Qubits forget — fast
A classical bit sits patiently as a 0 or a 1 for years. A qubit holds a fragile in-between state, and the surrounding world is constantly nudging it. After a tiny moment — often less than a thousandth of a second — the delicate state smears out and the information is gone. That fading is called losing coherence, and how long a qubit survives before it happens is its coherence time.
So the first job of a quantum chip is not really computing — it is keeping the qubit calm long enough to do anything at all. Almost every design choice on the chip, from the materials to the wiring, is ultimately a fight to buy a few more microseconds of memory.
The disturbance hides in the materials
Where does the nudging come from? A lot of it is not from the outside world at all — it is baked into the chip itself. The metals, the insulating layers, and especially the surfaces carry tiny imperfections. Two of them matter most.
The first is the TLS defect: a single atom or stray molecule, often sitting in a thin oxide skin on the surface, that can flip back and forth like a microscopic switch. When its rhythm happens to match the qubit's, it quietly steals the qubit's energy. The second is dielectric loss: the insulating materials are not perfectly lossless, so a little of the qubit's signal leaks away as heat each cycle, like sound dying in a slightly soft wall.
qubit energy ))) ----> drains into: [ TLS defect ] a single atom that flips, in tune with the qubit [ dielectric ] lossy insulator, leaks a little heat each cycle [ surfaces ] thin oxides where most defects live fewer / cleaner materials = longer memory
You cannot just print more
A tempting thought: if one qubit is fragile, just put thousands on a chip and average over the mess. But scaling up runs into walls that have nothing to do with cleverness, and everything to do with physics and manufacturing.
The biggest is frequency crowding (collisions). Each qubit is tuned to its own radio-like pitch so we can address it without disturbing its neighbours. But there is only so much room on the dial. Pack in more qubits and their pitches start to overlap; talking to one accidentally pokes another. On top of that, real fabrication is never perfect — two qubits meant to be identical come out slightly different (parameter variability), so a fraction simply land in the wrong place. The share of chips that come out usable is the yield, and today it drops fast as chips grow.
- Each qubit needs its own frequency slot, like a station on a radio dial.
- Fabrication scatters the actual frequencies, so some land too close together.
- Crowded neighbours interfere, and that whole chip's usefulness drops.
- More qubits make crowding worse, so yield falls faster than the qubit count rises.
So where does that leave us?
None of this means quantum chips are a dead end. It means the hard part is unglamorous: cleaner materials, better wiring, tighter manufacturing. Progress is real but incremental, measured in microseconds of coherence and percentage points of yield, not in sudden leaps.
It also means honesty matters. Today's quantum chips are small and noisy, and no qubit modality has won — superconducting, trapped-ion, spin, and photonic approaches each have their own walls. And to be clear: a quantum chip does not replace your laptop. It is a special-purpose device aimed at a narrow set of problems, and even that promise is still being proven.