Researchers at EPFL have demonstrated that noise accumulation in quantum circuits imposes a fundamental constraint on their complexity, limiting the practical depth of these circuits even as more operational steps are added. The team, led by Armando Angrisani and Yihui Quek at EPFL, along with colleagues at institutions including the Free University of Berlin and the University of Copenhagen, published their findings in Nature Physics after a broad theoretical analysis of noise’s impact. Their work reveals that in most noisy quantum circuits, “only the last few steps really matter,” as earlier operations are effectively erased by accumulating errors. This means that simply building deeper circuits will not necessarily increase their power for common tasks; instead, progress hinges on improved noise control and circuit design. The study also cautions that the trainability of these circuits is limited because noise has already diminished their overall capabilities.
Noise Accumulation Limits Quantum Circuit Depth
This discovery challenges the prevailing assumption that simply adding more layers to a quantum circuit will automatically enhance its computational capabilities. The research, published in Nature Physics on April 2, 2026, modeled quantum circuits comprised of two-qubit operations subjected to realistic noise after each step, meticulously tracking how this interference propagates through the system. The team’s mathematical treatment of the problem revealed that even complex circuits behave similarly to much shallower ones due to this noise-induced truncation. This limitation has significant implications for the development of near-term quantum machines, suggesting that progress will depend less on simply increasing circuit depth and more on achieving better noise control or designing circuits that leverage specific noise characteristics. The study also cautions that the ability to “train” these noisy circuits for simple tasks is itself a consequence of the weakened computational power caused by noise. The researchers explain that changing the circuit’s settings can still change the outcome, but only because the final layers remain active, highlighting that treating hardware noise as a simple blur could lead to unrealistic expectations about performance.
Two-Qubit Operations Reveal Trainable, Shallow Circuit Behavior
Unlike classical computing where increased complexity generally yields greater power, quantum circuits are susceptible to accumulating errors from qubit “noise,” which degrades performance and ultimately constrains the number of operations that can be reliably chained together. This analysis focused on circuits constructed from two-qubit operations, modeling the realistic impact of noise after each step to understand how it propagates through the system. Interestingly, the study also explains the surprising ability of these noisy circuits to remain “trainable” for certain tasks; the researchers found that the circuits can still be adjusted to produce desired outcomes because the final layers remain the most influential. This work emphasizes the need for advancements in noise control and careful circuit design to unlock the full potential of quantum computation, rather than relying solely on increasing circuit depth.
Simply stacking more layers onto noisy circuits is unlikely to unlock new power for common tasks based on local measurements.
