As recently as 2023, the best spin-qubit machine consisted of just two computing units that botched roughly every 25th operation in certain measurements. Three years on, two research groups independently report error rates of around 0.2 percent — and considerably larger chips. The two papers published in Nature in late July bring an underdog back into the race for a useful quantum computer.
Spin qubits store information in the intrinsic angular momentum of a single electron trapped in a tiny potential well etched into silicon. The appeal of the concept, formulated by Daniel Loss and David DiVincenzo in the late 1990s, lies in manufacturing: build qubits from the same materials and processes as ordinary silicon chips, and one day you might replicate them a million times over. In practice, the platform lagged behind superconducting circuits and trapped atoms for years.
Two bottlenecks, two answers
The new work targets exactly the two problems that have blocked scaling: how do you connect qubits that do not sit right next to each other, and how do you control very many of them without drowning in wiring?
The team around Lieven Vandersypen at TU Delft solved the first problem through movement. It shuttled a single qubit about 1.2 micrometres across the chip so that it could interact in turn with four stationary neighbours. That made possible the parity measurements that quantum error correction requires — and a shared entanglement of all five qubits.
HRL Laboratories in Malibu, California, went after control instead. Their chip carries 54 quantum dots and operated 18 qubits; the necessary signals no longer come from racks of electronics at room temperature but from a custom CMOS component sitting inside the cryostat at around minus 268 degrees Celsius. The two are joined by a superconducting ribbon cable roughly a centimetre wide that bundles hundreds of control lines while carrying almost no heat to the qubits. The system ran error detection and error correction on its own and undercut previous control errors for this type of qubit by about a factor of ten.
For comparison: superconducting quantum computers have long worked with more than 100 qubits, and neutral-atom machines with thousands. On the quality of individual operations, though, spin qubits have now caught up. Other groups are following: Japan's RIKEN institute reported a five-qubit system with an error rate below 0.01 percent in early July, and the Delft start-up Groove Quantum presented an 18-qubit germanium processor in April, albeit without completed peer review.
The progress is "great" for the field, says Daniel Loss, now at the King Fahd University of Petroleum and Minerals. A reviewer of the HRL study called it a significant milestone for the technological maturity of semiconductor-based spin qubits. The commercial shift is visible too: IBM, which backs superconducting qubits itself, announced its acquisition of HRL.