Raman spectroscopy: how it works, how to read a Raman spectrum and what it tells you about materials

When laser light hits a sample, almost all of it is scattered elastically, at the same energy (Rayleigh scattering). A tiny fraction is scattered inelastically: the photon gives energy to, or takes energy from, a molecular vibration or a lattice phonon and leaves with a different energy (Jones et al. 2019). That energy difference is the Raman shift, and it tells you the vibrational modes of the sample (Silver et al. 2019).

When the scattered photon has less energy than the laser photon (red-shifted), the process is Stokes scattering; when it has more (blue-shifted), it is anti-Stokes. Stokes lines are the stronger ones, because fewer molecules start in an excited vibrational state (Jones et al. 2019). A vibration is Raman active when it modulates the polarisability of the molecule or crystal (Jones et al. 2019), whereas infrared absorption needs a change in dipole moment (Baker et al. 2014). In materials with a centre of inversion, a mode is either Raman active or IR active, not both (Jones et al. 2019), which is why Raman and FTIR are used together.

A Raman microscope focuses the laser through an objective, filters out the Rayleigh line and disperses the scattered light onto a CCD detector; a confocal pinhole passes only light from the focal plane (Jones et al. 2019). See the characterisation techniques hub for how Raman compares with other methods.

A Raman spectrum plots intensity against Raman shift in cm⁻¹, the wavenumber of the laser minus that of the scattered light; for Stokes lines the shift equals the vibration wavenumber (Jones et al. 2019). The IUPAC convention runs the shift axis from right to left, but many papers plot it left to right (Jones et al. 2019). As an example, liquid CCl₄ gives Raman bands at 218, 314, 459, 762 and 790 cm⁻¹ beside the Rayleigh line (Jones et al. 2019).

Graphene is the classic materials example (Silver et al. 2019):

• G band, ~1580 cm⁻¹: first-order scattering from two degenerate phonon modes at the Γ point.
• 2D (or G′) band, ~2700 cm⁻¹: a second-order, two-phonon process. In monolayer graphene it is a single peak; in bilayer graphene it splits into four components.
• D band, ~1350 cm⁻¹: appears only when defects are present, at about half the 2D frequency.

The 2D band changes shape, width and position as layers are added while the G band shifts slightly down, which allows unambiguous, non-destructive counting of graphene layers (Ferrari et al. 2006). The D-to-G intensity ratio, I(D)/I(G), measures point-defect density, but it depends strongly on the laser excitation energy and peaks at an inter-defect distance of about 3 nm, so one ratio can match two defect densities; the G band width resolves the ambiguity (Cançado et al. 2011).

For MoS₂, the in-plane E′ mode (~382 cm⁻¹) and out-of-plane A′₁ mode (~403 cm⁻¹) move apart as layers are added, and their separation gives the layer number (Silver et al. 2019).

• Little preparation. Raman usually needs no extensive preparation and no labels, and water does not normally interfere (Butler et al. 2016). Water and polar solvents absorb strongly in the infrared, which is one reason Raman is preferred for wet samples (Jones et al. 2019).
• Choose the laser. UV and short visible wavelengths cause strong photoluminescence in organic materials that can mask the Raman peaks, so longer visible or near-IR lasers (500–830 nm) are often used; 1064 nm is common for strongly fluorescent materials (Jones et al. 2019). The Raman signal falls at longer wavelengths, so acquisitions take longer, and silicon CCD detectors lose efficiency above about 800 nm (Jones et al. 2019). 785 nm is the most common compromise, and visible 500–830 nm lasers suit inorganic materials such as graphene and carbon nanotubes (Jones et al. 2019).
• Laser power. Raising the power and focusing tightly increases the signal but can photobleach or damage the sample; continuous-wave lasers are preferred for spontaneous Raman for this reason (Jones et al. 2019).

2D materials. Raman is a routine first check for graphene: it gives the number and orientation of layers, edge type, and the effects of strain, doping, disorder, electric and magnetic fields and functional groups (Ferrari & Basko 2013). The D band, absent in pristine graphene, is the quick test for defects introduced by chemical modification, as in graphene oxide, and I(D)/I(G) at a stated laser energy quantifies point defects (Silver et al. 2019; Cançado et al. 2011). For MoS₂, the E′–A′₁ separation gives the layer count (Silver et al. 2019), and in general Raman can determine 2D layer thickness with atomic-level precision from inter- or intra-layer modes (Jones et al. 2019).

Crystals and phonons. Raman probes phonon modes in crystals, and in 2D materials it also reports inter-layer breathing and shear modes, in-plane anisotropy, doping, disorder, thermal conductivity and strain (Jones et al. 2019).

Mapping and enhancement. Confocal Raman maps composition across a sample at about the diffraction limit; surface-enhanced Raman (SERS) on rough gold or silver raises the signal by orders of magnitude, and tip-enhanced Raman (TERS) reaches 10–30 nm resolution (Jones et al. 2019).

Pair Raman with FTIR for IR-active modes, XRD for crystal phase, AFM for thickness of 2D flakes, and XPS for surface chemistry.

• Fluorescence background. Photoluminescence from organic materials can mask the Raman peaks; move to a longer-wavelength laser such as 785 or 1064 nm (Jones et al. 2019).
• Laser damage. Too much power or too tight a focus can photobleach or damage the sample (Jones et al. 2019).
• Quoting I(D)/I(G) without the laser energy. The ratio depends strongly on excitation energy, and a single value can correspond to two defect densities (Cançado et al. 2011).
• Weak signal. Spontaneous Raman is a very weak process, so dilute species may need SERS or longer acquisitions (Jones et al. 2019).
• Missing modes. Some modes are absent because of selection rules for the scattering geometry or because the Rayleigh filter cuts off low shifts (Jones et al. 2019).
• Spatial resolution. Conventional Raman cannot resolve features much smaller than about half the laser wavelength (Jones et al. 2019).

The LabCanvas figure editor Library has the Raman microscope as an editable object, including a cutaway of the beam path from the laser through the objective to the sample and back through the Rayleigh filter, pinhole and grating to the detector, and an editable data plot for the spectrum. Plot your spectrum against Raman shift, label the peaks (for graphene: D, G and 2D, with I(D)/I(G) and the laser wavelength), overlay spectra for different layer numbers, and export a high-resolution image for a methods figure or graphical abstract.