IBM links two 120-qubit chips with around 99 percent transfer fidelity

Started by Mitchell93, Oct 30, 2026, 03:01 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: IBM links two 120-qubit chips with around 99 percent transfer fidelity   Views(Read 104 times)

Mitchell93

IBM has taken an important step towards modular quantum computers. ScienceAlert reports that IBM Research has linked two 120 qubit Nighthawk processors and moved quantum information between them with an average fidelity of about 99 percent

The chips were connected by four superconducting aluminium coaxial cables, each about 80 centimetres long, which IBM calls L-couplers. Every link reached at least 98.86 percent, and the best managed 99.29 percent. Gates acting across the two chips were less accurate, averaging around 94.75 percent according to The Quantum Insider's write up

The team also spread entanglement across both chips, including four and six qubit states. Raw fidelities were 77 and 68 percent, rising to about 91 percent after correcting for measurement errors. The links survived three warm up and cool down cycles without needing adjustment

The authors describe it as the first multichannel quantum interconnect between large superconducting processors, and it is a key building block for IBM's plan to build a fault tolerant machine by 2029. Jay Gambetta, director of IBM Research, said the cryogenic modules are speeding up progress. The work is a preprint with around 50 IBM authors and has not yet been peer reviewed

There are caveats. Magnetic fields during cool down degraded the cables, and the shielding used to fix it was a stopgap. Only qubits near the links were used, and some higher fidelity figures are model projections rather than measurements. Still, linking big chips reliably is exactly what modular quantum computing needs

How important are chip to chip links for scaling quantum computers? And is IBM's 2029 target still realistic?


RVD17

Eighty centimetre cables inside a dilution fridge is an engineering feat in its own right. Space and cooling are limited. Running four of them cleanly is no small thing

Nathan75

IBM keeps hitting its roadmap milestones fairly consistently. Starling in 2029 looked ambitious, but steps like this make it more believable. Peer review will tell us more
Normal is overrated

Yokozuna_Real

Surviving three thermal cycles without adjustment is underrated. Real machines get warmed up for maintenance all the time

Robust connections save huge amounts of recalibration. Fridges take days to cool down

Avoiding extra cycles saves weeks of work. Operators will appreciate that

Rob72

Ninety one percent after error correction on six qubit entangled states is decent. Raw numbers in the 60s and 70s are less so. Measurement errors remain a big issue for superconducting systems

Readout is a weak spot on most platforms. IBM knows that well

RandyOrton04

Modular designs are the only realistic way to reach millions of qubits. Nobody can build one giant chip that size

Links like this are the foundation of IBM's whole roadmap. Scaling by linking modules is how classical supercomputers grew too
Here more than I should be

NeuralSeer

Remote gates at about 95 percent are the weak point. State transfer is impressive, but computation across chips needs higher fidelity. Error correction would struggle at that level. That is the next number to watch. Ninety nine percent for gates would change everything
Have you tried turning it off and on again?