Gapless tunable intense terahertz pulse generation in strained diamond
Paper details
Section titled “Paper details”| Field | Value |
|---|---|
| Journal | Light Science & Applications |
| Published | 2026-03-31 |
| Authors | Yudan Su, Yuxuan Wei, Chaonan Lin, Li Huang, Jiaming Le |
| DOI | 10.1038/s41377-025-02092-6 |
Straining a 2 mm CVD diamond cube along its [001] axis made collinear phase matching of coherent Stokes Raman scattering possible. The source produced 30 nJ of 10 THz pulse energy per pulse. That output is about three times the 10 nJ measured with the earlier noncollinear geometry in a 0.5 mm diamond. Focusing the 60 fs pulse to a near diffraction limited spot gave a peak electric field of 2.3 MV/cm, and the demonstrated band ran from 5 to 12 THz.
The new THz gap and the CSRS mechanism
Section titled “The new THz gap and the CSRS mechanism”The 5 to 15 THz range coincides with the reststrahlen band of most nonlinear optical crystals, which the authors call the new THz gap. Sources in this range matter because they can drive narrow material resonances. A 10 THz pulse induces a transient superconducting state in K3C60 more easily than a mid infrared pulse. Resonant excitation of the 10.8 THz chiral phonon mode in rare-earth halides creates effective magnetic fields above 1 tesla.
The generation scheme is coherent Stokes Raman scattering. Two input pulses at ω1 and ω2 coherently and resonantly excite the Raman active 40 THz phonon of diamond. A third femtosecond mid infrared pulse at ωMIR beats with the coherent phonons to produce a THz pulse at ωTHz = ωMIR minus (ω1 minus ω2). The pump pulses are kept shorter than the phonon dephasing time of diamond so that the excitation stays transient.
Strain makes collinear phase matching possible
Section titled “Strain makes collinear phase matching possible”Collinear phase matching requires the mismatch vector to vanish, and in strain free diamond that vector is positive. Compressive stress along the [001] axis lowers the refractive index for light polarized along that axis. This shortens the wave vectors of the pump and the THz output until the mismatch is cancelled.
The calculation gives a strain of 0.14% for a phase matched frequency of 5 THz and 0.12% for 25 THz. Measured strains for phase matched generation at 5, 10, and 12 THz were close to those values. The measured output energy at 10 THz followed the predicted collinear phase matching curve. The authors also model strain inhomogeneity, which means the specimen has to hold a reasonably uniform stress for the phase matched frequency to stay on target.
In the earlier noncollinear work, phase matching in normal diamond was possible only with noncollinear geometry. That arrangement limits the interaction length and the output energy, and it forces the beam geometry to be readjusted whenever the THz frequency changes.
Setup: three input pulses and a strained cube
Section titled “Setup: three input pulses and a strained cube”A 33 fs, 5 W, 1 kHz Ti:sapphire laser pumped two optical parametric amplifiers seeded by a common white light source. The 50 fs signal and idler pulses at 6900 cm⁻¹ and 5600 cm⁻¹ from the first amplifier were chirp stretched to about 1 ps in dispersive ZnSe rods. The chirp rate was 7 THz/ps times 2π, so the time lag between the two pulses sets the detuning of the Raman excitation. A lag of 50 fs corresponds to a detuning of 0.35 THz. The mid infrared pulses came from a difference frequency generation stage after the second amplifier, ran 50 fs, and were tunable from 45 THz to 65 THz. All three beams were combined collinearly with a controllable delay.
The sample was a 2×2×2 mm³ CVD diamond cube sandwiched between two steel plates, with the compressive stress applied along the [001] axis.
| Pulse | Center frequency | Energy per pulse | Peak intensity on diamond | Beam area |
|---|---|---|---|---|
| ω1 signal | 6900 cm⁻¹ | 64 μJ | 110 GW/cm² | π(185 × 165)/2 μm² |
| ω2 idler | 5600 cm⁻¹ | 55 μJ | 55 GW/cm² | π(235 × 215)/2 μm² |
| ωMIR | 45 to 65 THz | 14 μJ | 0.7 TW/cm² | π(140 × 120)/2 μm² |
Output energy, efficiency, and field strength
Section titled “Output energy, efficiency, and field strength”At 10 THz the output reached 30 nJ per pulse, which corresponds to a MIR to THz energy conversion efficiency of 0.25% and a photon conversion efficiency of 1.25%. Extrapolating the measured scaling suggests the energy conversion efficiency can reach about 0.56% at 15 THz.
The output beam was near Gaussian in two orthogonal directions. A 2:1 parabolic mirror telescope focused it to a near diffraction limited spot of π(55 × 75)/2 μm² at the 1/e² radius. The focused 30 nJ, 60 fs, 10 THz pulse corresponds to a peak electric field of 2.3 MV/cm.
The output scaled linearly with the pump and mid infrared input energies, while the idler energy showed saturation like behaviour. Detuning the Raman excitation by 50 fs raised the output by 10%, because the weaker excitation spreads the phonon distribution over the length of the crystal. The optimum delay for the mid infrared pulse was 0.6 ps.
| Quantity | Collinear, strained 2 mm diamond | Noncollinear, 0.5 mm diamond |
|---|---|---|
| Diamond length | 2 mm | 0.5 mm |
| THz output at 10 THz | 30 nJ per pulse | 10 nJ per pulse |
| Peak electric field at focus | about 2.3 MV/cm | about 1 MV/cm |
Pump depletion and length scaling
Section titled “Pump depletion and length scaling”The ω1 pump was strongly depleted in the 2 mm sample. An infrared power meter measured 56% depletion of the pump energy at the highest input values. Integrating the measured intensity over time gave more than 78% depletion at the beam centre, with the pulse nearly fully consumed except for a small portion at the pulse front.
The coupled wave equations describe the behaviour. The phonon amplitude grows as the pump is depleted and the idler gains, reaching a maximum near z = 0.5 mm where the photon numbers of the two pulses become equal. Further along the crystal the Raman process runs in reverse, transferring energy back to the pump with a π phase shift, so the recovery of the pump pulse is only partial.
Because the depletion is strong, the depletion free estimate of (0.5/2)² = 1/16 for the 0.5 mm sample does not hold. The measured 0.5 mm output was only about three times lower than the 2 mm output.
For maximum output the phonon distribution should stay high and broad across the whole crystal. The calculation uses a pump intensity of 110 GW/cm², which is 80% of the 140 GW/cm² damage threshold of diamond. It gives an optimum idler intensity near 30 GW/cm² in the local calculation. The experiment instead found a broad maximum in output versus idler energy near 55 μJ, and the output would drop if the idler energy were raised further.
Extending the calculation to longer crystals gives an output ratio of roughly 1 to 3 to 6 and 9 for lengths of 0.5, 2, 5, and 10 mm. A 5 mm crystal with a beam overlap of π(150 × 150)/2 μm² would deliver about 60 nJ per pulse at 10 THz. The focused peak field would be 3.2 MV/cm. Length costs bandwidth: 5 mm gives a 14 THz bandwidth and a 31.5 fs output pulse, while 10 mm gives 7 THz and 63 fs. The authors judge lengths below 5 mm to be close to ideal.
Comparison with earlier schemes
Section titled “Comparison with earlier schemes”Two comparisons frame the result. The first is with the authors’ earlier noncollinear source, which delivered 10 nJ at 10 THz from a 0.5 mm diamond and a peak field of about 1 MV/cm at focus. The second is with difference frequency generation in GaSe, where the near infrared to THz energy conversion efficiency is 0.07%, against the 0.25% measured here and the 0.56% extrapolated at 15 THz.
Other routes into this band have their own limits. Organic crystals offer only a few narrow transparent windows above 5 THz and carry instability and cost problems. Air plasma and ferromagnetic film sources produce very broadband radiation but with low energy density at specific frequencies, which suits them poorly for driving narrow resonances.
Limits of the measurements and implications for diamond selection
Section titled “Limits of the measurements and implications for diamond selection”The polyethylene filter used to block the collinear mid infrared input was not effective, so the measured spectra were degraded, most visibly at the high frequency end of the 5 THz pulse where the signal is weak. The absorption cutoff of that filter above 15 THz also stopped the measurements at higher frequencies, so the part of the band above 12 THz is predicted rather than measured. The two pump inputs were limited by optical damage in diamond, and the mid infrared input was limited by the available pulse energy from the setup, at 0.7 TW/cm² against a 7 TW/cm² damage threshold. Numerical curves agreed fairly well with the data, with differences at the tail of the output pulse attributed to input parameters in the calculation deviating from the real ones.
For material selection, the source needs single-crystal CVD diamond that accepts a uniaxial compressive load and holds a uniform strain, since the strain sets the phase matched frequency within the narrow 0.12% to 0.14% window. Thickness matters, since 2 mm gave 30 nJ and the calculation puts 5 mm near the practical limit at about 60 nJ, while crystals thicker than 5 mm are not easy to obtain. The pump ceiling of 140 GW/cm², and with it the achievable output, is set by the crystal itself. On the system side, the collinear geometry reduces alignment complexity, tolerates beam misalignment better than noncollinear setups, and suits standard terahertz time domain spectroscopy.
Materials and support
Section titled “Materials and support”The paper’s material requirement is single-crystal CVD diamond, 2 mm thick for the demonstrated source and up to 10 mm for the crystals used in the length scaling study. The diamond materials catalog from 6CCVD lists single-crystal diamond grown by MPCVD with substrates up to 10 mm, which covers both the strained cube and the thicker crystals the output scaling depends on. Specimen geometry and strain fixture questions can be taken to engineering support.
View Original Abstract
We demonstrate collinear phase matching of transient coherent Stokes Raman scattering (CSRS) in strained diamond, enabling generation of intense, high-quality, femtosecond THz pulses over a gapless frequency band from 5 to beyond 15 THz. Our experimental results confirm this mechanism, supported by the theoretical calculations that agrees well with measurements. The theoretical model also provides a clear picture on how the input pulses should be adjusted to optimize the THz output from a diamond of given length.
Single Crystal Cvd Diamond Plate
View Specs →