Optimising Circuits

Artificial intelligence or, better, neural networks accelerate compilation in quantum computing: this is a finding of the Quantum Integration Centre (QIC) at the Leibniz Supercomputing Centre (LRZ). For this purpose, the experts simulated possible quantum circuits of a smaller system and tested their performance and possibilities for optimisation with the help of a specially constructed neural network: "Quantum computing is often noisy, errors can creep into the final results, so the quality of the circuits is important," explains Burak Mete, researcher in the team Quantum Computing and -Technologies (QCT). "Our neural network tests these and predicts optimisability." To execute algorithms on a quantum computer, its smallest computational units, the quantum or qubits, are entangled with each other. Not every circuit created in this way is efficient or can be adapted to a desired application. Testing takes even longer than in classical computing, which in turn hinders the development of operating software and the integration of quantum processors into supercomputing. The neural network, meanwhile, pre-sorts circuits and thus shortens the compilation time: "It predicts optimisability with 96 percent accuracy," Mete continues. It will now be confronted with larger quantum systems as well as different quantum technologies. "It is quite possible that we will be able to train the network in such a way that it not only reveals the best circuits, but also determines the selection of the appropriate quantum technology," Mete continues. The way qubits are created–  the LRZ works primarily with semiconductor technology, but ion traps are also possible, and quantum bits can also be obtained from diamonds, photons or neutral atoms– possibly influences later applications. This experience also influences the software development for the future technology.

Link to the paper