Industrial silicon growth delivers purified layers and working spin qubits

Why the isotope matters

An electron confined in a tiny silicon region called a quantum dot can carry quantum information in its spin. Maintaining that information requires controlling disturbances from its surroundings. One source is silicon-29: its nucleus has a spin that interacts magnetically with the electron, disturbing the phase relationships needed for quantum operations.

Natural silicon contains about 4.7% silicon-29. Enriching the material in silicon-28 reduces this nuclear-spin disturbance. The practical question is whether industrial growth equipment can produce suitable material that also survives device fabrication.

In an experimental preprint, Zhenzhen Kong and colleagues report purified silicon layers grown by industrial reduced-pressure chemical vapor deposition, or RPCVD. The supplied evidence includes the paper text, not just its abstract. The findings should be treated as preprint results rather than established, independently replicated performance.

Growing and measuring a purified layer

The researchers used an already isotope-enriched silicon-containing gas to grow crystalline layers on high-resistivity silicon substrates at 650°C. The reactor therefore deposits purified material; it does not itself perform the initial isotope separation.

One structure was an approximately 83-nanometre-thick silicon-28 layer. Secondary-ion mass spectrometry, which measures composition while sputtering through a surface, found 99.997% silicon-28, about 29 parts per million silicon-29 and about one part per million silicon-30. For context, 29 parts per million means roughly 29 silicon-29 atoms per million silicon atoms, compared with about 47,000 in natural silicon.

These are nominal, calibrated measurements. The authors note that relative measurement uncertainty increases near the instrument’s detection limit. Atomic-force microscopy measured surface roughness of about 0.07 nanometres, while electron microscopy showed an ordered crystal lattice in the sampled regions. Those observations do not establish that every part of a wafer is defect-free.

Heat can erase the engineered structure

The team separately grew a test stack containing three 10-nanometre enriched layers alternating with two natural-silicon layers. This made isotope redistribution easier to track during vacuum heating. Its buried enriched layers were less pure than the separate 83-nanometre layer, so the two structures should not be conflated.

After five minutes at 850°C, the stack retained clear isotope layering and a smooth surface. At 950°C, small pits appeared and isotope interfaces began to broaden. At 1050°C, roughness increased substantially and the layers mixed more strongly. Heating to 1150°C produced severe surface breakup within 15–30 seconds.

The authors interpret the progression as native surface oxide decomposing at defects, followed by enhanced atomic movement and eventual breakup into islands. Microscopy and composition profiles document the structural changes; the detailed vacancy and surface-transport account is their mechanistic interpretation.

The results support a processing window below roughly 950°C under the tested vacuum conditions. They do not establish a universal temperature limit independent of heating duration, surface protection or fabrication environment.

Working qubits, with distinct performance measures

Electrical transport measurements gave a peak Hall mobility of about 12,858 square centimetres per volt-second. Mobility describes how readily electrons move under an electric field. A separate scattering analysis yielded a quantum scattering lifetime of 0.659 picoseconds. That is an electronic transport quantity, not the lifetime of stored quantum information.

For spin control, the researchers operated a double-quantum-dot device at about 20 millikelvin. Ramsey measurements gave a dephasing time of 616.6 ± 24.5 nanoseconds. A Hahn-echo sequence, which reverses some effects of slowly varying disturbances, extended coherence to 28.8 ± 0.9 microseconds—roughly 47 times longer. These uncertainties are reported fit standard errors, not measures of device-to-device reproducibility.

Randomized benchmarking returned a reference Clifford-operation fidelity of 99.70%. Interleaved measurements estimated 99.91% and 99.90% for individual X and Y rotation gates. A fidelity near 99.9% corresponds to an average error near one part in a thousand under that protocol, not a guarantee that an entire computation succeeds with that probability.

What remains unproven

The study connects industrial deposition with functioning single-qubit control. It does not isolate how much isotope purification alone improved performance, demonstrate a two-qubit logic gate or establish fault-tolerant computation. Wafer-wide uniformity, thickness control and integration with 300-millimetre manufacturing substrates remain to be demonstrated. Those are essential distinctions between a promising materials process and a reproducible manufacturing platform.