Quantum computers were supposed to be forgetful machines. Each operation should start fresh, with errors popping up randomly and independently, like static on a radio. But it turns out these hyper-sensitive devices have a memory problem. Errors linger, echo forward in time, and link together in ways that can quietly sabotage an entire calculation before it finishes.
A team of Australian researchers has produced the first complete map of how this happens. Using superconducting quantum processors chilled to 15 millikelvin—colder than deep space—they watched qubits whisper mistakes to each other across time, like a digital game of telephone. This phenomenon, called non-Markovian noise, contradicts a foundational assumption baked into most quantum computing theory: that what goes wrong now has nothing to do with what went wrong a moment ago.
The discovery matters because the entire field has been designing error-correction schemes around the idea of independent, random glitches. If errors are actually correlated across time, those schemes may fall short. Christina Giarmatzi, who led the work at Macquarie University, puts it plainly: “A lot of quantum protocols assume quantum computers have no such memory (known as Markovian) but that’s simply not true.”
Watching Errors Unfold Across Multiple Moments
The breakthrough involved a technique called multi-time quantum process tomography. In the past, scientists could only capture snapshots of a quantum system at one or two points in time, making it impossible to see how an error at the start of a process might shape the outcome at the end. The new method reconstructs the entire history of a quantum operation across three or more time steps.
That’s harder than it sounds. Measuring a quantum system mid-experiment usually collapses its fragile state, making it difficult to restart cleanly. The team got around this with clever post-processing: instead of resetting the system based on each individual measurement, they statistically reconstructed what must have been happening across many runs. “The hardware could do it,” said co-author Fabio Costa. “What we figured out was how to actually prepare the system after a mid-circuit measurement.”
They tested the approach on processors at the University of Queensland and through IBM’s cloud-based quantum systems. Even the most stable IBM chips showed clear signs of time-linked noise. On the Queensland chip, where qubits were more tightly coupled, the memory effects were more pronounced. Some of that noise was classical—slow drifts in temperature or electronics. But some was genuinely quantum, tied to interactions between neighboring qubits.
“We can think of it as quantum computers retaining memory of the errors, which can be classical or quantum depending on the way these errors are linked,” Giarmatzi explains.
Rethinking Error Correction From the Ground Up
For engineers trying to build fault-tolerant quantum computers, this complicates the roadmap. Error-correction schemes often assume that fixing today’s error is sufficient. If errors echo forward in time, those strategies need redesigning. Think of it like proofreading a document where earlier typos quietly reappear later, even after you fix them. You need a strategy that understands the pattern, not just individual mistakes.
The team used mathematical tools that treat the process as a “density operator over time,” quantifying exactly how far each real device deviated from an ideal memoryless one. Even the best-performing machines showed subtle but measurable time correlations. That finding isn’t limited to superconducting qubits—the same framework could apply to trapped ions, spin qubits, and other quantum platforms.
The researchers have made their code and data openly available, inviting the broader quantum community to build on the work. As we try to pack more qubits onto a single chip, these correlated errors will only become a bigger obstacle. Understanding how noise evolves over time is a vital step toward machines that can correct their own mistakes as they happen. For now, the field has a clearer, if less comforting, picture of what it’s up against: quantum computers that can’t stop remembering what they should forget.
Quantum: 10.22331/q-2025-12-02-1582
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.