Researchers assessed the need for supercomputers for quantum systems — ForkLog

# Researchers assessed the need for supercomputers for quantum systems

Researchers from Amazon Web Services (AWS), Nvidia, the Lawrence Berkeley National Laboratory, and NASA proposed a model to evaluate how quantum processors interact with classical supercomputers.

It makes it possible to determine when quantum hardware needs to be placed close to high-performance computing infrastructure, and when remote access via the cloud is enough for operation.

Requirements depend on the type of task

The researchers divided quantum-classical interaction into two layers:

1. The first is related to real-time hardware control. This includes calibration of physical qubits and processing data for quantum error correction.
2. The second covers hybrid algorithms in which a classical computer and a quantum processor alternately execute parts of the computation.

During error correction, the classical system analyzes syndrome measurements and determines which failures need to be fixed. Processing must fit into time windows ranging from fractions of a microsecond to several microseconds. For such tasks, a direct connection with minimal latency is required. Remote connection over a standard network will not work.

In the second case, latency affects the runtime duration, but it does not always determine whether the algorithm can be executed. Therefore, some tasks only need cloud access to the quantum device.

Remote access is not suitable for all algorithms

The authors broke down the runtime of a hybrid algorithm into classical computation, operation of the quantum processor, and data exchange.

If most of the loop is devoted to information processing, reducing network latency does not nearly speed up the calculation. If the algorithm requires frequent exchanges, a fast connection becomes critical.

The first example was sample-based quantum diagonalization (SQD), used in quantum chemistry. The authors used experimental data from a 77-qubit IBM Heron processor. In this scenario, most of the time was spent on classical processing of the results. Communication overhead accounted for about 0.0001, so remote access to the quantum processor was sufficient.

A different result was shown by the Quantum-accelerated Markov chain Monte Carlo method (QE-MCMC), tested on a 10-qubit IBM device. The algorithm requires frequent sequential exchanges. With remote access, communication overhead exceeded computation time by about 1000 times.

However, it did not need the power of a full-fledged supercomputer. A small classical controller with a low-latency connection was enough.

Error correction will increase requirements

As systems move to logical qubits and fault-tolerant quantum computers, the role of classical infrastructure will grow. The system must continuously process measurements, detect errors, and send corrective commands. The speed of the classical portion will start to determine the pace of the quantum computer’s work.

The authors expect that large fault-tolerant installations will require tighter integration with high-performance computing systems. At the same time, there is no universal architecture. Requirements depend on the algorithm, hardware characteristics, the number of exchanges, and the volume of data being transferred.

The researchers propose using the model to regularly reassess infrastructure as quantum processors and computing methods evolve.

Recall that in July Nvidia opened the code for an AI module for pre-processing quantum error signals. In the simulation, coupling with a classical decoder reduced the rate of logical failures and accelerated processing.