RESOURCES

Diamond Optical Windows: The Last Material Standing in High-Power Laser and Extreme-Environment Optics

Above a few kilowatts, the first-order criterion for a window is not its absorption coefficient but how much optical path distortion each deposited watt produces. CVD diamond absorbs one to two orders of magnitude more than ZnSe at 10.6 µm and still wins, because thermal lensing scales with (dn/dT)/κ — where diamond leads by nearly three orders of magnitude. This article assembles verifiable literature and vendor data on diamond's intrinsic optical parameters, a quantitative comparison against ZnSe, Ge, GaAs and sapphire, real engineering cases from ITER gyrotron windows to synchrotron beamline foils, and a tolerance-and-acceptance checklist that can go straight into a purchase specification.

diamond optical window2026-07 · 10 min

01Bottom line first: diamond wins on heat removal, not on absorption

The most common error in high-power window selection is treating the absorption coefficient as the whole story. At 10.6 µm, ZnSe absorbs at just 0.0005 cm⁻¹ (Crystran data sheet), while CVD diamond is specified at <0.07 cm⁻¹ across 8–12 µm, with measured single-crystal and polycrystalline samples typically at 0.02–0.05 cm⁻¹ (Balmer et al., 2009). Diamond therefore absorbs one to two orders of magnitude more bulk power — and still wins. The controlling figure is (dn/dT)/κ: diamond, at 9.6×10⁻⁶/°C and 2600 W/m·K (273 K), gives roughly 3.7×10⁻⁹; ZnSe, at +61×10⁻⁶/°C and 18 W/m·K, gives roughly 3.4×10⁻⁶ — close to three orders of magnitude apart on the same two data sheets. Diamond does not leak less heat into the beam path; it refuses to hold a temperature gradient long enough to bend the wavefront. Any window trade study that stops at the transmission curve is answering the wrong question.

02Intrinsic optics: one continuous corridor from 226 nm to microwave

Diamond's 5.45–5.47 eV band gap puts the intrinsic absorption edge near 226 nm, and transparency then runs unbroken into the far infrared — measured out to 500 µm — and onward through radar, microwave and THz. The only intrinsic gap is the multiphonon band between 2.5 and 6.5 µm, and the physics there is worth understanding: diamond's single-phonon mode sits at 1332.5 cm⁻¹ (7.5 µm), but cubic lattice symmetry forbids first-order absorption at that frequency, so infrared loss arises only from two- and three-phonon processes, peaking near 5 µm at roughly 12 cm⁻¹. The refractive index falls from 2.60 at 266 nm to 2.46 at 405 nm and 2.4175 at 589 nm, settling near 2.38 across 8–14 µm. Laser calorimetry at 1064 nm gives 0.003–0.07 cm⁻¹ for single-crystal CVD material; Webster et al. (JOSA B, 2015; samples at 10–40 ppb nitrogen) show that absorption from the long-wavelength side of the mid-UV into the visible is dominated by nitrogen and nitrogen-related defects — which is precisely why visible and near-IR applications must specify nitrogen, not just thickness. Thermal conductivity is 2000–2600 W/m·K at room temperature and still 730 W/m·K at 500 °C.

03Versus ZnSe, Ge, GaAs and sapphire: where the orders of magnitude appear

Running (dn/dT)/κ across one consistent set of vendor data sheets: diamond ≈3.7×10⁻⁹; sapphire, 13.1×10⁻⁶ ÷ 27.21 ≈ 4.8×10⁻⁷ (≈130×); GaAs, 147×10⁻⁶ ÷ 48 ≈ 3.1×10⁻⁶ (≈830×); ZnSe ≈3.4×10⁻⁶ (≈900×); germanium worst at 396×10⁻⁶ ÷ 58.62 ≈ 6.8×10⁻⁶ (≈1800×). The thermal-stress side tracks the same way: diamond expands at 1.0×10⁻⁶/°C against 7.1 for ZnSe, 6.1 for Ge, 5.7 for GaAs and 5.6/5.0 for sapphire, leaving diamond roughly three orders of magnitude ahead on κ/α. Germanium carries an additional hard ceiling — a little above 350 K the band gap floods with thermal electrons and the material goes opaque at all wavelengths, which alone disqualifies it wherever aerodynamic heating is in play. Balmer's review reproduces Kemp's thermal-lensing figure of merit at 1 µm: diamond 1440 versus YVO₄ 20, KGW 3 and Ba(NO₃)₂ 1 (diamond evaluated over an 8 mm path, the others over 25 mm). On damage threshold, Element Six reports its sub-wavelength-structured diamond windows surviving about 2.5 MW/cm² where comparison AR-coated windows failed near 0.25 MW/cm².

04Case one: gyrotron and laser exit windows, where nuclear codes set the spec

Fusion heating is the most demanding productionized use of diamond windows. ITER's electron cyclotron system — one equatorial launcher plus four upper launchers — is designed to inject up to 20 MW at 170 GHz. Each window unit is a polycrystalline disc grown by microwave-plasma-assisted CVD, Ø80 mm × 1.11 mm, the thickness chosen to satisfy the half-wavelength resonance condition at 170 GHz for maximum transmission. It is simultaneously part of ITER's first confinement system (nuclear facility INB No. 174) with a tritium confinement function, classified SIC1/PIC1. KIT's qualification campaign produced two directly citable data sets: five Ø30 mm × 1.111 mm specimens measured 5.2–7.8×10⁻⁶ loss tangent in a Fabry–Perot resonator, and ring-on-ring testing (DIN 51105) gave a mean flexural strength of 314.28 MPa with a standard deviation of ±16.19 MPa (individual values 295.55–332.73 MPa); an Ø80 mm disc brazed to copper cuffs was pressure-tested to 2 bar. Earlier work on a Ø100 mm polycrystalline disc reported tanδ as low as (0.6±0.5)×10⁻⁵ at 145 GHz. On the laser side the same material has been pushed hard: Antipov et al. (Optics Letters, 2019) reached 1.2 kW from an external-cavity diamond Raman laser at 83% slope and 53% optical-to-optical efficiency, with M² improving from 2.95 to 1.25 as power rose.

05Case two: synchrotron beamline windows, IR domes, and the diamond anvil cell

On synchrotron and FEL beamlines, diamond is displacing beryllium for reasons that are half thermal and half occupational-health. Diamond Materials publishes the head-to-head: atomic number 6 versus 4; hardness 12,000–15,000 versus 150–200 kg/mm²; tensile strength >1200 versus 310–550 MPa; Young's modulus 1140 versus 290 GPa; thermal expansion 1.1 versus 11.6 ppm/K at room temperature; thermal conductivity 2000 (RT)/730 (500 °C) versus 180/97 W/m·K — and beryllium can cause chronic beryllium disease in trace quantities while diamond is non-toxic. Their X-ray windows are built from 20–300 µm foils polished below 10 nm rms and bakeable to 250 °C. Infrared domes follow a different logic: high-Mach aerodynamic heating drives germanium straight into the 350 K ceiling noted above, whereas diamond still conducts 730 W/m·K at 500 °C and resists rain and sand erosion — Crystran lists diamond explicitly as a window and dome material. The most extreme use makes the diamond both anvil and window at once: Dewaele et al. (Nature Communications, 2018) used toroidal culets to reach 603 GPa in static compression (16 µm central flat, ~5–6 µm sample thickness), against roughly 400 GPa for a conventional DAC.

06How to write the spec: an executable requirements checklist

Six groups of numbers can go straight onto a drawing. Size and thickness: commercial infrared windows typically run 5–50 mm in diameter and 300–1000 µm thick. Wedge and figure: wedge 0–1°, flatness <1 fringe/cm at 633 nm. Vacuum integrity: helium leak rate <10⁻⁹ mbar·l/s, bakeable to 250 °C. Surface roughness must be specified per process stage — Yuan et al. (Micromachines, 2024) give the full chain: as-grown polycrystalline CVD surfaces at Ra 13–18 µm with asperities up to 180 µm; laser planarization to 200–600 nm; a 2000-grit wheel to 120–200 nm; 10,000-grit to about 10 nm; W0.1 abrasive to 6–8 nm; CMP to 3–5 nm; and single-crystal mechanical polishing below 1 nm. One critical caveat: below roughly Ra 20 nm, pores and grain boundaries in polycrystalline material block further improvement. Birefringence and strain: Δn ≤1.2×10⁻⁵ in high-dislocation samples, ≤3×10⁻⁶ after dislocation reduction, and ≤5×10⁻⁷ with carefully selected substrates and optimized growth; polycrystalline material cannot serve low-birefringence applications because of inter-grain stress. Nitrogen: optical-grade single crystal should be specified in ppb, with reference samples at 10–40 ppb. AR: bare diamond reflects about 17% per surface for roughly 71% transmission, an ideal quarter-wave layer needs n≈1.55, and sub-wavelength structuring reaches >99% transmission with <0.5% reflectance.

07Single crystal or polycrystalline: a clean dividing line, and how to run acceptance

The dividing line is not fuzzy. Polycrystalline material wins on size and cost: windows exceeding 100 mm diameter at 1.5 mm thick were already routine by 2009, and megawatt millimetre-wave windows — including ITER's Ø80 mm disc — are polycrystalline by design. But it carries two hard constraints: Rayleigh scattering becomes significant below roughly 1 µm, so UV-to-near-IR work essentially requires single crystal; and inter-grain stress rules it out of any low-birefringence requirement. Single crystal wins on optical purity: in high-quality material residual scatter is dominated by surface finish, just 0.04–0.6% at 1064 nm, and the surface can be polished below 1 nm. The price is available aperture — the 2009 review records optical components with 'at least one dimension in excess of 5 mm', which has been the governing bottleneck for single-crystal windows ever since. For acceptance, four practices are worth insisting on: request measured tanδ or α reports rather than grade nomenclature; require full-aperture birefringence imaging, not a centre-point reading; establish flexural strength statistically on small same-lot coupons (as in the 314 MPa ± 16 MPa work above) rather than destructively testing finished large discs; and use the Raman line width — under 1.8 cm⁻¹ at a 1332 cm⁻¹ shift in high-quality samples — as a cheap, fast crystalline-quality screen.

08Where ENTASK sits in this chain

Xi'an ENTASK Semiconductor Technology Co., Ltd. (founded September 2020, registered capital RMB 20 million) supplies diamond optical windows from 20×20 to 45×45 mm in custom thicknesses of 0.5–1.2 mm (material type per the order's technical agreement — the company's existing materials describe optical parts as both single-crystal and polycrystalline, so it must be specified in the contract) — squarely in the band where single crystal is mandatory and aperture has historically been the constraint. ENTASK holds both capabilities in-house — MPCVD system development and diamond material production: its in-house E-MG010K microwave generator operates at 2450 MHz, with the 10 kW configuration supporting a 92 mm growth area and the 15 kW configuration 120 mm; with InsightAction and WiseAction intelligent control, chamber pressure stability is ±0.005 kPa, MTBF exceeds 10,000 h, and full-load continuous operation has run beyond 1,500 h without failure. Process stability matters disproportionately for optics — the three acceptance parameters that recur throughout this article (birefringence, nitrogen content, absorption coefficient) are all, at bottom, readouts of growth-run consistency. The company holds published patent application CN119269411A, a 360° stress analyser for transparent crystals, aimed directly at the birefringence and residual-stress inspection problem. ENTASK is certified to ISO 9001:2015 and supplies mainland China plus India, the Middle East, Europe and North America. (The ENTASK specifications and performance figures below are supplied by the company and have not been independently verified by a third-party test house; buyers should inspect the raw records on site.)

REFERENCES