Resolving catastrophic error bursts from cosmic rays in large arrays of superconducting qubits

Classified as probably not a substantial claim.

Using a matched filter, 415 events were extracted from the datasets.

“Using our matched lter, we extract 415 events from these datasets”

May combine two claims.

Using a typical qubit frequency of 2.6 GHz, the quasiparticle density at the peak of the event is ~2.1e-5, corresponding to an absolute density of 87 m^-3; producing this density evenly over the full aluminum qubit layer on the 10 mm x 10 mm qubit chip would require an absorbed energy in the aluminum of ~160 keV, which corresponds well to the expected energy depositions from large events.

“Using a typical qubit frequency ! q = 2 6 GHz, we calculate a quasiparticle density of x qp 2 : 1 10 5 at the peak of the event, once the error has spread over a large area. We can convert this to an absolute density of n qp = 87 m 3 . Producing this density evenly over the full aluminum qubit layer on the 10 mm x 10 mm qubit chip would require an absorbed energy in the aluminum of E = n qp V 160”

• AssumptionRests on an assumption: The quasiparticle density is distributed evenly over the full aluminum qubit layer when estimating the energy required to produce the observed T1 time.

May combine two claims.

In indium, for an energy absorbed of 50 keV, the quasiparticle density is ~1e-6, the recombination time is ~36 μs, and accounting for the ratio of phonon absorption length (200 nm) to bumpbond height (5 μm), the effective relaxation time of the indium bumps is ~0.9 ms, similar to the spread of energy from recombination in the groundplane, so recombination in indium is compatible with the observed

“Taking the same hotspot area of 10 mm 2 , and considering an energy absorbed in the indium to be E In = 50 keV, the recombination timescales are similar. In indium, n cp 13 10 6 m 3 , so we nd a small quasiparticle density x qp 1 10 6 due to the large volume. The short characteristic phonon timescale ph 0 = 0 : 799 ns results in a recombination time r 36 s. However, only recombinations taking”

• AssumptionRests on an assumption: The speed of sound in indium is 1.2 km/s.

May combine two claims.

The decay timescale of extracted events is tightly grouped in the 25 to 30 ms range, and peak heights range from the minimum identifiable up to the full number of qubits used.

“We nd that the decay timescale is tightly grouped in the 25 to 30 ms range, and that peak heights range from the minimum identi able by our analysis up to the full number of qubit used.”

May combine two claims.

For a median impact delivering 100 keV of energy to the substrate, with 20% of the energy remaining in free charges, the energy absorbed by the initial hotspot in the aluminum layer is around 80 keV; taking the hotspot area as 10 mm2 and aluminum thickness 100 nm, the quasiparticle density is ~1.1e-4 and the recombination time is ~180 μs, which agrees well with the experimentally observed time for

“For a median impact delivering 100 keV of energy to the substrate [2], where 20% of the energy remains in free charges, the energy absorbed by the initial hotspot in the aluminum layer will be around E hs = 80 keV. We take area of the hotspot to be A = 10 mm 2 , similar to that seen in segment 1 of Fig. 4 of the main paper. The aluminum thickness d = 100 nm. For aluminum, 0 : 18 meV and n cp 4 10”

• cites [2] Terhal, B. M. Quantum error correction for quantum memories. Rev. Mod. Phys. 87, 307–346 (2015). sourceNot checked yet

  Not checked in the free run.

• cites [8] Martinis, J. M., Ansmann, M. & Aumentado, J. Energy decay in superconducting josephson-junction qubits from nonequilibrium quasiparticle excitations. Phys. Rev. Lett. 103, 097002 (2009). sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

At the start of the process, the timescale is estimated to be 180 us.

“At the start of this process, we estimate a timescale of 180 s.”

May combine two claims.

The event saturates the qubit patch, with all qubits experiencing a high probability of reporting an error, indicating total failure of coherence on the chip.

“This event has e ectively saturated the qubit patch, with all qubits experiencing a high probability of reporting an error, indicating total failure of the coherence on the chip.”

May combine two claims.

Heatmaps show that high error rates are initially localized to a small number of qubits but spread through the device over the course of the event.

“High error rates are initially localized to a small number of qubits, but spread through the device over the course of the event.”

May combine two claims.

For our carrier chip of 20 mm x 26 mm, the equivalent event rates are 1/(7.6 s) for gamma rays and 1/(38 s) for cosmic rays.

“For our carrier chip of 20 mm x 26 mm, the equivalent event rates are 1 = (7 : 6 s) and 1 = (38 s) respectively.”

May combine two claims.

The peak-finding technique is capable of detecting peaks of heights greater than 2 errors over the background error value around 4 errors, with the smallest identified peaks having absolute heights of 6 errors.

“This technique is capable of detecting peaks of heights greater than 2 errors over the background error value around 4 errors, with the smallest identi ed peaks having absolute heights of 6 errors.”

May combine two claims.

Initializing j0i produces much lower background error rates, as T1 error is suppressed.

“We note that initializing j 0 i produces much lower background error rates, as T 1 error is suppressed.”

May combine two claims.

In higher-time-resolution measurements, an event shows three distinct timescales: an immediate jump in error from baseline at 4 to 10 errors in only 10 us, a slower saturation up to a maximum of 15 over the following 1 ms, and a typical 25 ms exponential decay back towards baseline.

“We nd three distinct timescales; an immediate jump in error from baseline at 4 errors to 10 errors in only 10 s, a slower saturation up to a maximum of 15 over the following 1 ms, and a typical 25 ms exponential decay back towards baseline.”

May combine two claims.

Each of the three events shows a different location for the initial area of high error rates: Event A on the right side, Event B on the bottom left, and Event C on the top left of the qubit patch.

“However, each event shows a di erent location for the initial area of high error rates; Event A features an initial impact centered on the right side of the qubit patch, Event B features an initial impact on the bottom left side of the qubit patch, and Event C sees an initial impact on the top left of the patch.”

May combine two claims.

While there is localization on short timescales, events generically affect all qubits enough to correspond well to a simple homogeneous model.

“While there is localization on short timescales, the events generically a ect all qubits enough to correspond well to a simple homogeneous model.”

• AssumptionThe authors state this as an assumption: The global T1 model assumes a homogeneous T1 across all qubits equal to the average T1 extracted from relative peak heights.

May combine two claims.

The time periods between events in the same dataset correspond to an exponential distribution with an average event rate of 1/(10 s).

“We also extracted 326 time periods between events occurring in the same dataset and nd a strong correspondence to an exponential distribution with an average event rate = 1 = (10 s).”

May combine two claims.

Smaller events are found more frequently, qualitatively matching the distribution of deposited energies for impinging particles.

“Smaller events are found more frequently, qualitatively matching the distribution of deposited energies for impinging particles [2].”

• cites [2] Terhal, B. M. Quantum error correction for quantum memories. Rev. Mod. Phys. 87, 307–346 (2015). sourceNot checked yet

  Not checked in the free run.

May combine two claims.

All three immediate impacts primarily affect a small patch of qubits, producing a jump from 4 simultaneous errors at baseline to 10 errors.

“All three immediate impacts primarily a ect a small patch of qubits, producing a jump from 4 simultaneous errors at baseline to 10 errors.”

May combine two claims.

All three highly time-resolved events display the same overall timescales: an initial jump in a few tens of μs, a saturation to peak value in around 1 ms, and an exponential decay back to baseline with a timescale of 25 ms.

“All three events display the same overall timescales, showing an initial jump in a few tens of s, a saturation to peak value in around 1 ms and an exponential decay back to baseline with a timescale of 25 ms.”

May combine two claims.

In Fig. 4 of the main paper, the authors extracted an effectively global T1 time around 1 μs at the peak of large events, which is compatible with estimates based on similar device structures.

“In Fig. 4 of the main paper we extracted an e ectively global T 1 time around 1 s at the peak of large events, which is compatible with estimates based on similar device structures [1].”

• cites [1] Fowler, A. G., Mariantoni, M., Martinis, J. M. & Cleland, A. N. Surface codes: towards practical large-scale quantum computation. Phys. Rev. A 86, 032324 (2012). sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

The largest detected events correspond to a chip-wide reduction in T1 to less than 1 us.

“The largest detected events correspond to a chip-wide reduction in T 1 to less than 1 s.”

• AssumptionThe authors state this as an assumption: The global T1 model assumes a homogeneous T1 across all qubits equal to the average T1 extracted from relative peak heights.

May combine two claims.

The event shape is a sharp jump and exponential decay that is characteristic of all events.

“Fig. S2a shows time slice of raw data including an event, illustrating the sharp jump and exponential decay that is characteristic of all events.”

May combine two claims.

If the TiN film has a higher gap and no subgap conduction, the recombination time for quasiparticles trapped in indium can be estimated similarly, and the effective relaxation time of the indium bumps is ~0.9 ms, compatible with the observed spread.

“we calculate the e ective relaxation time of the indium bumps as e r = r z=l 0 : 9 ms. This timescale is similar to the spread of energy from recombination in the groundplane, so recombination in indium is also compatible with the observed spread of the hotspot.”

• AssumptionRests on an assumption: The speed of sound in indium is 1.2 km/s.

May combine two claims.

Energetic impacts from cosmic rays and latent radioactivity violate the assumptions that physical errors remain small and sufficiently uncorrelated as devices scale.

“However, energetic impacts from cosmic rays and latent radioactivity violate both of these assumptions.”

Probably not a substantial claim: may not pass S4 (based).

Physical errors can remain both small and sufficiently uncorrelated as devices scale, so that logical error rates can be exponentially suppressed.

“The key premise is that physical errors can remain both small and su ciently uncorrelated as devices scale, so that logical error rates can be exponentially suppressed.”

• cites [1] Fowler, A. G., Mariantoni, M., Martinis, J. M. & Cleland, A. N. Surface codes: towards practical large-scale quantum computation. Phys. Rev. A 86, 032324 (2012). sourceNot checked yet

  Not checked in the free run.

• cites [2] Terhal, B. M. Quantum error correction for quantum memories. Rev. Mod. Phys. 87, 307–346 (2015). sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The event shape is reliable, allowing the use of a matched filter to optimally reduce noise and locate peaks.

“Because the event shape is reliable, we can use a matched lter to optimally reduce noise and locate peaks.”

May combine two claims.

As the event continues, energy is distributed over a larger volume, slowing the rate of recombination, explaining the qualitative shape of the rising behaviour.

“As the event continues, the energy will be distributed over a larger volume, slowing the rate of recombination, and explaining the qualitative shape of the rising behaviour seen here.”

May combine two claims.

Quasiparticles in the superconductor rapidly cool to energies near the superconducting gap Δ; when tunneling across the Josephson junction, they can absorb the qubit energy E10, producing a decay error.

“Quasiparticles in the superconductor rapidly cool to energies near the superconducting gap . When tunneling across the Josephson junction, these quasiparticles are able to absorb the qubit energy E 10 , producing a decay error.”

May combine two claims.

The majority of electron-hole pairs immediately recombine and emit photons, but 20% avoid immediate recombination and propagate in the substrate.

“The majority of electron-hole pairs immediately recombine and emit photons but 20% of them avoid immediate recombination and propagate in the substrate [2].”

• cites [2] Terhal, B. M. Quantum error correction for quantum memories. Rev. Mod. Phys. 87, 307–346 (2015). sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The baseline performance of the experiment is well understood and the peaks represent anomalous correlated error events.

“This indicates that the baseline performance of the experiment is well understood and that the peaks represent anomalous correlated error events.”

May combine two claims.

The normalized distribution of times between events shows a strong correspondence to the probability density function for a Poisson process with rate = 1/10 s, indicating that events occur independently, as expected for events produced by radioactive decays or isolated cosmic ray impact events.

“Fig. S2e shows the normalised distribution of these times, along with the probability density function for a Poisson process with = 1 = 10s. The strong correspondence indicates that events occur independantly, as one might expect for events produced by radioactive decays or isolated cosmic ray impact events.”

May combine two claims.

The initial localization is caused by a single point of impact and in a location that is independent between different events.

“This provides further evidence that the initial localization is caused by a single point of impact and in a location that is independent between di erent events.”

May combine two claims.

Events are initially localized but spread over the chip, providing strong evidence for a high-energy impact.

“Using ne time-resolved measurements, we show that events are initially localized but spread over the chip, providing strong evidence for a high-energy impact.”

May combine two claims.

While the incidence rate can be reduced by shielding, the ability of even a single event to cause widespread failure calls for mitigation efforts on the chip itself.

“While the incidence rate can be reduced by shielding [3, 4], the ability of even a single event to cause widespread failure calls for mitigation e orts on the chip itself [14].”

• cites [3] Cardani, L. et al. Reducing the impact of radioactivity on quantum circuits in a deep-underground facility. Nat. Commun. 12, 2733 (2021). sourceNot checked yet

  Not checked in the free run.

• cites [4] Vepsäläinen, A. P. et al. Impact of ionizing radiation on superconducting qubit coherence. Nature 584, 551–556 (2020). sourceNot checked yet

  Not checked in the free run.

• cites [14] Karatsu, K. et al. Mitigation of cosmic ray effect on microwave kinetic inductance detector arrays. Appl. Phys. Lett. 114, 032601 (2019). sourceNot checked yet

  Not checked in the free run.

May combine two claims.

For a smaller sample with area 40 mm2, gamma ray and cosmic ray events were reported with rates of 1/(100 s) and 1/(500 s), and deposited energies around 100 keV and 1 MeV respectively.

“For a smaller sample with area 40 mm 2 gamma ray and cosmic ray events where reported with a rate of 1 = (100 s) and 1 = (500 s), and deposited energies around 100 keV and 1 MeV respectively [2].”

• cites [2] Terhal, B. M. Quantum error correction for quantum memories. Rev. Mod. Phys. 87, 307–346 (2015). sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The selection of decay = 20 ms influences the symmetry and scale of the resulting peak, but not the location of the maximum.

“Our selection of decay = 20 ms will in uence the symmetry and scale of the resulting peak, but not the location of the maximum.”

May combine two claims.

The occurrence of events is independent over time, occurring on average every 10 s without significant bunching or anti-bunching.

“This indicates that the occurrence of the events is independent over time, occurring on average every 10 s without signi cant bunching or anti-bunching.”

May combine two claims.

The event timescale is short compared to the projected run time of error-corrected algorithms of several hours, so any attempt to preserve a logical state for computation is very likely to see such an event.

“However, this timescale is quite short compared to the run time of error corrected algorithms, which is projected to be several hours [25], so any attempt to preserve a logical state for computation is very likely to see such an event.”

• cites [25] Google Quantum AI. Exponential suppression of bit or phase errors with cyclic error correction. Nature 595, 383–387 (2021).Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The event rate found is sensitive to the lower cutoff for size of events located, as less sensitive methods will locate fewer events.

“We also note that the event rate found is sensitive to the lower cuto for size of events located, as less sensitive methods will locate fewer events.”

May combine two claims.

High-energy radiation has been identified as a source of quasiparticles in transmon qubits, and single events have been reported in small numbers of specialized devices such as resonators and charge-sensitive qubits.

“High-energy radiation has been identi ed as a source of quasiparticles in transmon qubits [4], and single events have been reported in small numbers of specialized devices such as resonators [3] as well as qubits that are charge-sensitive [5].”

• cites [4] Vepsäläinen, A. P. et al. Impact of ionizing radiation on superconducting qubit coherence. Nature 584, 551–556 (2020). sourceNot checked yet

  Not checked in the free run.

• cites [3] Cardani, L. et al. Reducing the impact of radioactivity on quantum circuits in a deep-underground facility. Nat. Commun. 12, 2733 (2021). sourceNot checked yet

  Not checked in the free run.

• cites [5] Wilen, C. D. et al. Correlated charge noise and relaxation errors in superconducting qubits. Nature 594, 369–373 (2021). sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The findings are relevant for other solid-state quantum systems, such as spin qubits sensitive to induced charges and Majorana fermion devices sensitive to induced quasiparticle population.

“Our ndings are also relevant for other solid-state quantum systems, such as spin qubits that would be sensitive to the induced charges [26], and Majorana fermion devices that would be sensitive to induced quasiparticle population [27, 28].”

• cites [26] Houzet, M., Serniak, K., Catelani, G., Devoret, M. H. & Glazman, L. I. Photon-assisted charge-parity jumps in a superconducting qubit. Phys. Rev. Lett. 123, 107704 (2019). sourceNot checked yet

  Not checked in the free run.

• cites [27] Gidney, C. & Ekerå, M. How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits. Quantum 5, 433 (2021). sourceNot checked yet

  Not checked in the free run.

• cites [28] Zwanenburg, F. A. et al. Silicon quantum electronics. Rev. Mod. Phys. 85, 961–1019 (2013). sourceNot checked yet

  Not checked in the free run.

May combine two claims. The quote only partly states this claim.

The absence of correlated excitation-error peaks indicates that events are produced by an asymmetric decay error mechanism, compatible with quasiparticle poisoning across the chip.

“indicating that events are produced by an asymmetric decay error mechanism, which is compatible with quasiparticle poisoning across the chip.”

May combine two claims.

The authors have shown direct measurements of widespread, long-lived and highly detrimental correlated events in a practical quantum processor, presenting an existential challenge to quantum error correction.

“We have shown direct measurements of widespread, long-lived and highly detrimental correlated events in a practical quantum processor, presenting an existential challenge to quantum error correction.”

May combine two claims.

The superconducting gap of TiN films depends sensitively on thickness and contaminants, and there is evidence for a smeared gap that might allow conduction.

“The superconducting gap of TiN lms depends sensitively on the thickness and presence of contaminants, and there is evidence in literature for the possibility of a smeared gap that might allow conduction [9].”

• cites [9] Lenander, M. et al. Measurement of energy decay in superconducting qubits from nonequilibrium quasiparticles. Phys. Rev. B 84, 024501 (2011). sourceNot checked yet

  Not checked in the free run.

May combine two claims.

In the case that the TiN film has a lower superconducting gap than indium, quasiparticles will rapidly flow from indium into aluminum, cool to the aluminum gap, and recombine in aluminum at around the previously found rate.

“In the case that the TiN lm has a lower superconducting gap than the indium, quasiparticles will rapidly ow from the indium into the aluminum, where they will cool to the aluminum gap. Unable to ow back into the indium, they will recombine in the aluminum at around the rate found previously.”

May combine two claims.

The indium may play a role in the initial absorption of energy, but the presence of aluminum will soon downconvert phonons to below the indium gap, effectively freezing the indium out of the dynamics at longer timescales.

“As such, the indium may play a role in the initial absorption of energy. However, the presence of the aluminum will soon downconvert phonons to below the indium gap, e ectively freezing the indium out of the dynamics at longer timescales.”

• AssumptionRests on an assumption: Indium has a superconducting gap of 0.52 meV, nearly three times higher than aluminum.