Quantum Supremacy Is Unlikely, Scientist Says
Google announced this fall to much fanfare that it had demonstrated “quantum supremacy” — that is, it performed a specific quantum computation far faster than the best classical computers could achieve. IBM promptly critiqued the claim, saying that its own classical supercomputer could perform the computation at nearly the same speed with far greater fidelity and, therefore, the Google announcement should be taken “with a large dose of skepticism.”
This wasn’t the first time someone cast doubt on quantum computing. Last year, Michel Dyakonov, a theoretical physicist at the University of Montpellier in France, offered a slew of technical reasons why practical quantum supercomputers will never be built in an article in IEEE Spectrum, the flagship journal of electrical and computer engineering.
So how can you make sense of what is going on?
What’s a quantum computer?
To understand why, you need to understand how quantum computers work since they’re fundamentally different from classical computers.
A classical computer uses 0s and 1s to store data. These numbers could be voltages on different points in a circuit. But a quantum computer works on quantum bits, also known as qubits. You can picture them as waves that are associated with amplitude and phase.
Qubits have special properties: They can exist in superposition, where they are both 0 and 1 at the same time, and they may be entangled so they share physical properties even though they may be separated by large distances. It’s a behavior that does not exist in the world of classical physics. The superposition vanishes when the experimenter interacts with the quantum state.
Due to superposition, a quantum computer with 100 qubits can represent 2100 solutions simultaneously. For certain problems, this exponential parallelism can be harnessed to create a tremendous speed advantage. Some code-breaking problems could be solved exponentially faster on a quantum machine, for example.
There is another, narrower approach to quantum computing called quantum annealing, where qubits are used to speed up optimization problems. D-Wave Systems, based in Canada, has built optimization systems that use qubits for this purpose, but critics also claim that these systems are no better than classical computers.
Regardless, companies and countries are investing massive amounts of money in quantum computing. China has developed a new quantum research facility worth US$10 billion, while the European Union has developed a €1 billion ($1.1 billion) quantum master plan. The United States’ National Quantum Initiative Act provides $1.2 billion to promote quantum information science over a five-year period.
Breaking encryption algorithms is a powerful motivating factor for many countries — if they could do it successfully, it would give them an enormous intelligence advantage. But these investments are also promoting fundamental research in physics.
Many companies are pushing to build quantum computers, including Intel and Microsoft in addition to Google and IBM. These companies are trying to build hardware that replicates the circuit model of classical computers. However, current experimental systems have less than 100 qubits. To achieve useful computational performance, you probably need machines with hundreds of thousands of qubits.
Noise and error correction
The mathematics that underpin quantum algorithms is well established, but there are daunting engineering challenges that remain.
For computers to function properly, they must correct all small random errors. In a quantum computer, such errors arise from the non-ideal circuit elements and the interaction of the qubits with the environment around them. For these reasons the qubits can lose coherency in a fraction of a second and, therefore, the computation must be completed in even less time. If random errors — which are inevitable in any physical system — are not corrected, the computer’s results will be worthless.
In classical computers, small noise is corrected by taking advantage of a concept known as thresholding. It works like the rounding of numbers. Thus, in the transmission of integers where it is known that the error is less than 0.5, if what is received is 3.45, the received value can be corrected to 3.
Further errors can be corrected by introducing redundancy. Thus if 0 and 1 are transmitted as 000 and 111, then at most one bit-error during transmission can be corrected easily: A received 001 would be a interpreted as 0, and a received 101 would be interpreted as 1.
Quantum error correction codes are a generalization of the classical ones, but there are crucial differences. For one, the unknown qubits cannot be copied to incorporate redundancy as an error correction technique. Furthermore, errors present within the incoming data before the error-correction coding is introduced cannot be corrected.