Quantum Computers Retain Memory of Errors, Researchers Map Noise Across Time for First Time

A group of Australian and international physicists has produced what they describe as the first complete picture of how errors unfold over time inside a quantum computer. The work was led by Dr. Christina Giarmatzi of Macquarie University, and the full study has been published in the journal Quantum.

The central finding is that the small errors that plague these machines are not simply random flickers that appear and vanish. They can linger, change, and become linked across different moments in time. Giarmatzi puts it this way: the computer retains a memory of its errors, and those errors can be classical or quantum depending on how they are connected. Many quantum protocols, she says, assume the opposite, that the machine has no such memory, a property known as being Markovian. In her words, that assumption is simply not true.

That matters because this kind of time-linked behavior is one of the main obstacles to building practical, large-scale quantum computers. If noise carries forward from one step of a computation to the next, tools designed around memoryless noise will miss part of the problem.

The team ran its experiments on superconducting quantum processors. Some of the work was done on hardware in the lab at the University of Queensland, and some through IBM's cloud-based quantum computers. What they reconstructed was the evolution of a quantum process across multiple points in time, something Giarmatzi says had not been done before. It shows, she said, not only when noise occurs but how it carries through time.

Previous attempts to map quantum systems over time ran into the same obstacle. Once you measure a system midway through an experiment, you cannot freely set it up again for the next step, because the correct preparation depends on whether the measurement came out as 0 or 1. The new method gets around this with a twist: it assumes that half the time the result was 1 and half the time it was 0, then uses software to work backward through the data and determine what state the system was in.

Co-author Dr. Fabio Costa of Nordita in Stockholm said the hardware was already capable of this. What the team worked out, he said, was how to actually prepare the system after a mid-circuit measurement.

Applying the method to today's machines, the researchers found that even the best of them show subtle but significant time-linked noise patterns. Some of that noise is quantum in nature and originates from neighboring qubits on the same chip. The team says the approach should extend beyond superconducting chips to other platforms, including trapped ions and spin qubits.

The researchers have made their experimental data and code openly available. Tyler Jones, who worked on the project as a Ph.D. student at the University of Queensland, said it is rewarding to see theoretical models brought to life on real hardware, especially when they help develop the hardware itself. Robust characterization of time correlations, he said, is needed on the path to powerful quantum machines.

The practical payoff is expected in better tools for characterizing noise and correcting errors, a necessary step toward dependable, fault-tolerant quantum computers. With the data and code public, other groups can now test the method on their own devices.