Researchers have experimentally encoded and decoded 1568 real values on IBM’s Phoenix and Fez quantum computers using a new method called Tessa. This hardware-aware scheme encodes classical data across an entire quantum processor, demonstrating a significant data capacity achieved on current hardware. Tessa builds upon earlier work like QPIXL and QCrank, but differs in its address and data qubits, eliminating the need for SWAP gates or routing. As an example of its versatility, the team encoded a “City-pop cat by M. Fish (Leviatic)” pixel art image, using two values per pixel.
Tessa: Device-Wide Classical Data Encoding on Quantum Computers
This accomplishment used a lattice structure embedding the address register within the device’s coupling graph, allowing for increased storage capacity without requiring complex circuit routing. The system’s efficiency stems from a design that shares address qubits between neighboring data qubits, a technique that minimizes the need for SWAP gates and streamlines data transfer.
The encoding process, detailed in a recent publication, successfully processed diverse datasets including pixel art, audio, MIDI files, DNA sequences, and protein amino acid sequences. Decoding required 8192 shots per basis, a figure determined by the local coupling degree of the quantum processor and representing a trade-off between data fidelity and computational cost.
This method differs from earlier approaches like QPIXL and QCrank through its checkerboard arrangement of qubits, a structural choice that directly addresses the challenges of data encoding. The architecture of Tessa is not limited to specific coupling graphs. It functions with any bipartite coupling graph, offering flexibility in hardware implementation. While amplitude encoding theoretically allows for storing values on n qubits, Tessa’s focus is on practical implementation and maximizing data throughput on current hardware.
The researchers emphasize that the method’s scalability is controllable, allowing for adjustments in circuit depth and the number of shots required to balance performance and resource use. “Sharing address qubits raises the storage density while allowing the data to remain recoverable,” the authors write, highlighting a key advantage of their approach. The code and data used in the experiments are publicly available, facilitating further research and development in the field of quantum data encoding.
All experiments were conducted on IBM’s ibm_fez, a 156-qubit IBM Quantum Heron r2 chip, and ibm_phoenix, a 120-qubit IBM Nighthawk r2 quantum processor, using the Qiskit framework and the IBM quantum platform. The successful reconstruction of recognizable images from every qubit on these devices demonstrates the method’s robustness and accuracy, with errors aligning closely with the team’s local noise model. This work represents a step toward addressing the data encoding problem, a fundamental bottleneck in quantum machine learning and a key area for achieving quantum advantage.
QPIXL and QCrank Foundations for Tessellated Encoding
Tessa builds upon the foundations of QPIXL and QCrank, yet distinguishes itself through a unique architectural approach to data encoding that optimizes resource use on quantum hardware. While QPIXL stores a vector of 2k values using k address qubits and a single data qubit, and QCrank parallelizes encoding by adding data qubits sharing address qubits, Tessa implements a “checkerboard” arrangement. This design strategically positions data qubits to share only a subset of address qubits, a key difference from earlier methods.
This tessellated structure allows Tessa to achieve a higher encoding capacity than a simple arrangement of QPIXL blocks, effectively maximizing the use of available qubits. The method addresses a common challenge in quantum computing: limited qubit connectivity. Unlike QPIXL and QCrank, which require a “star connectivity” from data to address qubits and thus expensive operations, Tessa’s design minimizes the need for long-range gates.
This is particularly beneficial for processors like trapped-ion and neutral-atom systems where non-local gates can be slower or require physical qubit movement. The researchers note that an encoding that functions with a chosen connectivity graph, and fewer long-range gates, can also help on these machines. The development of Tessa was motivated by a desire to move beyond the limitations of existing encoding schemes.
Previous approaches, including divide-and-conquer state preparation and low-depth state preparation, often rely on adding ancilla qubits or splitting preparation across disjoint registers. Tessa, however, aims to improve efficiency by using the existing connectivity of the quantum processor.
Checkerboard Lattice Design for Address and Data Qubits
The tessellated architecture central to Tessa uses a checkerboard lattice to distribute address and data qubits, a design visualized in the paper as a 4×4 example for illustrative purposes. This sharing is not merely an aesthetic choice. It directly impacts the efficiency of data encoding by minimizing the need for complex qubit routing. The construction functions on any bipartite coupling graph, but benefits most from devices where no SWAP gates or routing are required to implement the encoding.
The degree to which data qubits share address qubits directly influences the number of shots required for readout, a critical metric for practical quantum data encoding. The paper details that the shots needed are determined by the local coupling degree, meaning a higher degree, more shared address qubits, divides the required shots among more values.
The construction begins by 2-colouring a desired coupling graph. Each data qubit is the target of a multiplexer controlled by its neighbouring address qubits which are themselves shared. Beyond demonstrating the technical feasibility of this approach, the researchers presented its versatility by encoding diverse data types.
They moved beyond simple numerical values, successfully encoding grayscale and colour images, including 28×28 pixel images, using only hue and value information. This ability to handle varied data types highlights Tessa’s potential for applications beyond basic data storage, suggesting a pathway for quantum computers to process and manipulate complex information.
Hardware Implementation: Encoding Pixel Art and Sequences
The Tessa scheme successfully encoded pixel art images, including a “City-pop cat by M. This encoding, performed on IBM’s Phoenix quantum computer, required 8192 shots per basis for readout, a metric indicating the computational cost of retrieving the encoded information. The team also encoded single-basis images, achieving a correlation of 0.95-0.96 with the original input, and used 28×28 pixel images on Phoenix and 21×21 pixel images on Fez for testing. Experiments revealed a reduction in value error when employing dynamic-range recovery techniques, though this benefit diminished with pair encoding, a method using Z and X readouts to represent both value and hue. The correlation between the encoded value and the input decreased from 0.96 with single-basis encoding to 0.69-0.73 with pair encoding, indicating a trade-off between data complexity and accuracy. The performance of pair encoding differed between the two quantum computers used, with the hue error on Fez measuring 0.084 compared to 0.083 in local-noise simulation, suggesting a closer alignment between simulated and hardware results for that metric. While the value error in hardware was two to three times higher than in the local-noise simulation, the hue error remained comparable on Fez, indicating that the hardware’s performance was more consistent for hue representation. The remaining unused slots in the encoding were filled with zeros, a detail of the implementation that affects the overall data density and error characteristics. “The pair encoding is less accurate than the single encoding,” the researchers noted, acknowledging the inherent challenges of representing more information within the same quantum state.
Amplitude and Angle Encoding with Multiplexers
Tessa achieves a data capacity of 1568 encoded values from 120 qubits, exceeding the scale of previous experiments defining data encoding and decoding on a quantum processor. However, it checkerboards address and data qubits across a coupling graph with neighbouring data qubits sharing only a subset of address qubits. Tessa needs no ancilla qubits, because its parallelism comes from the geometry of the device.
Unlike those approaches, Tessa focuses on angle encoding, where each value dictates a rotation angle, and the number of shots sets the precision of the recovered value. The construction of Tessa’s encoding unitary, U, involves applying rotations, first , then multiplexers, across the quantum processor, doubling the circuit depth but avoiding the complexities of all-to-all hardware connectivity and the need for circuit transpilation to account for routing.
Data qubits and their neighbors form storing a single value, , mapped to an angle within each slot (d,i), with the total capacity, C, determined by the sum of across all data sets, D. This method restricts the polar angle to θ ∈ [c,π-c], where c = 0.2π minimizes error, as demonstrated through detailed analysis in the supporting appendices.
Data Encoding Precision and Scaled Fast Walsh-Hadamard Transform
This surpasses previous experiments in the number of values encoded and decoded under the same definition, using 120 qubits for the task. Beyond simple data storage, the team demonstrated the encoding of complex images, notably a “City-pop cat by M. This contrasts with the QCAT implementation, which used one value per pixel encoded in the Z basis. The underlying principle relies on an extension of the scaled fast Walsh-Hadamard transform (sfwht) and Gray-code permutation, originally developed for QPIXL. This transform maps values to angles, and setting values below a certain threshold allows for a lossy compression, effectively removing unnecessary rotation gates and simplifying the circuit. The researchers note in their paper that this process is analogous to the rotation applied in QPIXL, where each value is linearly mapped to an angle . Unlike encodings requiring ancilla qubits, Tessa’s geometry-based parallelism achieves this without additional qubit resources. The researchers emphasize that while other encodings can theoretically store more values, the need for transpiled circuits to account for routing often limits the practical amount of data that can be encoded and decoded.
👉 More information
🗞 Tessa: Encoding Classical Data Across an Entire Quantum Processor
✍️ Daniel Bultrini (Moth Quantum)
🧠 ArXiv: https://arxiv.org/abs/2611.18440v1




See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
