Susceptibility to interference is the weakest point of quantum computers. Classical solutions cannot be used to protect against quantum computing errors. Either extreme redundancy or sophisticated architectures come to the rescue. Amazon has bet on the latter, promising to pave the way for practical quantum platforms.
Image source: AI generation Grok 3/3DNews
In the new work, Amazon researchers demonstrate a combination of fault-tolerant hardware and an error-correcting circuit. At least Amazon’s qubits don’t require new physics, as Microsoft’s new error-tolerant Majorana fermion quantum architecture does. Simply put, Amazon is using tried-and-true solutions in a novel way, and that approach has clear promise.
Amazon’s quantum computing platform combines two different types of hardware qubits to improve the stability of quantum information—the quantum states of qubits. The idea is that one type of qubit is resilient to one type of error, while the other can be used to implement error-correcting code that detects emerging problems in the computational (algorithmic) chain. Amazon’s demonstration was not large-scale, but at a basic level it showed the potential validity of the approach.
In a normal computer, there is only one type of error: a bit flip, where circumstances can cause a 0 to become a 1 and vice versa. As with most things in quantum computing, qubits are much more complicated. Since they contain probabilities rather than binary values, if a qubit flips erroneously, returning it to its true state will be difficult.
But bit flipping isn’t the only problem that can arise. Qubits can also suffer from so-called phase-swapping errors. These have no analogues in classical computers, but they will also prevent quantum computers from working properly.
Back in 2021, Amazon employees showed that qubits can be created that are extremely resistant to one type of error — bit flipping. The new work confirms this concept and, in fact, becomes the basic one. So-called cat qubits are used for this. The qubits are named after Schrödinger’s cat, which is in a state of superposition.
In fact, this is a group quantum state, spread over several elementary particles, in particular, photons. Erroneous switching of one photon in a group does not affect the quantum state of the group. Such a qubit is conditionally not subject to bit switching errors and they can be forgotten, which means that the architecture of the calculator will be simpler a priori. It will be necessary to correct not two, but only one type of errors associated with phase switching. However, the more photons in a group, the higher the probability of erroneous phase switching, and when scaling the platform, something will also have to be done about this.
Detecting and correcting phase shift errors using a repetition code. Image source: Nature
It was because of these phase flips that a second set of qubits, called transmons, were introduced. Transmons are a widely used type of qubit based on a loop of superconducting wire connected to a microwave cavity, and are used by companies like IBM and Google. Superconducting transmons were used to link the cat qubits, allowing the team to create an error-correcting logical qubit using a simple error-correcting code called a repetition code.
Transmons that found errors raised red flags. Image source: Nature
In the above schematic, each cat qubit is connected to a neighboring transmon. This allows the transmons to monitor what is happening in the cat qubits using what are called weak measurements. Such measurements do not destroy the quantum state, as in a full measurement, but they allow changes in neighboring cat qubits to be detected and the information needed to correct errors if they occur.
Thus, the combination of these two methods means that almost all errors that occur are phase shift errors that are detected and corrected, since bit flip errors should not occur in the system by definition, although in fact they do. However, if two types of errors had to be corrected simultaneously, as was common until now, a lot of physical qubits were required to implement each logical qubit. In the case of Amazon’s design, only one type of error is supposed to be corrected, and each logical qubit will require significantly fewer physical qubits.
In the study, Amazon showed that when comparing a chain of three cat qubits and two transmons to a chain of five cat qubits and four transmons, the error rate decreased as the architecture became more complex. Typical quantum systems typically do the opposite: the more qubits, the higher the error rate, at least at large scales. Amazon thus claims the advantage of its approach, which will allow it to scale up the number of physical and logical qubits while keeping the probability of computational errors from growing.
In reality, everything is much more complicated. Transmons themselves are subject to both types of errors, and a failure of one of them will collapse the entire calculation. Also, cat qubits cannot boast a complete absence of bit-flip errors, and if such an error occurs, the calculations will also be meaningless. However, the idea proposed by Amazon has potential and the right to further development.