EPFL Research Reveals Noise Accumulation Imposes Strict Depth Limits on Near Term Quantum Computing Circuits

New research in Nature Physics shows how noise erases memory in deep quantum circuits, limiting near term hardware to shallow operational depths.

By: AXL Media

Published: Apr 2, 2026, 8:02 AM EDT

Source: The information in this article was sourced from Eurekalert

The Cumulative Decay of Quantum Operations

Theoretical physicists have identified a fundamental constraint on the complexity of quantum circuits, likening the build up of internal interference to a wobbly chain of dominoes. As a quantum circuit processes information through a series of tiny operational steps, "noise" or environmental interaction begins to accumulate at each stage. According to the research led by Armando Angrisani and Yihui Quek, this noise eventually reaches a crescendo that overwhelms the system, effectively erasing the computational work performed in the earlier stages of the chain.

The Vanishing Memory of Deep Circuits

The study utilized mathematical modeling to track how individual two qubit operations propagate through a circuit under realistic noise conditions. The results showed that in deep, noisy circuits, the final output is predominantly shaped only by the last few layers of operations. This "fading memory" effect means that the initial steps, which are often the most critical for complex problem solving, are wiped out by the time the measurement is taken. This finding suggests that simply adding more layers to a noisy hardware setup does not necessarily increase its total computational power.

Revisiting the Trainability of Noisy Hardware

One surprising outcome of the analysis involves the "trainability" of these systems for specific tasks. Because noise weakens the overall power of the circuit, it prevents the formation of "barren plateaus," which are regions in a mathematical landscape where a system becomes impossible to optimize. While this makes it easier to adjust or train a noisy circuit, the researchers warn that this is a double edged sword. The system remains trainable only because it has been functionally reduced to a much shallower version of itself, limiting its ability to outperform classical machines.