Picture a choir where every singer is assigned a fixed note at birth and can never change it. As long as the voices stay spread across the scale, harmonies work. But let two neighbors land on nearly the same pitch and the music turns to mush. That, in miniature, is the problem haunting one of the most popular designs in quantum computing: the fixed-frequency transmon.
A pitch set by a tiny junction
A transmon qubit is essentially an electrical oscillator, and its resonant frequency is set by the properties of its Josephson junction, a sliver of insulator sandwiched between two superconductors. During fabrication, the thickness and area of that junction determine how much current tunnels through it, and that in turn fixes the qubit's frequency, usually somewhere around 5 gigahertz.
Unlike tunable qubits, which carry a magnetic knob that shifts their frequency on command, fixed-frequency transmons have no such dial. That is a deliberate choice. A qubit that cannot be tuned also cannot be jostled by stray magnetic noise coming through that same knob, so it tends to hold its quantum state longer. IBM built much of its hardware around this bargain, trading flexibility for stability.
Why neighbors need to disagree
The catch shows up when two qubits need to interact. The cross-resonance gate, the workhorse entangling operation on these chips, works by driving one qubit at the frequency of its neighbor. For that trick to produce a clean, controllable interaction, the two qubits must sit at specific frequency offsets from each other. Too far apart and the gate is weak and slow. Too close and something worse happens.
When two coupled qubits drift into near-resonance, a frequency collision, their energy levels start to mix in ways the control electronics never intended. A pulse meant to nudge one qubit leaks into the other. The gate loses fidelity, and in the worst cases the qubit climbs out of its two computational states entirely, bleeding information into higher energy levels. A single bad pair can poison the circuits that run across it.
The yield problem
Here is where fabrication physics collides with ambition. Junctions cannot be manufactured to land on an exact frequency. There is always scatter, historically on the order of a couple of percent, which translates to tens or even hundreds of megahertz of uncertainty per qubit. On a small chip with a handful of qubits, you can usually get lucky and avoid collisions. Scale up to a heavy-hex lattice with over a hundred qubits, each with two or three neighbors, and the odds that at least one pair lands in a forbidden zone climb toward certainty.
This is a yield problem in the classic chipmaking sense. Fabricate a wafer full of processors and only a fraction come out with a clean frequency map. For a field trying to grow qubit counts by orders of magnitude, throwing away most of your chips is not a sustainable plan.
Retuning after the fact
The fix that moved the field forward was learning to adjust a junction after it was already built. IBM developed a technique, sometimes called laser annealing, that uses a focused laser to gently heat an individual junction and nudge its tunneling resistance. Change the resistance and you change the frequency. By measuring where every qubit actually landed, computing a target map that avoids collisions, and then dialing each junction toward its assigned slot, engineers can rescue a chip that would otherwise be scrap.
The approach turns a random scatter of frequencies into something closer to a deliberate arrangement. It does not make the junctions perfect, but it tightens the distribution enough to push collision probabilities down and yields up. Competing strategies attack the same problem from other directions: tunable couplers that let designers sidestep strict frequency requirements, and tunable qubits that accept flux noise in exchange for the freedom to move voices out of each other's way.
Why it matters for the roadmap
Frequency collisions sit at the awkward intersection of physics and manufacturing, and they are one reason scaling superconducting machines is harder than simply printing more qubits. Every added qubit is another voice that must find room on a crowded dial, and every new neighbor is another chance for a clash. Solving it reliably, whether through post-fabrication tuning, better junction uniformity, or circuit designs that forgive imperfection, is part of the quiet engineering work that has to succeed before fault-tolerant processors with thousands of qubits become routine rather than lucky.
