ETH Zurich's chip routes quantum states into mechanical vibrations instead of electromagnetic cavities, recreating the CPU/RAM split in quantum hardware and easing the on chip layout bottleneck.
ETH Zurich has rebuilt the oldest trick in classical computing, separating the processor from memory, inside a quantum chip. The move is not a qubit upgrade but a memory swap: quantum states normally stored in electromagnetic cavities are now offloaded into mechanical vibrations, packing more memory cells into a smaller area than electromagnetic cavities do.
Adding one more superconducting qubit to a chip costs more floor space than the last. The electromagnetic cavities that store and route quantum states consume an outsized share of the surface area inside the dilution refrigerator, the cryogenic enclosure that keeps a quantum processor near absolute zero, and the wiring that links qubits scales faster than the qubits themselves. Sharper individual qubits will not change that. A different shape of memory will.
The ETH Hybrid Quantum Systems Group, led by Yiwen Chu, builds that different shape on a single 7.5 mm by 2.5 mm chip. Superconducting circuits stay on top as the processing layer; a grid of mechanical resonators underneath rings at distinct acoustic frequencies, each a separate store for a quantum state handed down by its qubit partner. The group's own July 2026 central release likens the arrangement to a CPU talking to working memory, a metaphor rather than a product claim, and an earlier D-PHYS item describes the resonators as guitar strings vibrating in the dark, each storing a slice of quantum information in its tone.
The mechanism is not new in theory. Chu's group and others have spent years on hybrid systems that marry superconducting circuits to mechanical oscillators, and phonon-based quantum memory traces back to circuit-QED acoustics from the 2010s. What is new here is the integration: a chip designed from the start around the processor-memory split rather than around individual qubits. The acoustic layer is meant to pack more memory cells into a smaller footprint, lengthen the time stored states survive, and free surface area for the qubits and control lines that still need to live there.
The Hybrid Quantum Systems lab describes a broader program of building processors around acoustic modes. Independent pickups by Quantum Computing Report and an HPCwire wire summary corroborate the headline mechanism without adding primary numbers, leaving the arXiv preprint and the forthcoming Science paper as the underlying source.
The limits are deliberate. The work shows storage, retrieval, and coherent swapping between the two layers, not computation beyond those memory operations. Acoustic damping still bounds how long a phonon-encoded state survives before decoherence sets in, an interval the preprint reports as an open optimization target rather than a finished specification. Treating this as "quantum RAM" in a commercial sense would overstate the result; the work is a memory primitive, not a memory product.
Mechanical memory cells packed beside the qubit layer will host more useful processors in the next generation of dilution refrigerators, before the wiring runs out of room, which is why a single-chip demonstration still shifts the design choices the field considers. Classical computing took decades to settle on a memory hierarchy; quantum hardware is now working through the same trade for the first time, with acoustic storage as one of the early options. The next milestone to watch is whether Chu's group, or competing hybrid teams, can chain several acoustic memory cells into a coherent working store, since that step would turn a single vibrating chip into a quantum machine's working memory.