Circuit Folding: Scalable and Graph-Based Circuit Cutting via Modular Structure Exploitation

TL;DR

CIFOLD is a novel graph-based framework that exploits modular structures in quantum circuits to improve the scalability and efficiency of circuit cutting, enabling better partitioning with less computational effort.

Contribution

The paper introduces CIFOLD, a new method that leverages repetitive modular structures to optimize quantum circuit cutting and reduce computational complexity.

Findings

CIFOLD reduces the number of cuts by an average of 31.6%.

It lowers sampling overhead by 3.55×10^9.

It outperforms existing circuit-cutting techniques in quality and efficiency.

Abstract

Circuit cutting is a promising technique that leverages both quantum and classical computational resources, enabling the practical execution of large quantum circuits on noisy intermediate-scale quantum (NISQ) hardware. Recent approaches typically focus exclusively on either gate cuts or wire cuts, modeling quantum circuits as graphs. However, identifying optimal cutting locations using this representation often results in prohibitively high computational complexity, especially under realistic hardware constraints. In this paper, we introduce CIFOLD, a novel graph-based framework that exploits repetitive modular structures inherent in quantum algorithms, significantly enhancing the scalability and efficiency of circuit cutting. Our approach systematically folds quantum circuits into compact meta-graphs by identifying and merging common gate sequences across entangled qubits, dramatically simplifying subsequent partitioning tasks. We define folding factor and variance to quantify circuit compression and ensure balanced folding. Using these condensed representations, CIFOLD precisely identifies cut locations without exhaustive global graph searches. We perform extensive experiments, comparing CIFOLD with state-of-the-art circuit-cutting techniques. Results demonstrate that CIFOLD achieves superior partition quality and computational efficiency, reducing the number of required cuts by an average of 31.6% and lowering the sampling overhead substantially by 3.55*10^9. Our findings illustrate that CIFOLD represents a significant advancement toward scalable quantum circuit cutting.