Cryogenic Attenuator
Why Drive Lines Need Distributed Cold Attenuation
A superconducting qubit operating near 5 to 6 GHz has a transition energy hf that is only a few tenths of a kelvin in temperature units, so it is exquisitely sensitive to thermal microwave photons. Control electronics that generate the drive and readout tones sit at room temperature, and the coaxial line that carries those tones down into the cryostat would, if left alone, deliver a 300 K Johnson noise floor straight onto the qubit. That noise both saturates the readout resonator and dephases the qubit, destroying coherence. The job of the cryogenic attenuator chain is to attenuate the intended signal by a controlled amount while replacing the upstream noise with the much colder thermal emission of a millikelvin-anchored resistor.
The reason engineers cannot simply place one large attenuator at the bottom is that an attenuator is itself a noise source. A pad with attenuation L (a linear power ratio) passes a fraction 1/L of its input noise and adds its own thermal emission scaled by (1 - 1/L), all referenced to the attenuator's physical temperature. Putting 60 dB at the 4 K plate would still hand the qubit 4 K noise from that single pad, which corresponds to a thermal photon occupancy far above what coherent operation tolerates. Splitting the loss so the final increment sits at the 10 mK mixing chamber drives the effective noise temperature of the line down to the qubit's own bath temperature.
The penalty is heat. Every decibel of attenuation turns absorbed signal power into heat that the cold stage must remove, and the mixing chamber typically has only tens of microwatts of cooling power. Designers therefore keep drive powers at the coldest stage in the picowatt range, choose attenuator bodies and connector interfaces with strong thermal links, and verify that worst-case continuous-wave drive never exceeds the cooling budget of the plate the attenuator is anchored to.
Noise Temperature and Photon Occupancy
Tout = Tin/L + Tp × (1 − 1/L) K
Mean thermal photon occupancy (Bose-Einstein):
n̄ = 1 / (exp(hf / kBT) − 1) ≈ exp(−hf / kBT) for hf >> kBT
Dissipated heat at a stage:
Pdiss = Pin × (1 − 1/L)
Where L = linear power ratio (20 dB → L = 100), Tp = attenuator physical temperature, h = Planck constant, kB = Boltzmann constant, f = signal frequency. Example: f = 6 GHz, T = 10 mK → hf / kBT ≈ 28.8 → n̄ ≈ 3 × 10−13.
Typical Drive-Line Attenuation Budget
| Stage | Physical Temp | Typical Attenuation | Cooling Power | Role on the Line |
|---|---|---|---|---|
| 50 K shield | ~50 K | 0 to 3 dB | ~1 W | Thermal break, optional pad |
| 4 K plate | ~4 K | 20 dB | ~1 W | Dumps most 300 K noise |
| Still | ~0.8 K | 10 dB | ~1 mW | Intermediate re-anchoring |
| Cold plate | ~0.1 K | 10 dB | ~100 μW | Pre-final noise reduction |
| Mixing chamber | ~10 mK | 20 dB | ~10 to 20 μW | Sets qubit noise floor |
Frequently Asked Questions
How much attenuation is distributed on a qubit drive line and why is it split across stages?
A typical XY drive line carries about 60 dB total, often 20 dB at the 4 K plate, 10 dB at the still (~0.8 K), 10 dB at the cold plate (~0.1 K), and 20 dB at the mixing chamber (~10 mK). It is split because each pad re-emits Johnson noise at its own physical temperature; a single 60 dB room-temperature pad would still pass 300 K noise down the line below it. Distributing the loss lets each warmer stage dump most upstream noise where cooling power is plentiful, while the final cold pad references the noise to ~10 mK.
Why can't you just use a room-temperature attenuator and a long cable instead?
A resistive attenuator both reduces the signal and re-emits noise at its own physical temperature. A 300 K pad attenuates the tone 100-fold yet injects 300 K Johnson noise into its output, which then reaches the qubit unattenuated by the pad that created it. Only a pad that is physically cold and thermally anchored to a millikelvin stage emits the low noise floor a qubit needs. A long cable adds poorly controlled, temperature-dependent loss with no defined heat sink, so discrete attenuators bolted to each plate are preferred.
What thermal photon occupancy does a mixing-chamber attenuator achieve at 6 GHz?
The mean photon number is n̄ = 1 / (exp(hf / kBT) − 1). At 6 GHz and 10 mK, hf / kBT ≈ 28.8, giving n̄ ≈ 3 × 10−13. Real lines rarely reach a true 10 mK noise temperature; an effective 40 to 50 mK pushes n̄ into the 10−3 range, which is the practical target for keeping dephasing and readout-resonator population low. The final 20 dB pad is sized so warmer-stage residuals sit below this level.
How is heat dissipation in a cryogenic attenuator managed against the cooling budget?
A 20 dB pad absorbs 99 percent of incident power, so signal levels at the coldest stage are held to picowatts because the mixing chamber offers only tens of microwatts of cooling. Designers pick attenuators with strong thermal links, beryllium-copper or stainless bodies with gold-plated mating interfaces, and confirm that worst-case continuous-wave drive times the absorbed fraction stays well under the cooling power of the anchoring stage.