What do pharmaceutical drugs, rechargeable batteries, and solar cells have in common? They share the potential to reap great benefits in design from the simulation of their behavior at the level of quantum mechanics. The trouble is, simulating the quantum mechanics of these systems can be extraordinarily difficult for even the largest supercomputers. The simulation entails keeping track of and performing calculations on a number of variables that grow exponentially with the number of electrons in each molecule.

For many systems, a full quantum-mechanical simulation would take even a supercomputer thousands of years to complete. This difficulty of simulating quantum systems inspired physicist Richard Feynman to propose, in the early 1980s, the development of a computer that operates quantum mechanically in a fundamental way.

“Feynman’s idea was that it would be natural to use a computer that itself was quantum mechanical to simulate quantum-mechanical systems,” says Peter L. McMahon, Applied and Engineering Physics. “A quantum computer would not suffer from the exponential size of the state of the system being simulated because it can be mapped to the internal state of the quantum computer, which is also exponentially large.”

What’s Next in Computing? Exploring the Possibilities

McMahon’s research revolves around using physical systems to do computation in better ways than the current means, employing conventional computers. Along with exploring the possibility of quantum computers, the McMahon lab also delves into alternative classical computer architectures, such as photonic neural networks, in which computations are performed using light instead of electrons. McMahon and his colleagues see their research as exploring what might come next, given that the computer industry is close to hitting the limit for scaling down the size of transistors in silicon chips, which is the current primary source of improvement in computer performance.

Rather than trying to find ways to push the capabilities of current computer processors, the researchers are addressing the problem from a more fundamental, applied physics perspective. “Let’s stand back and ask, ‘If we had to do this all over again, what would we do?’” McMahon says. “What is really the most effective way to make a processor?”

The Quantum Way

In the case of quantum computing, there are many technological candidates for creating the basic units of information, called quantum bits or qubits. The McMahon lab is currently focusing on two: superconducting circuits and photonics.

“In both these technologies, we have a bunch of ideas we’d like to try out,” McMahon says. “In our photonic approach, the goal is to construct a prototype quantum computer that is restricted in what it can do but will still be a machine that a classical computer could not easily simulate. In the near term, this will teach us about how feasible our photonic architecture is to realize in practice. And in the more distant future, we can potentially build machines that are universal quantum computers—ones capable of running any algorithm.”

“How do you get photonic qubits to talk to each other? It’s possible to make them do so, but they don’t naturally want to.”

To create quantum computers, McMahon and his colleagues will need to overcome innate disadvantages of the technologies they are using. Photonic qubits, for example, generally don’t interact with each other. “That is a very big problem for quantum computation,” McMahon says. “How do you get photonic qubits to talk to each other? It’s possible to make them do so, but they don’t naturally want to—at least not in most situations.”

Superconducting Circuits

Superconducting circuits, on the other hand, present their own problems. On the surface, they might appear to be an easy technology to harness; they look remarkably similar to classical electronic circuits. A major catch is that they have to be operated at a temperature near absolute zero.

“If you don’t run them at very low temperature, they don’t behave as quantum-mechanical systems,” McMahon says. “In other words, the quantum-mechanical effects get washed out.” Keeping the circuits cold enough while increasing their number and the complexity of computations that they perform is a tall order.

What Can You Do with a Quantum Computer?

McMahon also wants to explore the potential uses for a quantum computer, including and beyond Feynman’s original suggestion that they be used for simulating quantum systems. Toward that end, he is collaborating with Thomas Hartman and Paul H. Ginsparg, Physics. “We’re trying to find something useful we can do with a near-term quantum computer that would answer a question in quantum gravity, or high-energy physics more generally, that couldn’t be answered otherwise,” McMahon says. “For instance, can we simulate a model of a black hole on a quantum computer? Would that be useful? We don’t know if we’ll find anything, but it’s very interesting to try.”

In the future, quantum computers may be just one of many different processors best suited to do specialized tasks, McMahon says. “Can we find a way of doing computing that maybe doesn’t solve all the tasks we have, but solves some of the more important ones more effectively?” he asks. He points out that many research groups, including his team, are trying to find ways to make processors that are dedicated to machine learning.

Using Light versus Electrons for Processors

One of the core tasks a processor executes for machine learning—and in particular for neural networks—is matrix-vector multiplication. These multiplications simulate information transmission between layers of artificial neurons. Classical computers that run modern neural-network algorithms use a huge part of processing power to perform matrix calculations. McMahon is tackling matrix-vector multiplication by building photonic processors that perform the multiplications and additions with light instead of electrons—a continuation of his pursuit of physical systems that are naturally suited for computing.

“If you think of a beam of light, you can interpret different parts of the beam as encoding different elements of a vector, or equivalently in the case of neural networks, different neuron values,” McMahon says. “As a consequence of how light propagation works, if you send light through a designed medium or optical apparatus that makes different parts of an incoming beam scatter in different directions by different amounts, you can describe what happens as a matrix-vector multiplication. So just by shining light on a carefully designed optical apparatus, you can perform a desired matrix-vector multiplication.”

An Interminable Fascination

McMahon has had a lifelong fascination with computing, but reading about quantum computation as a freshman in college sparked his interest in unconventional physical computers.

“It was a revelation,” he says. “I was fascinated by the idea that physical laws, such as the laws that govern the motion of electrons, have a decisive impact on what we can compute and how quickly. I’ve never let go of that. I’m drawn to this intersection between physics and computation.”