The Wiring Bottleneck and Cryo-CMOS

A few cables per qubit, and it adds up fast

Picture the chip at the bottom of a dilution refrigerator, colder than deep space. To run it, room-temperature electronics upstairs must send microwave pulses *down* to drive and read each qubit. Those signals travel along coaxial cables — stiff, finger-thick, gold-plated lines that snake through every cold stage of the fridge.

The trouble is the count. A typical superconducting qubit needs several lines all to itself: one to drive single-qubit gates, often one to tune its frequency, and at least one for readout. Multiply by the number of qubits and the cables pile up. A 50-qubit machine already runs on a forest of well over a hundred coax lines; scale that to the millions of qubits a useful error-corrected machine would need, and you simply cannot fit the cables — let alone cool them. This is the [[wiring-bottleneck|wiring bottleneck]].

  roughly how the cable count grows

   qubits   lines/qubit   total coax lines (order of)
   ------   -----------   --------------------------
      1         ~3                 ~3
     50         ~3                ~150
    100         ~3                ~300
  1,000,000     ~3            ~3,000,000  (cannot fit)

  'lines/qubit ~3' = one drive + one flux/tune + one readout
  the fridge has finite space and finite cooling power;
  millions of independent coax lines fit in neither.

Heat, and the bulky parts that make it worse

A cable is not just a path for signal — it is also a path for *heat*. Each coax line is a thermal bridge connecting the warm top of the fridge to the cold chip, and every line you add leaks a little more heat into the coldest stage. That stage has a cooling power measured in fractions of a milliwatt; spend it all warming up cables and the fridge can no longer hold the chip cold enough to work. The same cables that carry your signals are quietly fighting your refrigerator.

Readout makes it worse, because reading a qubit usually needs a circulator: a small component that lets a signal pass one way but not back, so the faint echo from the chip reaches the amplifier while the amplifier's own noise is kept from washing back onto the qubit. Circulators work beautifully — but each one is a heavy puck the size of a coin, full of magnetic material, and you need roughly one per readout line. Stack a few hundred of those at the cold plate and you have a brass jungle that is bulky, heavy, and impossible to scale.

  signal path down a single readout line today

   300 K  room-temp electronics
   (warm)      |  coax (carries signal AND heat down)
              v
   ~4 K     attenuators / amplifier stages
              |  coax
              v
   ~10 mK   [ CIRCULATOR ]  <- coin-sized, magnetic, heavy
   (cold)        |             one per readout line
              QUBIT CHIP

  add up: many coax lines  +  one bulky circulator each
          = space runs out, cooling budget runs out

The leading escape: multiplexing and cryo-CMOS

If you cannot afford one cable per qubit, the obvious move is to make one cable serve many qubits. That is [[multiplexed-readout|multiplexed readout]]: you give several readout resonators slightly different frequencies and connect them all to a single line, then read them at once by sending a comb of tones and sorting the echoes apart by pitch. Ten qubits, one cable. This already works and is widely used — it is the cheapest, most mature relief for the bottleneck.

The bigger idea is to stop sending control signals all the way from room temperature at all. [[cryo-cmos|Cryo-CMOS]] means building ordinary silicon control chips — the same transistor technology in your phone, but designed to run at a few kelvin — and placing them *inside the fridge*, close to the qubits. Then the long warm-to-cold cables carry only a few digital instructions and a clock, and the actual microwave pulses are generated locally, down in the cold. Far fewer cables cross the boundary; the bottleneck eases dramatically.

  before  (one cable per qubit, signals made warm)

   300 K  [ generate every pulse here ]
            | | | | | | | | |   many long coax lines
            v v v v v v v v v
   10 mK  Q Q Q Q Q Q Q Q Q   chip

  after  (cryo-CMOS + multiplexing)

   300 K  [ a few digital lines + clock ]
            |   (thin, low-traffic)
            v
   ~4 K   [ CRYO-CMOS controller ]  <- makes pulses locally
            | shared / multiplexed lines
            v
   10 mK  Q Q Q Q Q Q Q Q Q   chip

  fewer wires cross the warm->cold boundary

Why cryo-CMOS is promising but still early

Here is the honest catch. A silicon control chip, even an efficient one, *dissipates power* — transistors switching is transistors making heat. Put that chip in the cold and its waste heat lands squarely on the same scarce cooling budget the qubits depend on. You have traded a heat-leaking cable problem for a heat-dumping transistor problem. The whole design game becomes: deliver clean control pulses while spending as little power, and therefore as little of the cooling budget, as possible.

This is why the controllers usually sit at a warmer, roomier stage of the fridge (around 4 kelvin), where there is far more cooling power to spare, rather than right on the coldest plate beside the qubits. It buys headroom for the heat, at the cost of still needing some wiring down to the chip. The field is making real progress — working cryo-CMOS controllers exist in the lab — but matching the signal purity of room-temperature electronics while staying inside a tight power budget is not yet solved at scale.

1. Estimate the cooling power available at the stage where the controller will sit (much larger at 4 K than at 10 mK).
2. Budget how much power each control channel may dissipate, then multiply by the number of qubits — the total must stay under that cooling power.
3. Design the cryo-CMOS circuit to hit that per-channel power while still producing pulses clean enough not to add qubit errors.
4. If the heat still overruns the budget, multiplex more aggressively or move the controller to a warmer stage — and accept the extra wiring that warmer stage implies.