search

qiboteam / qibolab

Quantum hardware module and drivers for Qibo.
https://qibo.science
Apache License 2.0
40 stars 10 forks source link

Several Errors - Comparison with 15/12/2023 #860

Open aorgazf opened 3 months ago

aorgazf commented 3 months ago

Dear @hay-k , @andrea-pasquale As explained on Friday, I've run calibration of spin10q q1 using old versions of qibocal and qibolab, dated 15/12/2023: qibocal: alvaro/latest_20231215 qibolab: alvaro/latest_20231215 qibolab_platforms_qrc: alvaro/latest_20231215

And then tried to run the same experiments using the main branches in both repositories. Several of the key characterisation routines in main no longer work: image Attached are all reports and debugging information generated. comparison_with_20231215.zip Thanks for looking into this.

I consider this issue as urgent, because currently, QM instruments are dedicated to controlling qw5q_platinum, and the only control electronics left to characterise the resot of the chips are 1 set of ZI and 3 qblox clusters.

hay-k commented 3 months ago

Hi @aorgazf, thanks for making the effort to report the issues. With @andrea-pasquale we skimmed through the reports with the aim of identifying concrete action points. It is a bit hard to navigate through the data since there is no one-to-one mapping for old vs new comparison, hence it is not always clear what is exactly the thing that is 'no longer working'. Further elaboration would be appreciated. Below are some quick comments from our side:

  1. 20240329_spinq10q_02_punchout_att: Not sure if there is an issue here. Everything looks good.
  2. 20240329_spinq10q_01_resonator_hp_fails: This should already be fixed via https://github.com/qiboteam/qibolab/pull/857
  3. 20240329_spinq10q_06_qubit_50shots_fails: Not sure if we understood what failed here. The acquisition was completed without errors. The data looks like just noise, so I wouldn't expect much from the fit. And I believe the fit behaved the same before as well, so there is no new behavior. We could perhaps consider improving the fit so that it refuses to fit complete noise in general, but this should be a separate discussion.
  4. 20240329_spinq10q_07_rabi_amp_5kshots: Not sure if we understood what is the problem here. For qubit 1 the data is nice and the fit is nice. For the rest - same as above.
  5. 20240329_spinq10q_07_rabi_amp_50shots: Same as above.
  6. 20240329_spinq10q_03_resonator_flux_1000shots_fails: This issue is already separately reported in https://github.com/qiboteam/qibolab/issues/845. Let's follow the progress there.
  7. 20240329_spinq10q_04_resonator_50shots: We can see that for some qubits the fit didn't work. We can explore possibilities to improve the fit in general. However, I have to note that the fitting logic has not changed recently, and in the report generated by older version of qibocal (where all fits worked) the sweep range is narrower. This makes the loretzian fit easier. So the comparison is not entirely apples to apples. Using narrow sweep ranges should mitigate the issue most of the time, however if you have any suggestion on how to evolve the fit beyond basic lorentzian, please feel free to open a separate feature request.
  8. 20240329_spinq10q_09_t2_sequences_takes_60xlonger: This one errored out, so I am not sure compared to what it took 60x longer. Nevertheless, the error itself is already separately reported in https://github.com/qiboteam/qibolab/issues/836. Let's follow the progress there.
  9. 20240329_spinq10q_03_resonator_flux_50shots: Fitting issue. Although, looking at the report from the older version of software, I can see the fit was not correct there either. So, not exactly a new behavior, but anyways the fit should certainly be improved.
  10. 20240329_spinq10q_05_qubit_flux_50shots: Fitting issue, however the data is also extremely noisy, plus the fitting was even worse in the older version. So again, no broken functionality compared to the previous version, but the fit could perhaps be improved.
  11. 20240329_spinq10q_05_qubit_flux_1000shots_fails: This is already separately reported in https://github.com/qiboteam/qibolab/issues/845. Let's follow the progress there.
  12. 20240329_spinq10q_04_resonator_1000shots_fails: Should already be fixed via https://github.com/qiboteam/qibolab/pull/857
  13. 20240329_spinq10q_06_qubit_5kshots_fails: Should already be fixed via https://github.com/qiboteam/qibolab/pull/857
  14. 20240329_spinq10q_10_ramsey_signal_fails: Not sure if we understood what is the issue here.

In summary, I think only 2. 6. 8. 11. 12. 13. are issues that can block the characterization team in one way or another, hence need asap attention. The most pressing ones I believe are 2. 12. 13., which all are different occurrences of the same issue, and fortunately it is already fixed. The others have simple workarounds explained in respective issue, so I would not consider them as blocking, but they are anyways in active progress, so I think everything is already happening as optimal as it gets! Let's follow the progress on the dedicated issues, and please elaborate more on points that you think we misunderstood/misinterpreted.

andrea-pasquale commented 3 months ago

  • 20240329_spinq10q_03_resonator_flux_50shots: Fitting issue. Although, looking at the report from the older version of software, I can see the fit was not correct there either. So, not exactly a new behavior, but anyways the fit should certainly be improved.

  • 20240329_spinq10q_05_qubit_flux_50shots: Fitting issue, however the data is also extremely noisy, plus the fitting was even worse in the older version. So again, no broken functionality compared to the previous version, but the fit could perhaps be improved.

I'm having a look at these two points in https://github.com/qiboteam/qibocal/issues/782