Researchers at HRL Laboratories reported in a peer-reviewed Nature paper on July 29, 2026, that they had run error-correction routines on silicon spin qubits with control electronics mounted inside the refrigerator that cools the qubits. The authors wrote that once they had started a program, the processor needed no further communication with electronics at room temperature.
Six days earlier, IBM announced a definitive agreement to acquire HRL, a research laboratory in Malibu, California, then owned jointly by Boeing and General Motors. The HRL team had first posted the work as a preprint in April.
HRL’s processor ran those routines without real-time help from outside the refrigerator, on up to seven of its 18 physical qubits, using codes that correct only one type of error or that detect errors without correcting them.
The processor
The qubit chip has 54 quantum dots, tiny regions of silicon that each trap one electron, arranged in three rows. The team can configure the dots as up to 18 exchange-only qubits, each storing information in the combined spin of three electrons, one in each of three neighboring dots. Spin is a magnetic property of electrons.
Researchers control exchange-only qubits with voltage pulses alone, without microwave signals. HRL designed a controller chip to generate those pulses inside the refrigerator.
The controller has about 70 million transistors and was made in a commercial 130-nanometer CMOS process, the standard technology for computer chips. It operates at 4 kelvin, about minus 452 degrees Fahrenheit, and draws 3.5 watts or less in typical operation, according to the preprint.
A superconducting ribbon cable of niobium on polyimide, with 296 signal lines, connects the controller to the qubit chip, which is mounted on the refrigerator’s coldest stage, at millikelvin temperatures. The authors reported that the cable conducts less than 10 microwatts of heat to that coldest stage of the refrigerator.
Nature featured the system on the cover of its July 30 issue. In the cover summary, the journal’s editors wrote that control electronics at room temperature need wiring across a difference of about 300 kelvin. They wrote that the wiring limits how large cryogenic quantum systems can grow.
Nature published a paper in the same issue from QuTech researchers in Delft. They moved one spin qubit along a chip to run parity checks, a type of error check, with four stationary qubits. Quantum Observer compared earlier silicon spin-qubit results, including QuTech’s, in March.
What the team measured
HRL said its control errors were 10 times lower than in any earlier demonstration with exchange-only qubits. The lowest reproducible error the team reported for a CNOT gate, a basic operation between two qubits, was 9 in 10,000. In the Nature paper, the team reported mean errors of 1.7 in 10,000 for single-qubit gates and 35 in 10,000 for CNOT gates.
In the largest test, the team ran a distance-5 repetition code on seven physical qubits for up to 200 rounds of error checks. In a repetition code, researchers spread one bit of information across several qubits and check repeatedly for flips of that bit. A code’s distance is a measure of how many such errors it can withstand.
The authors reported that the logical error rate, how often the encoded bit was corrupted, fell by a factor of 4.7 when they increased the code distance from 3 to 5.
In a second test, the team used six physical qubits to run a [[4,2,2]] code, which detects errors without correcting them. The code stores two logical qubits, units of quantum information protected by the code, across four physical qubits. Fidelity is a measure of how closely a result matches the intended one, on a scale where 1 is a perfect match.
After the researchers discarded the roughly 77% of runs in which the checks flagged an error, a step called post-selection, they measured a logical fidelity of 0.95. With every run kept, the logical fidelity was 0.59.
The authors attributed most remaining errors to miscalibrated control pulses, caused by imperfect signal generation and transmission, and to static magnetic field gradients. They estimated in the preprint that noise intrinsic to the qubit device contributed only about 0.02% to the absolute CNOT error.
The authors listed work still needed to scale the design, including a manufacturable interconnect, lower-power controllers, and more uniform devices.
IBM’s purchase
IBM said HRL’s “advanced expertise in silicon-spin qubit engineering will complement and extend IBM’s long-term mission.” IBM builds its own processors from superconducting qubits, electrical circuits that store quantum information when cooled close to absolute zero.
IBM also said it saw “an opportunity to partner even more closely with Anderon, including potential plans to develop spin qubit manufacturing,” through the acquisition. Anderon is the quantum wafer foundry that IBM announced with the Commerce Department in May.
IBM did not disclose a price. It said it expected the deal to close by the end of the third quarter of 2026, subject to regulatory approvals. Boeing and General Motors would continue to partner with IBM on quantum applications, the company said.
In an interview with Reuters, Jay Gambetta, director of IBM Research, said HRL would begin making chips at IBM’s facility in New York in the nearer term. He said smaller electron-spin circuits would become useful to IBM’s efforts after the company delivers Blue Jay, a planned superconducting system, in 2033.
Update, November 12, 2026: On August 26, IBM said it had completed its acquisition of HRL. It did not disclose financial terms.
