Germanium qubit team traces where spin readout loses accuracy
A Nature Electronics team dissected the steps used to read two hole-spin qubits in germanium. By tuning the magnetic field and holding a short-lived charge signal long enough to measure, it reported 97 percent average preparation-and-measurement fidelity in one device. The result pinpoints error sources that future designs can target; it does not demonstrate a large, manufacturable quantum processor.
Science··Night
A charge signal stands in for spin
The experiment used two single-hole quantum dots at one end of a germanium device. A nearby radiofrequency sensor detected changes in charge, which are easier to register than the spin state directly. To reach that sensor, the team used Pauli spin blockade: depending on how the two spins were arranged, movement between the dots was either allowed or blocked. The resulting charge configuration became the observable readout signal. This conversion is a useful route to measuring a qubit, but its intermediate steps can introduce mistakes before the sensor records anything.[1]
Field choice exposes a short-lived state
The researchers varied gate-voltage ramp timing and magnetic-field direction to isolate errors in the spin-to-charge conversion. They found that a stronger field shortened the lifetime of the blocked spin state at the readout point. Operating at a lower field helped that state survive longer. The team then used a double latch to move the temporary, spin-dependent charge signal into a state that remained measurable for longer. This separated a loss during conversion from a later loss during latching or from noise in the charge sensor.[1]
One device reaches 97 percent
With those settings, average fidelity for preparation plus readout reached 97.0 percent in the two-qubit experiment. Rabi-oscillation visibility and the measured-state confusion matrix supplied that estimate; it was not a count of error-free operations in a large processor. An earlier germanium hole-spin readout figure near 94 percent served as the comparison. The present work identifies which steps limit this device and why changing them helps. It did not test manufacturing batches or a large array of coupled qubits, where further constraints could alter readout performance.[1]
Related columns
For more information on this topic, you can read the related columns.
