At recent quantum technical conferences it has been emphasized that qubit quality is a very important factor in creating a viable quantum computer.  So kudos to IBM, Rigetti, IonQ and Google for documenting the specifications of their respective chips. However, the companies are using different formats making it difficult to compare so we decided to put them into a common format.  These parameters are shown below in the three tables below.   Table 1 shows the qubit count, qubit connectivity, T1 and T2 or T2* times.   Table 2 shows the single qubit gate fidelity, the two qubit gate fidelity and the readout fidelities.   And Table 3 shows the sources of the data for the first two tables.  Note that all of this data is for physical qubits without any error correction.

Google’s T1 times are roughly 2-4X worse than IBM’s. However, they have faster gate times which compensates for the lower T1 numbers because it is the ratio of T1 to gate times which is the important factor. Also note that IonQ’s ion trap gate delays are about three orders of magnitude slower than those of the superconducting implementations.  So this takes away a little of the advantage they may have in T1 and T2 decoherence times.

We encourage anyone else who has data on other implementations to contact us at info@quantumcomputingreport.com so we can include your data into these tables.  Please see the notes at the end of this page that provide additional details of how these tables were generated.

Table 1

    Qubit Connectivity T1 (µsec) T2 (µsec)
Computer Qubit Count Min Max Ave Min Max Ave Min Max Ave
IBM Q5 Tenerife
5
2
4
2.4
34.3
49.7
43.6
4.8
53.3
25.1
IBM Q5 Yorktown
5
2
4
2.4
36.04
73.66
58.83
25.16
73.80
54.65
IBM Vigo
5
1
3
1.6
71.09
124.55
100.45
18.39
127.74
80.77
IBM Oursense
5
1
3
1.6
69.04
117.51
92.66
30.14
126.47
60.50
IBM Q16 Melbourne
14
1
3
2.57
25.5
88.5
52.2
15.6
105.0
64.8
IBM Q20 Poughkeepsie
20
2
3
2.3
39.4
123.3
73.2
10.8
123.6
66.2
IBM Q20 Tokyo
20
2
6
3.9
42.2
148.5
84.3
24.3
78.4
49.6
IBM Q20 Austin
20
2
6
3.9
N/A
N/A
102.6
N/A
N/A
111.26
IBM Q System One
20
2
6
3.9
38.2
132.9
73.9
39.2
100.8
69.1
IBM Almaden Boeblingen  Singapore
20
1
3
2.3
TBD
TBD
TBD
TBD
TBD
TBD
IBM Rochester
53
1
3
2.15
TBD
TBD
TBD
TBD
TBD
TBD
Rigetti 16Q Aspen-4
16
2
3
2.25
N/A
N/A
25.24
N/A
N/A
19.89
Rigetti Aspen-7
28
2
3
2.375
N/A
N/A
41
N/A
N/A
35
Rigetti Aspen-8
31
2
3
2.375
N/A
N/A
29
N/A
N/A
18
IonQ 11 Qubit
11
11
11
11
>1010
>1010
>1010
TBD
TBD
~3x106
Honeywell 4 Qubit
4
4
4
4
TBD
TBD
TBD
TBD
TBD
TBD
Google Sycamore
53
1
4
3.25
9.7
27.8
16.04
N/A
N/A
N/A
QuTech Spin-2
2
1
1
1
N/A
N/A
>20,000
N/A
N/A
>6
QuTech Starmon-5
5
1
4
2.4
N/A
N/A
18
N/A
N/A
23

Table 2

  1-Qubit Gate Fidelity 2-Qubit Gate Fidelity Readout Fidelity
Computer Min Max Ave Min Max Ave Min Max Ave
IBM Q5 Tenerife
99.83%
99.93%
99.89%
93.78%
97.74%
95.54%
64.9%
95.7%
84.7%
IBM Q5 Yorktown
99.87%
99.92%
99.90%
98.46%
98.89%
98.66%
97.45%
98.70%
98.31%
IBM Vigo
99.86%
99.92%
99.89%
98.81%
99.29%
99.08%
93.60%
98.50%
97.22%
IBM Oursense
99.90%
99.95%
99.93%
99.04%
99.27%
99.20%
96.30%
98.90%
97.44%
IBM Q16 Melbourne
96.26%
99.80%
99.20%
84.52%
97.11%
92.99%
89.30%
96.59%
94.64%
IBM Q20 Poughkeepsie
99.72%
99.95%
99.89%
93.39%
98.89%
97.75%
TBD
TBD
TBD
IBM Q20 Tokyo
99.39%
99.94%
99.80%
92.88%
98.53%
97.16%
N/A
N/A
91.72%
IBM Q20 Austin
N/A
N/A
N/A
N/A
N/A
98.47%
N/A
N/A
91.55%
IBM Q System One
99.92%
99.98%
99.96%
97.15%
99.03%
98.31%
TBD
TBD
TBD
Rigetti 16Q Aspen-4
N/A
N/A
95.5%
N/A
N/A
90.35%
N/A
N/A
93.02%
Rigetti Aspen-7
N/A
N/A
99.23%
N/A
N/A
95.2%
N/A
N/A
96.4%8
Rigetti Aspen-8
N/A
N/A
99.79%
N/A
N/A
95.66%
N/A
N/A
N/A
IonQ 11 Qubit
 99.18%
99.64%
 99.46%
 95.1%
98.9%
 97.5%
N/A
N/A
99.3%
Honeywell 4 Qubit Simultaneous
 99.986%
99.991%
 99.989%
 99.12%
99.28%
 99.20%
99.7%9
99.8%9
99.7%9
Google Sycamore Simultaneous  
99.66%
99.92%
 99.84%
N/A
N/A
 99.38%
84.4%
99.6%
96.2%
Google Sycamore Isolated
N/A
N/A
 99.85%
N/A
N/A
 99.64%
N/A
N/A
96.9%
QuTech Spin-2
N/A
N/A
 ~99.0%
N/A
N/A
 N/A
N/A
N/A
~85%10
QuTech Starmon-5
N/A
N/A
 ~99.8%
N/A
N/A
 ~99.5%
N/A
N/A
~95%10

Table 3

Qubit Gate Delays
Computer
Min
Max
Average
Rigetti 16Q Aspen-1
188 ns.
464 ns.
323 ns. (2-Qubit)
~60 ns. (Single)
Rigetti Aspen-7
N/A
N/A
340 ns. (2-Qubit)
80 ns. (Single)
Google Sycamore
N/A
N/A
12 ns. (2-Qubit)
25 ns. (Single)
QuTech Spin-2
N/A
N/A
150 ns. (2-Qubit)
250 ns. (Single)
QuTech Starmon-5
N/A
N/A
80 ns. (2-Qubit)
20 ns. (Single)

Table 4

Computer Reference Date
IBM Q5 Tenerife
https://quantumexperience.ng.bluemix.net/qx/devices
5/31/2019
IBM Q5 Yorktown
https://quantum-computing.ibm.com/
10/3/2019
IBM Vigo
https://quantum-computing.ibm.com/
10/3/2019
IBM Oursense
https://quantum-computing.ibm.com/
10/3/2019
IBM Q16 Melbourne
https://quantum-computing.ibm.com/
10/3/2019
IBM Q20 Poughkeepsie
https://www.ibm.com/blogs/research/2019/03/power-quantum-device/
3/4/2019
IBM Q20 Tokyo
https://www.ibm.com/blogs/research/2019/03/power-quantum-device/
3/4/2019
IBM Q20 Austin
https://quantumexperience.ng.bluemix.net/qx/devices
3/29/2019
IBM Q System One
https://www.ibm.com/blogs/research/2019/03/power-quantum-device/
3/4/2019
IBM Almaden Boeblingen Singapore
https://www.ibm.com/blogs/research/2019/09/quantum-computation-center/
9/18/2019
IBM Rochester
https://www.ibm.com/blogs/research/2019/09/quantum-computation-center/
9/18/2019
Rigetti 16Q  Aspen-4
https://www.rigetti.com/qpu
5/30/2019
Rigetti Aspen-7
https://aws.amazon.com/braket/hardware-providers/#Rigetti
10/24/2019
Rigetti Aspen-8
https://www.rigetti.com/
5/20/2020
IonQ 11 Qubit
https://ionq.co/news/december-11-2018#appendix
https://arxiv.org/abs/1903.08181
3/19/2018
Honeywell 4 Qubit
https://www.honeywell.com/content/dam/honeywell/files/Beta_10_Quantum_3_3_2020.pdf
3/3/2020
Google Sycamore
https://www.nature.com/articles/s41586-019-1666-5
https://static-content.springer.com/esm/art%3A10.1038%2Fs41586-019-1666-5/MediaObjects/41586_2019_1666_MOESM1_ESM.pdf
10/23/2019
QuTech Spin-2 and Starmon-5
https://qutech.nl/wp-content/uploads/2020/04/1.-Technical-Fact-Sheet-Quantum-Inspire.pdf
4/20/2020

Notes

  1. For most of the parameters we show the Min, Max, and Average values.   Since both IBM and Rigetti publicize the individual values for every qubit, the Min shows the value for the worst of the qubits, the Max shows the value for the best of the qubits, and the Average shows the mean calculations for all of the qubits.
  2. The connectivity shows the number of connections from a qubit to a other qubits in the array for use in creating a CNOT gate.  The higher the connectivity, the easier it would be to fit a quantum calculation into the structure.  At this time, we do not differentiate on the flexibility of a connection.  For example, if qubit 1 is connected to qubit 2, many implementation require one of the qubits to be the CONTROL and the other qubit to be the TARGET.  Some implementations may be flexible enough so that either qubit can serve as the CONTROL and either qubit can serve as the TARGET.  That implementation may have some configuration advantages, but for the purposes of the table we are still only counting it as one connection and not as two.
  3. The T1 measure is called the relaxation time and the T2 or T2* measure is called the decoherence time.  For details of these definitions we refer you to this paper.  Note that IBM only publishes the T2 times while Rigetti only publishes the T2* time.  The measures are similar, but not exactly the same.
  4. Rigetti and Google are the only ones that publish detailed gate delay information.  We have included these measures in Table 3.
  5. The IBM reference link in Table 4 may require you to register for the IBM Q Experience.  If you click on this link it may ask you for a logon and password to see in more detail the referenced data.
  6. The IonQ Readout Fidelity measure of 99.3% includes both state preparation and measurement errors.
  7. Google publishes both isolated and simultaneous gate fidelity numbers.  For this table we are showing both the simultaneous numbers and the isolated numbers.  The simultaneous numbers are slightly worse, but more realistic, in our opinion. But to provide the best comparison we can, we also show the isolated numbers because we think that is what some of the other platforms are using.
  8. For the Rigetti Aspen-7, the Readout Value is the SPAM (State Preparation and Measurement) value which will always be slightly lower than the pure readout value because it includes any state preparation errors.
  9. Honeywell published SPAM values.  However, they believe that these errors are dominated by measurement errors.
  10. The QuTech readout numbers include both initialization plus readout.
  11. Questions, suggestions, and any additions you may have to the data are welcomed.   You can send them to info@quantumcomputingreport.com.