quantum computer to function correctly, each of the plates must be thermally shielded and isolated to prevent black body radiation from affecting it. Engineers vacuum-seal the whole device to keep out unwanted photons as well as other electromagnetic radiation and magnetic fields.
Qubits are controlled with microwave signals that range from 4 to 7 gigahertz. Classical electronics generate microwave pulses that travel down the cables to bring the input signals to the chip and carry the output signals back. As the signal travels down the chandelier, it goes through components like filters, attenuators, and amplifiers.
IBM works largely with superconducting qubits. They’re little pieces of metal that sit on the wafer, which is used to make the chip. The metal is made up of superconducting materials like niobium, aluminum, and tantalum. A Josephson junction, made by layering a very thin insulator between two superconducting materials, provides the essential nonlinear element needed to turn a superconducting circuit into a qubit.
“What we’re building is quantum examples of oscillators,” says Jerry Chow, director of quantum infrastructure at IBM. Oscillators convert a direct current from a power source (in this case, microwave photons) into an alternating current, or a wave.
Unlike typical harmonic oscillators, a nonlinear oscillator gives you an unequal spacing of energy levels, Chow says. “When you have that, you can isolate the lowest two to act as your quantum zero and your quantum one.”
Think of a hydrogen atom. From a physics standpoint, it has a set of energy levels. The right wavelengths of light hitting this atom could promote it to different states. When microwaves hit the qubit, it is doing something similar. “You effectively have this artificial atom,” Chow explains. “We have a quantum of energy, which we move around by putting the right amount of microwave photon at a certain pulse for a certain duration to either excite or de-excite a quantum of energy within this nonlinear microwave oscillator.”
In a classical computer, there’s an on-state (one), and an off-state (zero). For a quantum computer, the off-state is the ground state of the artificial atom. Adding a pulse of a particular microwave photon of energy would excite it, promoting it to one. If the qubit is hit again with that pulse, it would be brought back down to ground state. Let’s say it takes 5 gigahertz for 20 nanoseconds to promote a qubit fully to the excited state—if you were to halve the amount of energy or halve the amount of time, you would actually drive a superposition state, Chow says. That means if you were to measure the state of the qubit with a resonator, you would have a 50 percent chance of it being in zero, and 50 percent chance of it being in one.
Users can play around with the circuit elements, pulse frequencies, duration, and energy between different qubits to couple them, swap them, or perform conditional operations like building entangled states and combining single qubit operations to perform universal computation across the entire device. When waves intersect, it can either amplify or deconstruct the message.
What are qubits good for?
The practical uses for quantum computers have evolved over the last couple of years. “If I look at what people were doing with the system in that 2016, 2017, 2018 timeframe, it was using quantum to research quantum… condensed matter physics, particle physics, things like that,” says Katie Pizzolato, director of strategy and applications research at IBM Quantum. “The key part of this is going to be taking classical resources and making them quantum-aware. We have to make the people who are experts in their field understand where to apply quantum, but not be quantum experts.”
The interest IBM has been seeing in terms of quantum problems posed to their machines can be grouped into three buckets: chemistry and materials, machine learning, and optimization (finding the best solution to a problem from a set of possible options). The key is not to use a quantum computer in every part of the problem—but on the parts that are hardest.
The team at IBM has been continuously searching for real-world problems that are hard for classical computers to solve due to their structure or the math they involve. And there are many interesting places to look for them.
Classical computers solve basic math problems using binary logic and circuit components such as adders. However, quantum computers are really great at doing linear algebra—multiplying matrices, and representing vectors in space. This is due to unique features in their design. It allows them to perform functions like factoring relatively easily—a problem that is extremely difficult for a classical computer because of the exponentially increasing number of variables and parameters and the interactions between them. “There’s structures within that factoring problem that allow you to take advantage of the entanglements, all the things that you get with these devices. That’s why it’s different,” Pizzolato says.
And with chemistry and materials problems, qubits are just better at simulating properties like bonds and connected electrons.
“We’re thinking about what types of things you can map to quantum circuits that are not simulable classically, and then what do you do with them,” says Pizzolato. “A lot of the algorithm discussion is how do I exploit the underlying mechanics of this device. How do you map on higher dimensional spaces and how do you use this coordination and multiplications of these matrices to rise up the answer that you want.”
Remember, qubits can have a value of zero, one, or a combination of the two. Since qubits are waveforms, engineers can rotate the zero or one to give it a negative amplitude. Qubits can also get entangled—a unique quantum mechanics property that doesn’t have a classical analog. Entangled qubits can contain information not just in the zeros and ones themselves, but also in the interactions between all of them. Also, there are gates in quantum circuits that can rotate the qubit to change its phase, and oscillators can entangle those qubits.
“The art of doing a quantum algorithm is how you manipulate all of those entangled states and then interfere in a way that the incorrect amplitudes cancel out, and the amplitudes of the correct one come forward, and you get your answer,” Nazario says. “You have a lot more room to maneuver in a quantum algorithm because of all these entangled states and this interference compared to an algorithm that only allows you to flip between zeroes and ones.”
Qiskit, IBM’s open-source development kit for quantum computers, contains information on various types of quantum algorithms and programs at different levels of detail.
Still find it tricky to visualize what the qubit is doing? Let’s zoom out to some examples of how IBM’s partners are using quantum computers. For example, biopharmaceutical company Amgen is looking to use quantum computers and machine learning to predict the patients who would be best suited for a drug trial based on health records and other factors.
And Boeing is applying quantum computing to analyze corrosion coefficients on airplanes. Airplane wings require a certain density of materials. Engineers make them with different layers of various materials, but need help figuring out how they should arrange the layers in a way that makes the wings stronger, cheaper, and lighter. This boils down to a combinatorial optimization problem.
Goldman Sachs has been using it for options pricing. “These are very complex operations that are very computationally expensive. And they have complex distributions,” Nazario says. It has to do with calculating the derivatives of the variations in those options (a linear algebra operation), which will tell them about risks.
Finally, in the natural sciences, research groups have been interested in using quantum computers to study photosynthesis.
Building in parallel
Although IBM has been steadily increasing the processor size for its quantum computers, and building a community of partners from industry, national government hubs, and academic institutions, the company is still figuring out the best ways to move forward both with the hardware and software.
The company has previously said that it would have a machine capable of quantum advantage (in which it can reliably and accurately solve a problem better than a classical computer) ready by 2025. That means that in addition to developing new components, it needs to iron out some problem areas, and make what already works well, more efficient.
“This is a big part of the focus of the software. We’ve recognized that a lot of the tools, the error mitigation tools, the intelligent orchestration, the idea of circuit-knitting, how do we break down the problems to extend what we can do on the quantum computer, these are becoming much more prolific in how we can push the technology,” says Pizzolato.