Qubit error bursts in superconducting quantum processors of Quantum Inspire: quasiparticle pumping and anomalous time dependence

Abstract

We investigate qubit error bursts in 5- and 7-transmon processors of similar design, fabrication and packaging, but with different types of qubit Josephson junctions. Measurements for each are performed in two refrigerators to discern device-specific from refrigerator-dependent characteristics. The duration and rate of bursts are device specific but within the range of prior experiments and consistent with ionizing radiation. We observe two unforeseen signatures specifically in the processor with Dolan junctions. First, increasing the rate of $π$ pulsing in the detection scheme shortens the recovery time to equilibrium, which is explained by a quasiparticle pumping mechanism. The second signature is an anomalous time dependence in the burst rate: a surge happens days or weeks after cooldown, followed by a strong suppression that persists until thermal cycling.

Figures (15)

• Figure 1: Detection and characterization of simultaneous error events. (a) Repeated detection subcircuit described in the main text. Blocks labelled $W$ correspond to idling (waiting) periods. (b,c) Average of kept and rejected events for (b) $\mathrm{S\text{-}5}$ ($t_{\mathrm{cycle}}=100~\upmu\mathrm{s}$, $k=5$) and (c) $\mathrm{S\text{-}7}$ ($t_{\mathrm{cycle}}=10~\upmu\mathrm{s}$, $k=7$). Note the different timescales on horizontal axes. For kept events in $\mathrm{S\text{-}5}$ ($\mathrm{S\text{-}7}$), the baseline is fully recovered in $t_{\mathrm{rec}} \sim 2~\mathrm{ms}$$(\sim250~\upmu\mathrm{s})$. (d) Cumulative sum of kept events for $\mathrm{S\text{-}5}$ ($\mathrm{S\text{-}7}$) for $k=3$ and maximized $k=5$$(7)$. (e) Extracted rate of kept events $\gamma_{\mathrm{kept}}$ as a function of $k$ for both processors.
• Figure 2: Signature of QP pumping in $\mathrm{S\text{-}5}$ and absence thereof in $\mathrm{S\text{-}7}$. (a) Variant of the repeated detection subcircuit: an even number $m$ of $\pi$ pulses is inserted after readout. (b,c) Average of kept events for (b) $\mathrm{S\text{-}5}$ and (c) $\mathrm{S\text{-}7}$ with fixed $m=0$ and varying $t_{\mathrm{cycle}}$. $\mathrm{S\text{-}5}$ shows an increase in $t_{\mathrm{rec}}$ as $t_{\mathrm{cycle}}$ increases from $10$ to $100~\upmu\mathrm{s}$. $\mathrm{S\text{-}7}$ does not show any such dependence. (d,e) Average of kept events for (c) $\mathrm{S\text{-}5}$ and (d) $\mathrm{S\text{-}7}$ with fixed $t_{\mathrm{cycle}}=100$ and $30~\upmu\mathrm{s}$, respectively, but increasing $m$. In $\mathrm{S\text{-}5}$, but not in $\mathrm{S\text{-}7}$, $t_{\mathrm{rec}}$ decreases as $m$ increases from 0 to 6. In combination, these results demonstrate that increasing the net rate of $\pi$ pulses in the detection scheme shortens $t_{\mathrm{rec}}$ in $\mathrm{S\text{-}5}$.
• Figure 3: Schematics of the transmon Josephson junctions of $\mathrm{S\text{-}5}$ fabricated using the Dolan-bridge technique. (a) Top view. (b) Cross section along the long axis. (c) Energy diagram illustrating the superconducting energy gaps $\Delta_{\mathrm{Al}}$ in the thin (bottom) and thick (top) Al layers deposited by double-angle evaporation. The difference between these gaps $\delta \Delta_{\mathrm{S\text{-}5}}$ is smaller than the qubit transition energy $h f_{\mathrm{q}}$ in each transmon. [The much larger gap of the NbTiN transmon capacitor pads $(\Delta_{\mathrm{NbTiN}}\sim2.3~\mathrm{meV})$ is not shown.] With the qubit in $\ket{1}$, a QP on the thick left layer can tunnel across the dominant JJ into the thin right layer by relaxing the qubit to $\ket{0}$, then diffuse and tunnel onto the thick right layer, where it becomes trapped. A subsequent $\pi$ pulse re-excites the qubit, commencing another cycle of QP pumping to the trap.
• Figure 4: Example observation of the surge in $\mathrm{S\text{-}5}$ while in refrigerator A. (a) Cumulative sum of kept events (left axis) and average $T_\mathrm{1}$ (right) as a function of elapsed time at base temperature (bottom) and elapsed time running the detection circuit of Fig. \ref{['fig:Figure1']}(a) ($t_{\mathrm{cycle}}=30~\upmu\mathrm{s}$, $k=3$). After 26 days at base temperature with similar characteristics as Fig. \ref{['fig:Figure1']}, $\gamma_{\mathrm{kept}}$ suddenly surges by a factor $\sim\times10$ and average $T_\mathrm{1}$ drops by $\sim\times3$. Over the next several days, $\gamma_{\mathrm{kept}}$ monotonically decreases, ultimately settling to a value $\sim \times 100$ lower than at the start of the experiment, and average $T_\mathrm{1}$ gradually recovers. (b) Average of kept events ($t_{\mathrm{cycle}}=30~\upmu\mathrm{s}$, $k=5$) over the full observation period. We use a second pass of template matching (see SOM_CR) to discern $37$ bursts with atypically long recovery times (red), all of which are detected at the onset of the surge. See SOM_CR for other observations of the surge showing some refrigerator-specific characteristics. See SOM_CR for other observations of the surge in $\mathrm{S\text{-}5}$.
• Figure S1: (a,b) Optical images of the (a) $\mathrm{S\text{-}5}$ and (b) $\mathrm{S\text{-}7}$ quantum processors. Both have 7 transmons and the same nearest-neighbor connectivity, with only evolutionary differences in design and layout (e.g., signal routing). Two transmons of $\mathrm{S\text{-}5}$ are inoperable (indicated by red crosses). A key addition in $\mathrm{S\text{-}7}$ is the addition of 'shoelacing' airbridges allowing post-fabrication frequency trimming of readout and Purcell resonators Valles23. Both processors are laterally connected to a Cu PCB using Al wirebonds. These images were taken before the start of the experiment, when the initial density of wirebonds in $\mathrm{S\text{-}5}$ was roughly double that of $\mathrm{S\text{-}7}$. (c,d) Scanning electron microscope images of (c) Dolan and (d) Manhattan JJs in sister devices of $\mathrm{S\text{-}5}$ and $\mathrm{S\text{-}7}$, respectively, with added falsecolor: gray for the bottom (thin) electrode, light blue for the top (thick) electrode, and brown for the NbTiN arms of the superconducting quantum interference device (SQUID) loops. Note that we do not use any bandaging layers between the Al electrodes and the NbTiN arms.