Italy's Planckian tries a different chip wiring, with a US firm modeling the noise first

On July 14, Los Angeles-based Quantum Elements signed a development agreement with Italy's Planckian to build a noise model for the Italian hardware firm's shared-control-line quantum processor, letting the team evaluate error-correction (QEC) choices in software before fabrication.

Planckian's design uses shared, or "global," control lines, in which a single set of wires addresses many qubits through a closed-loop geometry and blockade-style ZZ coupling, rather than running a dedicated control wire to every qubit as conventional superconducting processors do. The trade is a different error landscape: errors that would look independent in a standard chip can become correlated across qubits that share a control path, and a decoder that assumes local errors can fail quietly.

A conventional superconducting processor's wiring count scales with its qubit count, and the wall shows up early. Industry coverage of the global-control approach puts the bottleneck at roughly two million wires for a one-million-qubit conventional machine, with about 80% of the cost of a roughly $5 million cryostat for a 150-qubit coaxial-wired prototype attributable to the wiring itself. Shared-control architectures try to break that wall by collapsing many per-qubit wires into a small, addressable set. They also change the optimization problem: instead of asking how to fit more cables into a dilution refrigerator, Planckian is asking what its correlated errors look like and how a QEC scheme should be redesigned around them.

Quantum Elements and AWS published the prior result the partnership builds on: a Digital Twins framework that modeled a 97-physical-qubit (49 data plus 48 ancilla) distance-7 rotated surface code in roughly an hour on a single compute node, using Amazon EC2 Hpc7a instances and a quantum Monte Carlo (QMC) algorithm from arXiv:2502.18929. A direct density-matrix simulation of the same circuit would need 4^97 entries, which is not computable on any classical machine that exists. The practical effect is that a hardware team can sweep over error-correction parameters and decoder choices in software before paying for a fabrication run.

The Planckian-specific twist is that the new noise model will not be the USC/Harvard/AWS model. The 97-qubit prior work used a conventional noise profile: T1 and T2 coherence times of 150 to 300 microseconds, residual ZZ crosstalk of 20 to 100 kilohertz, 25-nanosecond Gaussian single-qubit pulses, and 50-nanosecond Rzz two-qubit gates. Shared-control wiring is expected to push the model in a different direction, toward correlated and leakage-style errors that the prior parameter set does not capture. The partnership's first deliverable is the parameter set and code that does.

The deal, as of July 14, is a development agreement, with no published Planckian-specific noise model or decoder benchmark in the source basis. The architectural bet is real, the simulation method is real, and the two firms have prior co-development track records on their respective sides, including Planckian's earlier co-development work with Italy's QTLab on scalable superconducting designs. The partnership adds the simulation layer to the fault-tolerance supply chain, the layer that lets a hardware team choose its error-correction scheme with the same seriousness that a chip foundry chooses its photomask.

The next concrete thing to watch is whether Quantum Elements and Planckian publish a Planckian-specific noise profile and a decoder benchmark on a surface-code instance, and how soon. The 97-qubit prior result set a throughput a shared-control model would have to match or exceed to be useful as a design tool rather than a marketing artifact.