
Non-Polarizing Beamsplitter
A non-polarizing beamsplitter (NPBS) is an optic that divides an incident laser or broadband beam into reflected and transmitted beams by optical power, while keeping the split nearly the same for s- and p-polarization. It is specified by a power ratio (for example 50:50 R:T), not by an extinction ratio.
That distinguishes it from a polarizing beamsplitter (PBS), which separates light by polarization state, and from a dichroic beamsplitter, which separates by wavelength.
Why “non-polarizing” is hard:
At oblique incidence, Fresnel reflection is stronger for s-polarization than for p-polarization, so a simple dielectric interface or an ordinary multilayer coating is strongly polarization-dependent.
An NPBS coating is designed so that, over a stated wavelength band and angle of incidence,
∣Rs−Rp∣ and ∣Ts−Tp∣
stay small (often a few percent). The polarization state of the input is therefore approximately preserved in both outputs, and the split ratio does not swing with the input polarization.
Common coating approaches:
Hybrid metal–dielectric stacks (often with a thin silver layer) for broadband cubes. Polarization dependence is low, but the metal adds a few percent to about 10% absorption, so T+R<100.
All-dielectric multilayers for laser-line plates or cubes. Absorption is negligible, but the non-polarizing band is narrower.
Designs that combine interference with frustrated total internal reflection inside a cube for very flat Rs≈Rp over a visible band.
Forms:
Cube. Two right-angle prisms (typically N-BK7 or fused silica) are cemented or optically contacted, with the beamsplitter coating on one hypotenuse. Input is near normal to a face; the reflected beam exits at about 90°. Faces are usually AR-coated. A single internal reflection avoids the ghost reflections of a plate, and the optical path is equal in both arms if the cube is square. Cement limits pulsed or high-power use; optically contacted or air-spaced cubes are used when the adhesive would be damaged.
Plate. A flat substrate used at 45°, beamsplitter coating on the first surface and AR coating on the second. Lower dispersion and aberration than a cube for ultrafast beams, but the transmitted beam is laterally shifted and a residual second-surface reflection can produce a ghost unless the plate is wedged. Polka-dot (spatially patterned metal) plates are another broadband, polarization-insensitive option, at the cost of diffraction.
Typical commercial numbers for a 50:50 part (examples from Coherent, Newport, and Thorlabs datasheets): Rs≈Rp≈50%±5%, Ts≈Tp≈50%±5%, with ∣Rs−Rp∣ and ∣Ts−Tp∣ often within 5% or better at the design wavelength. Other ratios (10:90, 30:70, 70:30, 90:10) are common. Surface quality is often 40-20 or better, reflected-beam deviation a few arcminutes, and AOI tolerance about ±3∘.
Applications:
Sampling a laser for power monitoring or feedback without the sample depending on source polarization.
Splitting or combining beams in interferometers (Michelson, Mach–Zehnder) when both arms must keep the input polarization.
Fiber and telecom components (wavelength-selective switches, 40G/100G modules) where polarization-dependent loss must stay low.
Multi-path laser processing, pumping, and illumination.
Quantum optics, where a 50:50 NPBS routes single photons with a polarization-independent probability.
Low-dispersion plate NPBS versions in ultrafast experiments, where cube glass and cement would add group-delay dispersion and risk damage.
Selection trade-offs are bandwidth versus absorption, cube convenience versus plate dispersion and ghosts, and damage threshold (dielectric laser-line plates often rate around hundreds of W/cm² CW and ~0.5 J/cm² for 10 ns pulses at 1064 nm; hybrid metal cubes are lower). Match the coating to the laser line or band and to the actual angle of incidence; performance falls off outside the design window.