top of page


Quantum circuit emulation with unbeatable performance on every circuit


Fermioniq's Ava

< 34 qubits Nvidia's cuStateVec

On a single H100 GPU: time per gate for cuStateVec and Ava (pure-state, bond dimension 1024).

Circuit: transverse-field Ising model on a 2d lattice, depths 129-137, gate counts 2412-7020.

Scaling up

For few qubits the fastest way to emulate a circuit is with a full-state emulation.

For many qubits full-state methods become increasingly difficult (and expensive!) due to an exponential scaling in memory.


This can be avoided by using tensor networks to compress the quantum state.


Compressed emulation

Tensor networks compress quantum states by eliminating entanglement.

Less entangled states can be represented perfectly with few resources. For instance, the low-entanglement properties of the Quantum Fourier Transform allow for lossless emulation of 1000s of qubits on a single GPU, shown below:

More entangled states require lossy compression as the state is squeezed into the available memory. Using Ava, this loss can be analyzed and controlled, allowing users to trade accuracy for speed.

Cloud access

The Fermioniq client and API allows users to seamlessly integrate Ava into their workflows.

Jobs are easily configurable, allowing users to choose

  • The emulation mode (statevector / tensor network)

  • Backend hardware (CPU or GPU)

  • Type of output (samples / amplitudes / expectation values / state)

  • And a plethora of other options to tailor the emulation to their needs.


Our intuitive noise model builders allow the user to easily create noise models with features beyond those found in Cirq and Qiskit, such as cross-talk.


With support for server-side training of parameterized circuits, users can optimize variational quantum circuits via the client with a single job submission.


Intermediate measurements and classical control are fully supported, allowing users to simulate sophisticated routines such as error correction and state preparation.



We have thoroughly benchmarked Ava across a variety of realistic quantum circuits to give users an understanding of how Ava will perform for their application.


Ava's performance on circuits implementing time evolution of the 2D transverse-field Ising model on 36 to 100 qubits.

To read about our approach to benchmarking Ava, and to see the results,  check out our benchmarks page.


Fermioniq provides extensive support to users through detailed documentation and code examples in notebooks, ensuring a smooth and easy start as well as a guided exploration through all of our features. Our API reference explains how to integrate Ava into your current workflow.

bottom of page