The Signal
DLR researchers report shallower circuits for quantum data loading
A DLR and University of Cologne paper presents two methods for simplifying quantum state-preparation circuits and reports large reductions in circuit depth against Qiskit’s default method. The researchers also demonstrate the approach on IQM hardware, encoding a 20-qubit state on a square-lattice device.

Methods for reducing circuit depth
Researchers at the German Aerospace Center (DLR) Institute of Software Technology and the University of Cologne have introduced methods for encoding data into quantum states with shallower circuits. Their paper, “Scalable Unitary State-Encoding for Generic Large Data,” was submitted to arXiv on Nov. 6, 2026, as a preprint.
The work addresses data loading, which the researchers describe as a bottleneck for quantum computing protocols that require classical data to be encoded into quantum states. The paper presents deterministic amplitude-encoding methods with low two-qubit gate costs and adjustable trade-offs between circuit depth and encoding fidelity. The authors write: “Our approaches require no amplitude amplification or mid-circuit measurement and are capable of encoding arbitrary complex data or queryable functions by using tensor-networks to go beyond what can be held in classical memory.”
Two approaches to multiplexer circuits
The paper introduces two algorithms for simplifying fully controlled multiplexer circuits into one- and two-qubit gates. One frames the removal of redundant gate controls as a Boolean logic minimization problem. The other uses a single ancilla qubit and the Walsh–Hadamard spectrum of gate angles to express multi-controlled rotation gates as sequences of one- and two-qubit operations.
Quantum Computing Report identifies the methods as the Binary-Tree Boolean Encoder, or BT-BE, and the Binary-Tree Walsh–Hadamard Transform, or BT-WHT. It describes BT-BE as reducing multi-controlled rotation counts in shallow multiplexer layers. For deeper layers, the report says, BT-WHT keeps selected spectral components and arranges the corresponding rotations along an optimized Hamming path. The paper describes truncating small-magnitude Walsh coefficients.
The paper also describes using tensor networks to encode arbitrary complex data or queryable functions, including cases beyond what can be held in classical memory. Quantum Computing Report says Tensor Cross Interpolation can sample the angle tree to construct matrix product state representations without first filling dense arrays. The approaches address both circuit simplification and the handling of the angles needed for state preparation.
Benchmarks and comparisons
The researchers benchmarked their methods across a range of function classes, measuring circuit depth against system size and target fidelity in an all-to-all connectivity graph. Quantum Computing Report lists Lorentzian, polynomial and financial jump-diffusion distributions among its reported benchmark cases.
The arXiv abstract reports up to orders-of-magnitude improvement in circuit depth at equivalent state fidelity compared with the default state-preparation method in Qiskit. The abstract specifies that the benchmarks characterize total depth in an all-to-all connectivity graph.
Test on IQM hardware
The researchers also demonstrate their method on IQM hardware with square-lattice connectivity. Quantum Computing Report identifies the device as IQM Emerald and describes an experiment encoding a 20-qubit state representing a two-dimensional Gaussian distribution.
The report gives a total circuit depth of 56, including a CZ depth of 26, and a measured state infidelity of 0.011 ± 0.002 across 256,000 shots. It also reports that the experiment used Pauli twirling and dynamical decoupling. The arXiv abstract describes the hardware demonstration as achieving excellent fidelity with minimal noise mitigation. The paper argues that reducing encoding depth could allow applications to run on current and near-term hardware before noise overwhelms the computation.