Aggregated Control of Quantum Computations: When Stacked Architectures Are Too Good to Be Practical Soon

Google’s quantum supremacy demonstration pitted the world’s largest supercomputer against a single quantum chip. A path to further scale up may involve a large supercomputer working together with quantum chips, instead of in competition with them. The debate about universal large-scale quantum computers never being built is not a novelty. If the hardware qubits were (almost) perfect, available quantum machines would be practically useful. But this is not the case: qubits are error-prone. Quantum error-correction has to be implemented, and millions of hardware qubits are necessary to execute quantum computations of practical interest.  The main concern is that the control of quantum computers, comprising of millions of qubits, is practically impossible for scalability reasons.

A realistic approach could be to design and implement the control as a kind of quantum operating system (QCOS), based on message passing, executed on supercomputers. Such an architecture should enable the scalable control of millions of physical qubits, and support the fault-tolerance of the OS. A QCOS is classic software running on classic hardware, responsible for preparing, starting, and managing quantum computations. The presented OS architecture referred to aggregation and not distribution, because the OS components are network-attached. We argue for a splitkernel design in the sense that the components of the QCOS related to various tasks, such as compilation and error-correction, are aggregated. In a splitkernel the components are parts of the kernel, and have high execution priority. The aggregated QCOS is such that there is no top-down work-flow like in a stacked QCOS architecture. The impossibility of controlling a large scale quantum computer could be rebutted by a supercomputer aggregated QCOS.