The Quantum Processing Unit (QPU)

A quantum processing unit (QPU), also referred to as a quantum chip, is a physical (fabricated) chip that contains a number of interconnected qubits. It is the foundational component of a full quantum computer, which includes the housing environment for the QPU, the control electronics, and many other components.

This page describes how to use the Forest API for interacting with Rigetti QPUs, and provides technical details and average performance of Agave, the 8Q QPU currently available, that has been designed, fabricated and packaged by Rigetti.

Using the QPU


User permissions for QPU access must be enabled by a Forest administrator. QPUConnection calls will automatically fail without these user permissions. Speak to a Forest administrator for information about upgrading your access plan.

One establishes a connection to a Rigetti QPU in the same manner as a QVM:

from pyquil.api import QPUConnection
qpu = QPUConnection() # NOTE: This raises a UserWarning!

There is one caveat, however, as shown in the UserWarning that is raised by the above command: You must specify a device as an argument. This is described in the following section.

Accessing available devices with get_devices()

The initialization function for a QPUConnection object must be provided a speciffic Rigetti QPU as an argument, so that Forest knows on which quantum computer you want to execute your programs. The available QPUs, synonymously referred to as devices in Forest, can be inspected via the function get_devices in the api module:

from pyquil.api import get_devices
for device in get_devices():
    if device.is_online():
        print('Device {} is online'.format(


The Device objects returned by get_devices captures other characteristics about the associated QPU, such as its connectivity, coherence times, single- and two-qubit gate fidelities. For more information on the Device class, see Getting QPU Information from the Device Class.

Devices are typically named according to the convention [n]Q-[name], where n is the number of active qubits on the device and name is a human-readable name that designates the device.

Execution on the QPU

One may execute Quil programs on the QPU (nearly) identically to the QVM, via the .run(...) method (obviously, since the QPU is a real quantum computer, the .wavefunction(...) method is not available). We may fix the above example then by providing a device to the QPUConnection:

from pyquil.api import get_devices, QPUConnection

agave = get_devices(as_dict=True)['8Q-Agave']
qpu = QPUConnection(agave)
# The device name as a string is also acceptable
# qpu = QPUConnection('8Q-Agave')

You have now established a connection to the 8Q-Agave QPU. Executing programs is then identical to the QVM (we may ommit the classical_addresses and trials arguments to use their defaults):

from pyquil.quil import Program
from pyquil.gates import X, MEASURE

program = Program(X(0), MEASURE(0, 0)), [0])

In addition to the .run(...) method, a QPUConnection object provides the following methods:

  • .run(quil_program, classical_addresses, trials=1): This method sends the Program object quil_program to the QPU for execution, which runs the program trials many times. After each run on the QPU, all the qubits in the QPU are simultaneously measured and their results are stored in classical registers according to the MEASURE instructions provided. Then, a list of registers listed in classical_addresses is returned to the user for each trial. This call is blocking: it will wait until the QPU returns its results for inspection.
  • .run_async(quil_program, classical_addresses, trials=1): This method has identical behavior to .run except that it is nonblocking, and it instead returns a job ID string.
  • .run_and_measure(quil_program, qubits, trials=1): This method sends the Program object quil_program to the QPU for execution, which runs the program trials many times. After each run on the QPU, the all the qubits in the QPU are simultaneously measured, and the results from those listed in qubits are returned to the user for each trial. This call is blocking: it will wait until the QPU returns its results for inspection.
  • .run_and_measure_async(quil_program, qubits, trials=1): This method has identical behavior to .run_and_measure except that it is nonblocking, and it instead returns a job ID string.


The QPU’s run functionality matches that of the QVM’s run functionality, but the behavior of run_and_measure does not match its QVMConnection counterpart (cf. Optimized Calls). The QVMConnection version of run repeats the execution of a program many times, producing a (potentially) different outcome each time, whereas run_and_measure executes a program only once and uses the QVM’s unique ability to perform wavefunction inspection to multiply sample the same distribution. The QPU does not have this ability, and thus its run_and_measure call behaves as the QVM’s run.

For example, the following Python snippet demonstrates the execution of a small job on the QPU identified as “8Q-Agave”:

from pyquil.quil import Program
import pyquil.api as api
from pyquil.gates import *
qpu = api.QPUConnection('8Q-Agave')
p = Program(H(0), CNOT(0, 1), MEASURE(0, 0), MEASURE(1, 1)), [0, 1], 1000)

When the QPU execution time is expected to be long and there is classical computation that the program would like to accomplish in the meantime, the QPUConnection object allows for an asynchronous run_async call to be placed instead. By storing the resulting job ID, the state of the job and be queried later and its results obtained then. The mechanism for querying the state of a job is also through the QPUConnection object: a job ID string can be transformed to a pyquil.api.Job object via the method .get_job(job_id); the state of a Job object (taken at its creation time) can then be inspected by the method .is_done(); and when this returns True the output of the QPU can be retrieved via the method .result().

For example, consider the following Python snippet:

from pyquil.quil import Program
import pyquil.api as api
from pyquil.gates import *
qpu = api.QPUConnection('8Q-Agave')
p = Program(H(0), CNOT(0, 1), MEASURE(0, 0), MEASURE(1, 1))
job_id = qpu.run_async(p, [0, 1], 1000)
while not qpu.get_job(job_id).is_done():
    ## get some other work done while we wait
    ## and eventually yield to recheck the job result
## now the job is guaranteed to be finished, so pull the QPU results
job_result = qpu.get_job(job_id).result()

Getting QPU Information from the Device Class

The pyQuil Device class provides useful information for learning about, and working with, Rigetti’s available QPUs. One may query for available devices using the get_devices function:

from pyquil.api import get_devices

devices = get_devices(as_dict=True)
# E.g. {'8Q-Agave': <Device 8Q-Agave online>, '19Q-Acorn': <Device 19Q-Acorn offline>}

agave = devices['8Q-Agave']

The variable agave points to a Device object that holds useful information regarding the QPU, including:

  1. Connectivity via its instruction set architecture (agave.isa of class ISA).
  2. Hardware specifications such as coherence times and fidelities (agave.specs of class Specs).
  3. Noise model information (agave.noise_model of class NoiseModel).

These 3 attributes are accessed in the following ways (note that the specs shown below may be out of date):

  # Specs(qubits_specs=..., edges_specs=...)

  [_QubitSpecs(id=0, fRO=0.7841, f1QRB=0.9575, T1=1.07e-05, T2=1.06e-05),
   _QubitSpecs(id=1, fRO=0.9099, f1QRB=0.9513, T1=1e-05, T2=9.2e-06),
   _QubitSpecs(id=2, fRO=0.9427, f1QRB=0.9825, T1=1.55e-05, T2=1.25e-05),
   _QubitSpecs(id=3, fRO=0.9122, f1QRB=0.9703, T1=1.42e-05, T2=1.85e-05),
   _QubitSpecs(id=4, fRO=0.6777, f1QRB=0.9693, T1=1.46e-05, T2=2.62e-05),
   _QubitSpecs(id=5, fRO=0.8319, f1QRB=0.9624, T1=1.49e-05, T2=1.28e-05),
   _QubitSpecs(id=6, fRO=0.7526, f1QRB=0.969, T1=1.42e-05, T2=1.29e-05),
   _QubitSpecs(id=7, fRO=0.8954, f1QRB=0.9322, T1=1.32e-05, T2=1.77e-05)]

  # ISA(qubits=..., edges=...)

[Edge(targets=[0, 1], type='CZ', dead=False),
 Edge(targets=[0, 7], type='CZ', dead=False),
 Edge(targets=[1, 2], type='CZ', dead=False),
 Edge(targets=[2, 3], type='CZ', dead=False),
 Edge(targets=[3, 4], type='CZ', dead=False),
 Edge(targets=[4, 5], type='CZ', dead=False),
 Edge(targets=[5, 6], type='CZ', dead=False),
 Edge(targets=[6, 7], type='CZ', dead=False)]

  # NoiseModel(gates=[KrausModel(...) ...] ...)

Additionally, the Specs class provides methods for access specs info across the chip in a more succinct manner:

# {0: 1.07e-05, 1: 1e-05, 2: 1.55e-05, 3: 1.42e-05, 4: 1.46e-05, 5: 1.49e-05, 6: 1.42e-05, 7: 1.32e-05}

# {(0, 1): 0.9176, (0, 7): 0.9095, (1, 2): 0.9065, (2, 3): 0.8191, (3, 4): 0.87, (4, 5): 0.67, (5, 6): 0.9302, (6, 7): 0.9295}

With these tools provided by the Device class, users may learn more about Rigetti hardware, and construct programs tailored specifically to that hardware. The Device class can also be used to seed a QVM with characteristics of the device, supporting noisy simulation. For more information on this, see the next section.

Simulating the QPU using the QVM

The QVM is a powerful tool for testing quantum programs before executing them on the QPU. In addition to the module for generating custom noise models for simulating noise on the QVM, pyQuil provides a simple interface for loading the QVM with noise models tailored to Rigetti’s available QPUs, in just one modified line of code. This is made possible via the Device class, which holds hardware specification information, noise model information, and instruction set architecture (ISA) information regarding connectivity. This information is held in the Specs, ISA and NoiseModel attributes of the Device class, respectively.

Specifically, to load a QVM with the NoiseModel information from a Device, all that is required is to provide a Device object to the QVM during initialization:

from pyquil.api import get_devices, QVMConnection

agave = get_devices(as_dict=True)['8Q-Agave']
qvm = QVMConnection(agave)

By simply providing a device during QVM initialization, all programs executed on this QVM will, by default, have noise applied that is characteristic of the corresponding Rigetti QPU (in the case above, the agave device). One may then efficiently test realistic quantum algorithms on the QVM, in advance of running those programs on the QPU.

Retune Interruptions

Because the QPU is a physical device, it is occasionally taken offline for recalibration. This offline period typically lasts 10-40 minutes, depending upon QPU characteristics and other external factors. During this period, the QPU will be listed as offline, and it will reject new jobs (but pending jobs will remain queued). When the QPU resumes activity, its performance characteristics may be slightly different (in that different gates may enjoy different process fidelities).

Agave QPU Properties

The quantum processor consists of 8 superconducting transmon qubits with fixed capacitive coupling in the planar lattice design shown in Fig. 1.

The resonant frequencies of qubits 0, 2, 4, and 6 are tunable while qubits 1, 3, 5, and 7 are fixed. The former set has two Josephson junctions in an asymmetric SQUID geometry to provide roughly 1 GHz of frequency tunability, and flux-insensitive “sweet spots” near

\(\omega^{\textrm{max}}_{01}/2\pi\approx 4.8 \, \textrm{GHz}\)


\(\omega^{\textrm{min}}_{01}/2\pi\approx 3.3 \, \textrm{GHz}\).

These tunable devices are coupled to bias lines for AC and DC flux delivery. Each qubit is capacitively coupled to a quasi-lumped element resonator for dispersive readout of the qubit state. Single-qubit control is effected by applying microwave drives at the resonator ports. Two-qubit gates are activated via RF drives on the flux bias lines.


\(\textbf{Figure 1 $|$ Connectivity of Rigetti 8Q. a,}\) Chip schematic showing tunable transmons (green circles) capacitively coupled to fixed-frequency transmons (blue circles). \(\textbf{b}\), Optical image of an 8Q chip, representative of Agave.

1-Qubit Gate Performance

The device is characterized by several parameters:

  • \(\omega_\textrm{01}/2\pi\) is the qubit transition frequency
  • \(\omega_\textrm{r}/2\pi\) is the resonator frequency
  • \(\eta/2\pi\) is the anharmonicity of the qubit
  • \(g/2\pi\) is the coupling strength between a qubit and a resonator
  • \(\lambda/2\pi\) is the coupling strength between two neighboring qubits

In Rigetti 8Q, each tunable qubit is capacitively coupled to two fixed-frequency qubits. We use a parametric flux modulation to activate a controlled Z gate between tunable and fixed qubits. The typical time-scale of these entangling gates is in the range 100–250 ns.

Table 1 summarizes the main performance parameters of Rigetti 8Q. The resonator and qubit frequencies are measured with standard spectroscopic techniques. The relaxation time \(T_1\) is extracted from repeated inversion recovery experiments. Similarly, the coherence time \(T^*_2\) is measured with repeated Ramsey fringe experiments. Single-qubit gate fidelities are estimated with randomized benchmarking protocols in which a sequence of \(m\) Clifford gates is applied to the qubit followed by a measurement on the computational basis. The sequence of Clifford gates are such that the first \(m-1\) gates are chosen uniformly at random from the Clifford group, while the last Clifford gate is chosen to bring the state of the system back to the initial state. This protocol is repeated for different values of \(m\in \{2,4,8,16,32,64,128\}\). The reported single-qubit gate fidelity is related to the randomized benchmarking decay constant \(p\) in the following way: \(\mathsf{F}_\textrm{1q} = p +(1-p)/2\). Finally, the readout assignment fidelities are extracted with dispersive readouts combined with a linear classifier trained on \(|0\rangle\) and \(|1\rangle\) state preparation for each qubit. The reported readout assignment fidelity is given by expression \(\mathsf{F}_\textrm{RO} = [p(0|0)+p(1|1)]/2\), where \(p(b|a)\) is the probability of measuring the qubit in state \(b\) when prepared in state \(a\).

\(\textbf{Table 1 | Rigetti 8Q performance}\)
  \(\omega^{\textrm{max}}_{\textrm{r}}/2\pi\) \(\omega^{\textrm{max}}_{01}/2\pi\) \(T_1\) \(T^*_2\) \(\mathsf{F}_{\textrm{1q}}\) \(\mathsf{F}_{\textrm{RO}}\)
  \(\textrm{MHz}\) \(\textrm{MHz}\) \(\mu\textrm{s}\) \(\mu\textrm{s}\)    
0 5863 4586 10.72 10.6 0.957 0.784
1 5293 3909 10.04 9.2 0.951 0.910
2 5713 4524 15.52 12.5 0.982 0.943
3 5411 4054 14.17 18.5 0.970 0.912
4 5620 4660 14.58 26.2 0.969 0.678
5 5171 4081 14.86 12.8 0.962 0.832
6 5751 4760 14.17 12.9 0.969 0.753
7 5454 4110 13.19 17.7 0.932 0.895

Qubit-Qubit Coupling

The coupling strength between two qubits can be extracted from a precise measurement of the shift in qubit frequency after the neighboring qubit is in the excited state. This protocol consists of two steps: a \(\pi\) pulse is applied to the first qubit, followed by a Ramsey fringe experiment on the second qubit which precisely determines its transition frequency (see Fig. 2a). The effective shift is denoted by \(\chi_\textrm{qq}\) and typical values are in the range \(\approx 100 \, \textrm{kHz}\). The coupling strength \(\lambda\) between the two qubits can be calculated in the following way:

\[\lambda^{(1,2)} = \sqrt{\left|\frac{\chi^{(1,2)}_\textrm{qq} \left[\,f^\textrm{(1)}_{01}-f^\textrm{(2)}_{12}\right]\left[\,f^\textrm{(1)}_{12}-f^\textrm{(2)}_{01}\right]}{2(\eta_1+\eta_2)}\right|}\]

Figure 2b shows the coupling strength for our device. This quantity is crucial to predict the gate time of our parametric entangling gates.

alternate text

\(\textbf{Figure 2 $|$ Coupling strength. a,}\) Quantum circuit implemented to measure the qubit-qubit effective frequency shift. \(\textbf{b,}\) Capacitive coupling between neighboring qubits expressed in MHz.

2-Qubit Gate Performance

Table 2 shows the two-qubit gate performance of Rigetti 8Q. These parameters refer to parametric CZ gates performed on one pair at a time. We analyze these CZ gates through quantum process tomography (QPT). This procedure starts by applying local rotations to the two qubits taken from the set \(\{I,R_x(\pi/2),R_y(\pi/2),R_x(\pi)\}\), followed by a CZ gate and post-rotations that bring the qubit states back to the computational basis. QPT involves the analysis of \(16\times16 =256\) different experiments, each of which we repeat \(500\) times. The reported fidelity \(\mathsf{F}^\textrm{cptp}_\textrm{PT}\) is the average gate fidelity [Nielsen2002] of the ideal process and the process matrix inferred via maximum likelihood tomography under complete positivity (cp) and trace preservation (tp) constraints (cf. supplementary material of [Reagor2018]).

\(\textbf{Table 2 | Rigetti 8Q two-qubit gate performance}\)
  \(f_\textrm{m}\) \(t_\textrm{CZ}\) \(\mathsf{F}^\textrm{cptp}_{\textrm{PT}}\)
  \(\textrm{MHz}\) ns  
0 - 1 226 195 0.92
1 - 2 153 198 0.91
2 - 3 138 132 0.82
3 - 4 163 160 0.87
4 - 5 168 163 0.67
5 - 6 107 186 0.93
6 - 7 123 162 0.93
7 - 0 298 118 0.91
[Nielsen2002]Nielsen, M. A. (2002) ‘A simple formula for the average gate fidelity of a quantum dynamical operation’,
[Reagor2018]Reagor, M. et al. (2018) ‘Demonstration of universal parametric entangling gates on a multi-qubit lattice’,