Researchers from Amazon Web Services, Nvidia, Lawrence Berkeley National Laboratory and NASA have proposed a new model. This model evaluates the interaction of quantum processors with classical supercomputers. It helps to determine the optimal placement of quantum equipment. In some cases, physical proximity to high-performance infrastructure is required. In other situations, remote cloud access is sufficient. The authors divided the quantum-classical interaction into two main levels. The first level is related to real-time equipment management. This includes calibration of physical qubits and processing of quantum error correction data. The second level covers hybrid algorithms. In them, a classical computer and a quantum processor alternately perform parts of complex calculations.

When correcting errors, the classical system analyzes control measurements to identify failures. This processing should be carried out in intervals from fractions to several microseconds. For such tasks, direct communication with minimal network latency is required. The usual remote connection via a standard network is completely unsuitable here. In the second case, the delay affects the overall duration of the operation, but it does not always determine whether the algorithm can be executed. Therefore, cloud access to a quantum device is quite sufficient for some tasks. For example, the sampling-based quantum diagonalization method has been used in chemistry. There, the main part of the cycle was occupied by classical processing of the results. The share of communication costs was only about 0.0001, so remote access was sufficient.

The Monte Carlo method for Markov chains with quantum acceleration showed a different result. This algorithm requires frequent sequential data exchanges between the systems. With remote access, the communication costs exceeded the computing time by about a thousand times. However, he still didn't need the power of a full-fledged supercomputer. A small classic controller with low connection latency was enough to complete the task. As we move to logical qubits, the role of classical infrastructure will steadily grow. The system must continuously process measurements and send correction commands. The authors expect that large fault-tolerant installations will need closer integration. However, there is no universal architecture, as the requirements depend on the specific algorithm.