Researchers from the Ecole Polytechnique Fédérale de Lausanne (EPFL) and several international institutions have identified a fundamental constraint limiting the performance of modern quantum computers. Their study, published in Nature Physics, shows that accumulated noise causes quantum circuits to "forget" their initial operations, leaving only the final layers of a calculation to influence the result.
Quantum circuits function through a sequence of operations that rely on precise coordination to solve complex problems. However, this process is vulnerable to "noise," or environmental instability, which disrupts the delicate state of the qubits. The research team, led by Armando Angrisani and Yihui Quek, determined that this noise acts as a filter that degrades data as it progresses through the circuit.
The decay of computational depth
By modeling large groups of quantum circuits, the scientists tracked how information flows through layers of two-qubit operations. They found that in any noisy environment, the influence of early steps fades rapidly. Consequently, even if a circuit is designed to be deep, its actual computational output is dominated by the final few layers.
This behavior explains why deep quantum circuits often fail to outperform their shallower counterparts. Adding more steps to a noisy system does not necessarily increase processing power because the earlier, more complex operations become irrelevant to the final outcome. In many cases, the accumulated noise makes these deep systems act like much simpler, lower-capacity models.
These findings suggest that current strategies for scaling quantum technology face a practical ceiling. For many applications, particularly those involving local measurements, increasing the number of operations provides diminishing returns. The researchers emphasize that future advancements must prioritize noise reduction or the development of circuit architectures that can maintain coherence despite external interference.
Additionally, the study warns against overestimating the capabilities of current hardware. Because noise simplifies the effective complexity of a circuit, some deep systems may appear easier to train or simulate than their theoretical design suggests. Scientists must account for this "fading memory" effect to avoid setting unrealistic expectations for what today’s quantum machines can actually calculate.
