The reason nuclear clocks have been so hard to build lies in the sheer energy scale of nuclear physics. Almost all nuclear transitions sit at energies of kilo-electronvolts or higher, far beyond the reach of any conventional laser. Thorium-229 is the singular exception. Its lowest excited state, the isomer, lies astonishingly close to the ground state, so close that the transition falls in the vacuum ultraviolet at a wavelength of about 148.4 nanometres. That accident of nuclear structure makes thorium-229 the only known nucleus that can be addressed directly with laser light, opening the door to coherent quantum control of a nucleus and, crucially, to a clock in which the frequency reference is shifted from the electronic shell into the nucleus itself.

Decades of painstaking work established the existence and energy of this low-lying state, and recent years brought a cascade of breakthroughs: direct detection of the isomer, laser excitation of thorium-229 nuclei doped into calcium fluoride crystals, and precision frequency measurements using vacuum-ultraviolet frequency combs referenced to optical atomic clocks. Spectroscopy in crystalline hosts resolved the fine structure of the nuclear resonance, including quadrupole-split components, site-dependent shifts and temperature sensitivities. What remained missing was the final integration: a traceable, narrow-linewidth continuous-wave probe, a thorium-containing reference crystal, a fast and sensitive readout of the weak nuclear response, and a closed feedback loop that turns spectroscopy into timekeeping.

The new experiment supplies every one of those ingredients. The team generates continuous-wave 148.4-nanometre light by resonance-enhanced four-wave mixing in cadmium vapour. Two 375-nanometre photons drive a two-photon resonance in cadmium atoms, and a 710-nanometre photon completes the sum-frequency process, producing 10 microwatts of vacuum-ultraviolet radiation with an estimated linewidth below one hertz. The 375-nanometre light itself comes from second-harmonic generation of a fibre laser in a bow-tie enhancement cavity, while both fundamental lasers are phase-locked to a self-referenced erbium-fibre frequency comb. That comb, in turn, is anchored to a cryogenic-silicon-cavity-stabilized laser at 1397 nanometres and counted against a hydrogen maser, giving the vacuum-ultraviolet frequency absolute traceability.

Equally impressive is the crystal work. Thorium-229 is severely scarce, and the team grew their best crystal, dubbed TS1, from a solution containing only 1.4 micrograms, or 10 kilobecquerels, of the isotope, leaving essentially no room for iterative optimization. Using a temperature-gradient technique in which crystallization is driven by slowly cooling the heater rather than moving the crucible, they produced a bubble-free cylinder 1.09 millimetres in diameter and 4.94 millimetres long, with a thorium concentration of 2.2 × 10¹⁷ per cubic centimetre and a vacuum-ultraviolet transmission of 7 percent. In fluorescence and absorption spectroscopy, the crystal revealed a broad quadrupole-unresolved structure alongside a quadrupole-resolved manifold of five narrow transitions, including the line, known as line b, that serves as the clock reference. The excited-state lifetime measured 631 seconds, an extraordinarily long memory for a nuclear excitation.

Turning that spectroscopy into a clock required a clever readout. Because the on-resonance fractional absorption of the light is tiny, roughly 10⁻⁴, the team modulated the laser frequency sinusoidally and demodulated the transmitted signal with a lock-in amplifier, producing a dispersive error signal whose zero crossing marks the line centre. Detector choice proved decisive. A photomultiplier tube saturated at a few nanowatts of transmitted light, but a simple phototube operated in direct-photocurrent mode tolerated far more power. Combined with the higher-transmission TS1 crystal, the phototube readout boosted the detected photoelectron rate roughly thirtyfold and raised the one-second absorption signal-to-noise ratio about twentyfold compared with the initial configuration.

With the discriminator in hand, closing the loop was straightforward in principle: each cycle, the measured detuning is fed back through an acousto-optic modulator in the comb-referenced optical chain, steering the vacuum-ultraviolet frequency onto the nuclear resonance. The performance improved step by step as the detection scheme and crystal were upgraded, with fitted instabilities falling from 1 × 10⁻¹¹ per root second to 6 × 10⁻¹² and then 2 × 10⁻¹². After further optimization, raising the transmitted power to 350 nanowatts, increasing the measurement duty cycle to about 85 percent, and using a near-optimal modulation amplitude of 12 kilohertz, the clock reached its final instability of 5 × 10⁻¹³ per root second, averaging down toward the 10⁻¹⁵ level. That figure sits within a factor of two of the photon-shot-noise limit calculated from the measured discriminator parameters, meaning the clock is already operating close to the fundamental floor imposed by the quantum statistics of the detected light itself.

Just as important as the instability is the reproducibility of the clock transition. The team compared the centre frequency of line b in two independently fabricated crystals and found agreement at the 10⁻¹³ level, a fractional difference of 2.8 × 10⁻¹³. They then compared their measurements with previous vacuum-ultraviolet-comb data from JILA on entirely different crystals, evaluating the published temperature-dependent fit at their own crystal temperature of 301.4 kelvin. The offsets, −0.50 kilohertz and +0.06 kilohertz respectively, are both consistent with zero within the combined uncertainties. Systematic shifts were kept firmly under control: temperature fluctuations of 3 millikelvin correspond to frequency fluctuations of only about 6 hertz, and residual magnetic fields, suppressed by multilayer shielding, contribute less than 1 hertz of Zeeman shift.

A near-simultaneous independent report has demonstrated thorium-229 clock operation using a different vacuum-ultraviolet source based on randomly quasi-phase-matched strontium tetraborate, underscoring how rapidly the field is converging. The implications stretch well beyond metrology. Because the nucleus is shielded by the electron cloud, nuclear transition frequencies are predicted to be extraordinarily sensitive probes of variations in fundamental constants such as the fine-structure constant, making thorium-229 clocks potential detectors of new physics. And because a solid-state host packs a macroscopic number of nuclei into a crystal the size of a grain of rice, the platform points toward compact, robust, field-deployable frequency references and nuclear quantum sensors that no ion-trap clock could match.

There is still a long road to clocks that rival the best optical atomic standards, which already reach instabilities and accuracies in the 10⁻¹⁸ regime. The present device averages with a 10⁻¹³-level instability, limited by the weak absorption of the nuclear transition and the available vacuum-ultraviolet power. But the conceptual barrier has fallen: laser-addressed nuclei are now operational clock references, not proposals. The thorium-229 nucleus, hidden inside its calcium fluoride host, has joined the ranks of pendulums, quartz crystals and caesium atoms as a keeper of time, and with it comes a new window onto the deepest constants of nature.

Subject of Research: A solid-state nuclear clock based on the laser-locked 229Th isomeric transition

Article Title: A nuclear clock synchronized to 229Th

Article References: Huang, B., Yan, G., Xiao, Q., Bu, W., Zhao, C., Zhang, Z., Yan, C., Chen, Z.-A., Zhang, P., Penyazkov, G., Zhan, Z., Yan, L., Wang, Y., Li, L., Li, S., Jiang, D., Qian, X., Liu, X., He, Q., … Ding, S. (2026). A nuclear clock synchronized to 229Th. Nature. https://doi.org/10.1038/s41586-026-11122-1

Image Credits: AI Generated

DOI: 10.1038/s41586-026-11122-1

Keywords: nuclear clock, thorium-229, vacuum ultraviolet laser, frequency metrology, CaF2 crystal, isomeric transition, atomic clocks, precision spectroscopy, fundamental constants, four-wave mixing, quantum sensing, nuclear

The work, carried out by Abhishek Kumar Pandey, Deepak Jain, and Catherine Baskiotis, exploits spontaneous four-wave mixing, a nonlinear optical process in which two pump photons at the same frequency are annihilated to create a correlated pair of photons at two very different frequencies. Energy conservation forces the sum of the two new frequencies to equal twice the pump frequency, while momentum conservation, known as the phase-matching condition, dictates which frequency combinations can actually occur. In a waveguide, the phase-matching condition can be engineered by shaping the geometry of the silicon core, allowing the generated signal and idler photons to be tuned with remarkable precision across a wide spectral range.

What makes the new designs remarkable is the sheer distance between the two photons. The flagship design, targeted at atmospheric quantum key distribution, produces a signal photon at 3.905 micrometres, deep inside a mid-infrared atmospheric transparency window, while its idler partner lands at 1.541 micrometres in the telecom C-band. That separation of roughly 2,364 nanometres is dramatically larger than the record of about 1,125 nanometres theoretically reported in earlier all-solid silicon waveguide studies. Crucially, the design achieves this in the true quantum regime, where at most one photon pair is produced per pump pulse, rather than relying on the high pump powers used in classical demonstrations.

This distinction matters. Previous experiments with air-clad silicon waveguides demonstrated large signal-idler separations of about 2,058 nanometres, but only by using pump peak powers around 20 watts to achieve phase matching. In the quantum regime, pump power must be kept low to suppress multi-pair emission, and with those earlier designs the phase-matched separation collapsed to 1,780 nanometres, placing the mid-infrared photon outside any atmospheric transmission window. The new all-solid designs sidestep this problem by building the phase matching into the waveguide geometry itself, so that it holds even at the milliwatt pump levels required for single-pair emission.

The choice of an all-solid, fully oxide-clad structure is itself strategic. Air-clad waveguides suffer from significant scattering losses caused by sidewall roughness, because the optical mode interacts strongly with the etched surfaces. By surrounding the silicon core with fused silica cladding and a buried oxide layer, the proposed waveguides reduce the modal overlap with etched sidewalls, mitigating fabrication-induced scattering. The team’s calculations show that even at the longest signal wavelength, where silica is nominally opaque, the majority of the optical power remains confined to the silicon core, keeping total material attenuation over the 2-centimetre waveguide to acceptable levels.

To give their predictions quantitative teeth, the researchers adopted a model for the probability of photon-pair generation per pulse and validated it against published experimental data from photon-pair generation in photonic crystal fibre. By modelling the fibre as a simple step-index structure and feeding in the experimental parameters, they computed photon-pair rates that closely matched the measured values across a range of pump powers, outperforming the original theoretical treatment of those experiments. Because the model takes as inputs only the modal properties, the nonlinear index, and the operating conditions, the authors argue that the validation is generic and applies equally to the silicon waveguide geometries, whose modal fields they computed independently using full-vectorial finite element simulations.

Each of the three proposed waveguides is tailored to a specific application. Two designs generate signal photons at 3.265 and 3.461 micrometres, which coincide with absorption bands of methane and nitrogen dioxide respectively, the fingerprint region where these gases reveal their presence. Paired with telecom-band idlers, these sources could power schemes for sensing with undetected light, in which the mid-infrared photon probes the gas while only the telecom photon is ever detected, allowing the entire measurement to be performed with cheap, mature silicon-compatible detectors. The third design, optimised for atmospheric quantum key distribution, places the signal photon at 3.905 micrometres, where atmospheric transmission is dramatically better than at the 1.55-micrometre wavelength used in current daylight quantum communication experiments.

The shift to the mid-infrared could be transformative for free-space quantum communication. At 3 to 4 micrometres, the atmosphere exhibits reduced Rayleigh scattering, lower solar background noise, and greater resistance to weather-related variations, all of which translate into more reliable quantum links and an improved ability to operate in daylight. Shrinking the photon-pair source onto a chip also addresses a critical constraint for satellite payloads, where every gram of mass and cubic centimetre of volume counts. Silicon-on-insulator fabrication is cheap, scalable, and compatible with existing CMOS foundries, making the prospect of flight-qualified quantum sources considerably more realistic than with bulk nonlinear crystals.

The team also addressed the practical obstacles that have plagued mid-infrared quantum optics in silicon. Spontaneous Raman scattering, which can contaminate photon-pair spectra, is predicted to be negligible because the generated photons lie more than 330 Raman linewidths away from the silicon Raman gain peak, placing the Raman gain at the signal frequencies at the level of one part in a million. Two-photon absorption, the bane of silicon nonlinear optics at telecom wavelengths, falls off sharply for pump wavelengths above 1.9 micrometres, and the chosen pump wavelengths between 2.1 and 2.21 micrometres sit in a sweet spot combining strong Kerr nonlinearity with suppressed carrier generation. Numerical tolerance studies further showed that small fabrication imperfections of plus or minus 10 nanometres in the core dimensions cause only slight wavelength drifts that can be compensated by tuning the pump laser.

The predicted operating points are within reach of existing technology. Each design targets a probability of photon-pair generation per pulse of about 0.05, a value considered a practical compromise between source brightness and multi-pair suppression, and achieves this with peak pump powers between 9.2 and 32.2 milliwatts from 5-picosecond pulses. Such pump powers are already delivered by mature thulium and holmium-doped solid-state lasers around 2 micrometres, while integrated 2-micrometre laser sources continue to advance toward the required output levels. With mid-infrared single-photon detectors, including superconducting nanowire devices and frequency-upconversion schemes, progressing rapidly, the authors argue that their chip-scale sources could soon move from simulation to experiment, opening a path toward quantum communication and sensing that spans from the fingerprint region of the molecules to the heart of the telecom infrastructure.

Subject of Research: Silicon waveguide design for quantum-correlated photon-pair generation bridging mid-infrared and telecom bands via spontaneous four-wave mixing

Article Title: Quantitative study of silicon waveguides for the generation of quantum correlated photon pairs bridging mid-infrared and telecom bands

Article References: Pandey, A. K., Jain, D., & Baskiotis, C. (2026). Quantitative study of silicon waveguides for the generation of quantum correlated photon pairs bridging mid-infrared and telecom bands. Results in Optics, 25, Article 101158. https://doi.org/10.1016/j.rio.2026.101158

Image Credits: AI Generated

DOI: 10.1016/j.rio.2026.101158

Keywords: silicon photonics, quantum optics, photon pairs, four-wave mixing, mid-infrared, telecom C-band, quantum key distribution, gas sensing, silicon-on-insulator, phase matching, Raman scattering, integrated photonics