Skip to main content
European Commission logo print header

Neuromrophic Quantum Computing

Periodic Reporting for period 1 - Quromorphic (Neuromrophic Quantum Computing)

Período documentado: 2019-06-01 hasta 2020-05-31

The project aims to introduce human brain inspired hardware with quantum functionalities: The goal is to build superconducting quantum neural networks to develop dedicated, neuromorphic quantum machine learning hardware, which can, in its next generation, outperform classical von Neumann architectures. In contrast to established machine learning approaches that emulate neural function in software on conventional von Neumann hardware, neuromorphic quantum hardware can offer a significant advantage if applied to superpositions of multiple batches of real world data in parallel. This feature is expected to lead to a quantum advantage. Quromorphic aims to provide proof of concept demonstrations of this new technology and a roadmap for the path towards its exploitation. To achieve this breakthrough, we will implement two classes of quantum neural networks that have immediate applications in quantum machine learning, feed forward networks and non-equilibrium quantum annealers. This effort will be completed by the development of strategies for scaling the devices to the threshold where they will surpass the capabilities of existing machine learning technology and achieve quantum advantage. In preparation for future exploitation of this new technology, we will run simulations to explore its application to real world problems.
In the first 12 months of the project, implementation of the devices has started. For the feed forward quantum neural networks that we plan to build we have already been able to show nonlinear activation on single neurons, whereas for a different network, a Quantum Boltzmann Machine we have realized the tuneable coupler between the neurons and shown good control over the circuit.
For our aim to enhance the scalability of quantum neural networks, we have introduced the theory for a faster and more robust activation scheme for the neurons we build and devised a blueprint for a single step Toffoli gate that can be employed to control the activation of one neuron controlled by two other neurons in one single step.
To develop the training and applications of quantum neural networks, we have shown that some networks, e.g. the QAOA algorithm can be trained using Tensor Network techniques, where only the trained network then runs on the quantum processor and exploits its capabilities. In addition we have contributed to the development of TensorFlow Quantum.
Until the end of the project we expect to implement prototypes of quantum neural networks in superconducting circuits and to proof of principle demonstrations of their capabilities. We expect that this will trigger significant further research activity in this direction with the aim to ultimately exploit the newly emerging technology for applications of commercial relevance.
fridge at IBM
chip at ETH Zuerich
chip at TU Delft