Skip to content

Raman Microspectroscopy for Structural Indication in Ultrafast Laser Writing

FieldValue
JournalSmall Methods
Published2026-03-15
AuthorsXingrui Cheng, Eugenio Picheo, Zhixin Chen, Martin J. Booth, Patrick S. Salter
DOI10.1002/smtd.202502413

Depletion of the diamond 1332 cm⁻¹ sp³ Raman line predicted resistance in femtosecond laser-written electrodes more reliably than absolute sp² intensity. Wires written at 20 μm s⁻¹ and 200 μm s⁻¹ had resistances of 46.2 kΩ and 2.68 MΩ, respectively, an approximately 58-fold difference. The measurements provide a phase-specific metric for monitoring graphitic electrodes without destructive characterization.

Graphitic carbon absorbs visible light more strongly than transparent diamond, so the written track appears dark. Darkness is only qualitative because absorption, scattering, debris, and illumination artifacts can produce similar contrast. Photoluminescence also darkened the track through nonradiative recombination, but broadband emission reduced its phase specificity.

Raman contrast instead separated bonded phases through characteristic vibrational features. Diamond produced a sharp sp³ line at 1332 cm⁻¹. Graphitic carbon produced a D band near 1350 to 1360 cm⁻¹ and a G band near 1580 cm⁻¹. The study integrated 1325 to 1340 cm⁻¹ for sp³ and 1575 to 1610 cm⁻¹ for sp².

Most structural analyses cited by the authors were destructive or ex situ, including TEM, SEM, and X-ray diffraction microscopy. Raman mapping inspected the finished structures without destructively preparing the sample.

The substrate was optical-grade single-crystal ⟨100⟩ diamond grown by chemical vapor deposition, with nominal nitrogen below 1 ppm. The fabrication laser operated at 520 nm with 400 fs pulses, 10 kHz repetition, and 30 nJ per pulse. The beam was focused through a 20× objective with numerical aperture 0.50.

Wires were 1 μm wide and 200 μm long. Each wire was written in one pass without overwriting. Pads were wider than 25 μm and were 200 μm long. Pad electrodes were produced by stacking parallel wires at a constant 0.5 μm pitch.

Slower scanning increased deposited energy per unit length and produced a controlled range of conductivities. The surface wire geometry represented buried interconnects used in pixel detectors, biosensing electrodes, and microwave striplines. The pad geometry represented contact regions for wire bonding or later metallization.

Raman and photoluminescence were collected after fabrication with 532 nm continuous-wave excitation at 30 mW. A 60×, 0.95 numerical aperture objective collected the post-fabrication maps. Measurements used 500 nm lateral steps and 1 s exposures. Excitation power was below the laser modification threshold, so the imaging did not alter graphitization.

The microscope split the signal 99:1 between Raman and photoluminescence paths. The Raman path used a 1200 lines mm⁻¹ spectrograph and a cooled CCD camera. Raman spectra can contain nonspecific fluorescent background, which baseline correction may reduce.

Each map used nearby pristine diamond regions for normalization against tilt, focus drift, and excitation power changes. The brightest 200 wire pixels and 1000 pad pixels supplied each pristine reference. The darkest 60 wire pixels and 1500 pad pixels supplied each modified region.

For every pixel, the metric was the ratio of integrated sp² intensity to integrated sp³ intensity. Fixed windows made this a ratio of band integrals, not a ratio of intensity per unit wavenumber.

Mean spectra used 60 measurements per wire across 10 to 1270 μm s⁻¹. At 500 μm s⁻¹, the spectrum remained sp³ dominated, with D and G bands barely discernible. Lower scan speeds depleted the 1332 cm⁻¹ line and strengthened the D and G complex.

A 100 μm s⁻¹ pad showed depleted sp³ signal and an almost complete sp² conversion signature. Below 50 μm s⁻¹, apparent sp² intensity declined slightly even as sp³ depletion continued. The authors connect this response to increasing disorder and smaller sp² clusters.

At extreme doses, broad or merged D and G modes can lose contrast without a matching fall in sp² bonding. The downturn agrees with other diamond graphitization studies cited by the authors. Absolute sp² intensity can therefore misrepresent the structural state in the deeply graphitized regime.

A 50 μm s⁻¹ wire had 74.8 kΩ resistance and was used for the ratio and unmixing analysis. Pad writing caused stronger sp³ depletion than single-pass wire writing. The authors attribute this to stress-assisted conversion during overlapping passes.

Hyperspectral unmixing recovers phase maps

Section titled “Hyperspectral unmixing recovers phase maps”

Each spectrum was modeled as a linear mixture of two endmembers. Vector component analysis and N-FINDR estimated the endmembers. Fully constrained least squares then assigned nonnegative abundances that summed to one.

Over 1000 to 1800 cm⁻¹, the diamond endmember retained the 1332 cm⁻¹ line. The graphitic endmember carried a broad D/G complex and a suppressed diamond line. The diamond endmember also showed a band near 1430 cm⁻¹, consistent with neutral single nitrogen-vacancy defects.

Abundance maps tracked the sp³ window map and separated pristine diamond from written tracks. Excluding the 1332 cm⁻¹ line narrowed the region to 1345 to 1800 cm⁻¹. The expected spatial structure remained, but signal-to-noise ratio fell because the strongest phase-specific line was removed.

N-FINDR assumes that pure pixels are present in the dataset. Linear mixing can fail under strong fluorescence, self-absorption, re-emission, or stress-induced peak shifts. The authors suggest nonlinear unmixing for those cases, at the cost of more parameters and lower interpretability.

Although the number of endmembers was prescribed, the method avoided manual band selection. For diamond, unmixing did not offer a clear advantage over known band assignments.

Raman intensity tracks electrical resistance

Section titled “Raman intensity tracks electrical resistance”

Resistance came from linear least-squares fits to two-probe current-voltage sweeps, typically within ±500 mV. The main current-voltage traces were linear over ±500 mV, indicating ohmic behavior and no measurable Schottky barriers.

At 100 μm s⁻¹, a 200 μm wide pad had 332 Ω resistance. A wire written at the same speed had 1.46 × 10⁵ Ω resistance. The pad’s larger cross-section and overwriting explain its lower resistance. Pads remained more conductive than wires at identical scan speeds.

Across devices, resistance rose with scan speed and normalized sp³ intensity. Lower resistance accompanied stronger sp³ depletion and a more connected graphitic phase. The sp²/sp³ ratio was useful at moderate doses because normalization reduced sensitivity to power changes.

In the high-dose regime, sp³ depletion gave lower variance than the ratio. Broadband background also weighs more heavily on weak sp² signals in the low-dose regime because their signal-to-noise ratios differ.

At the lowest speeds, wire resistance approached that of the contact electrodes, reducing measurement accuracy. A wire written at 750 μm s⁻¹ had non-ohmic current-voltage behavior. It gave a higher apparent resistance than the wire written at 1000 μm s⁻¹. Electrical contacts used two micromanipulated platinum probes. Surface debris was removed with isopropyl alcohol before measurement.

The reported maps validate indicators for monitoring, but the measurements were not performed during writing. An integrated system could offset the Raman excitation from the fabrication beam and use subsecond integrations. Under that configuration, Raman sampling could provide feedback without interrupting fabrication.

Brightfield and photoluminescence should remain qualitative checks because neither isolates the relevant carbon phase. The self-referenced metric benchmarks each map against its local pristine background. For high-dose graphitization, sp³ depletion was the more stable indicator. The sp²/sp³ ratio can support moderate-dose control when experimental conditions vary.

Device evaluation still needs application-specific tests and may require methods beyond those used here. For engineers, the data make diamond quality and surface condition relevant checks when interpreting Raman feedback. This creates a path to specification-driven feedback during scaled femtosecond laser microfabrication.

Laser-written graphitic electrodes need a polished single-crystal diamond host for femtosecond writing and confocal Raman inspection. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal plates and wafers up to 125 mm, polished to Ra below 1 nm. The engineering support page can be used to check substrate and process requirements against the paper’s electrode design.

View Original Abstract

ABSTRACT Femtosecond laser fabrication enables the creation of a wide range of devices, but its scalability and yield can be limited by the lack of real‐time, in situ monitoring tools. In particular, there is a strong need for metrics that directly correlate with device performance. Raman microspectroscopy provides a non‐destructive route for in situ characterization. Here, we demonstrate its potential to assess the electrical performance of laser‐written graphitic electrodes in diamond. By combining hyperspectral mapping with electrical testing, we show that depletion of the 1332 sp3 Raman line serves as a monotonic and robust predictor of resistance, offering clear advantages over commonly used spectral features. We further introduce hyperspectral unmixing as a label‐free approach to identify relevant spectral signatures in fabrication processes where Raman markers are less defined. Importantly, the methodology we present is not restricted to diamond but can be adapted to other host materials and functionalities, offering a practical path toward specification‐driven fs‐laser microfabrication.