Probing Raman scattering for particle tracking: A novel spectroscopic analysis method

Benjamin T. Hogan,1,2 Evgeniya Kovalska1 & Anna Baldycheva1

1.Department of Engineering and Centre for Graphene Science, College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, EX4 4QF, UK.

2. EPSRC Centre for Doctoral Training in Electromagnetic Metamaterials, University of Exeter, EX4 4QL, UK.
Ben graduated in 2015 with a Master’s degree in Natural Sciences from the University of Bath, majoring in physics. In September 2015, he began a PhD with the Metamaterials CDT at the University of Exeter, where his project involves the investigation and application of 2D material- liquid crystal composites. 
We demonstrate a widely applicable technique for the spatial tracking of nanoparticles in multiple dimensions, through the use of Raman spectroscopy measurements. The technique presented can be applied to the tracking of any nanoparticle with well-defined Raman spectral characteristics. We show, both numerically and experimentally, that through suitable design of waveguide structures, significant enhancement of the Raman signal for liquid dispersed nanoparticles is obtained. We then describe how this enhancement can be used to facilitate the determination of individual nanoparticles’ positions.
We acknowledge financial support from: The Engineering and Physical Sciences Research Council (EPSRC) of the United Kingdom via the EPSRC Centre for Doctoral Training in Electromagnetic Metamaterials (Grant No. EP/L015331/1).
Corresponding author
Benjamin Hogan email: 
In the modern age, nanoparticle dispersions are increasingly becoming a ubiquitous element across the full breadth of scientific endeavour.
Applications in treating cancer, drug delivery, antibacterial coatings, cosmetics formulations, construction materials and countless other technologies have been developed; representing a huge outlay of human effort and resources. However, nanoparticles are not only synthetic. Biological systems- for example the human body rely heavily on nanoparticles to function efficiently. 
The potentially limitless pool of nanoparticle applications still holds great promise for researchers. However, to initially demonstrate the technique presented here, we will introduce and then focus on the specific case of liquid crystal dispersed two-dimensional materials, which demonstrate observable self-assembly processes.
These materials inherently possess several properties that make them of significant interest, such as: fluidity, hence they can be readily integrated into microfluidic systems; reconfigurability under applied electric field, magnetic field and thermal gradients; and scalability of the synthetic routes for these materials. 
In presenting options for reconfigurability however, a problem arises. The possible methods for monitoring the reconfiguration of dispersed nanoparticles are severely limited.
Let’s say, for example, we have a microfluidic waveguide. In this waveguide we have a particle of a 2D material. In order for the 2D material particle to have maximal interaction with light propagating in the waveguide, it should be positioned at the centre of the channel.
But how can we establish if it is at the centre? We could use microscopy and look at the particle, but that won’t give you information on its vertical position. Furthermore, what if the particle moves? Hence, we need a technique that can accurately determine the position of an individual nanoparticle, and then also be repeatable at a timescale suitable to monitor the variation in that nanoparticle’s position under the effects of external forces.
Electronic systems are the predominant and most ubiquitous technology in existence. In particular, silicon structures at the nanoscale are the driving force behind almost all computing globally.
However, as we approach the achievable limits of silicon-based devices, new technological paradigms are required to keep pace with Moore’s law.
One nascent technology of huge potential is optoelectronic devices capable of interacting with and controlling the propagation of light. Current limitations to the efficacy and uptake of optoelectronic and photonic devices cover a broad range, including: cost, scalability, achievability and reproducibility of desired results and compatibility with existing technologies. The development and characterisation of new materials, therefore, is a key requirement towards overcoming these hindrances.
Liquid crystalline nanocomposite materials are a novel paradigm of hybrid fluid materials, which are currently attracting significant interest from the optoelectronics community due to their unique capability to interact with light, utilising many possibilities such as photoluminescence, plasmonics and quantum optics.
Such nanocomposite materials are based on low-dimensional nanoparticles (graphene, transition metal dichalcogenides (TMDCs), metal nanoparticles etc.) dispersed in a fluidic host material. Liquid crystalline properties can be accessed through the solvent-facilitated self-assembly of particles possessing suitable shape anisotropy or chemical functionality. 
In particular, 2D nanoparticle liquid crystals (2DLC) hold great promise as drivers of a revolution in the development of optoelectronic and photonic devices.1
By combining the exotic and varied properties, within the few-layer limit, of the ever-expanding body of exfoliatable layered materials (graphene, transition metal dichalcogenides, MXenes, etc.), with the inherent reconfigurability of liquid crystals under applied fields, novel nanocomposites with huge potential can be produced.
Such 2DLCs can be synthesised in a number of ways; firstly, exfoliated particles of a 2D material can be dispersed in a liquid crystalline host matrix (Figure 1). Alternatively, the 2D nanoparticles themselves can act as the liquid crystal mesogens, through micro- or nano-scale self-assembly when dispersed at suitable concentrations within an organic solvent host fluid. 

Figure 1. a) Scheme for production of two-dimensional – liquid crystal composite materials. b) Starting from a bulk material in an organic solvent, by a process of ultrasonication, centrifugation and fractionation, a highly monodisperse solution of exfoliated 2D material can be produced. c) Optical reflection images under blue light of a dispersion of molybdenum disulfide nanoparticles on isopropanol.
Through application of the proposed technique below, we can successfully monitor the temporal variance of the spatial distribution of fluid dispersed nanoparticles of any type. 
We demonstrate a characterisation technique that can be used to elucidate the positions of individual nanoparticles within a confined fluid host at any given moment in time. 
Building upon our novel approach for wafer-scale integration of 2D materials on CMOS photonics chips, we herein demonstrate that the design of an optofluidic waveguide system can be optimised to enable simultaneous in-situ Raman spectroscopy monitoring of 2D dispersed flakes during the device operation. A novel method for simultaneous three-dimensional determination of the positions of individual nanoparticles is presented. 
This technique is not limited to the specific case discussed herein, but rather has applicability to the monitoring of a diverse range of Raman-active nanoparticles, representing a key step towards the characterisation of nanoparticle self-assembly and transport processes. 
As test samples to demonstrate the applicability of our technique, liquid crystalline nanocomposite materials consisting of molybdenum disulfide flakes dispersed in a nematic liquid crystal were synthesised via the following procedure.
Molybdenum disulfide was exfoliated to few-layer thicknesses from the bulk solid by use of a liquid phase exfoliation method,2 wherein ultrasonication of molybdenum disulfide dispersed in a suitably chosen organic solvent induces cleavage of the interlayer van der Waals bonding, with significantly less disruption of the intralayer bonding. This produces a solution containing a range of particle sizes, with platelet shapes typically being observed.
Resulting dispersions of few-layer molybdenum disulfide were then centrifuged to remove any residual bulk, or otherwise large, molybdenum disulfide particles.
The centrifuged dispersions were then mixed with commercial nematic liquid crystal formulations and subsequently ultrasonicated to ensure homogeneous mixing. The organic solvent was then removed from the mixture under vacuum using a Schlenk line to leave homogeneously dispersed molybdenum disulfide particles suspended in the liquid crystal host. The resultant liquid crystalline nanocomposite was then integrated into microfluidic structures on silicon-on-insulator chips.
Dispersions of other 2D materials can also be produced by the same method, and further integrated in the same way as well. Using scanning electron microscopy, and polarised light microscopy, the integration of molybdenum disulfide nanoparticles into microfluidic channels can be clearly observed (Figure 2).
Figure 2. a) SEM image of an empty microfluidic chip. b,e,f) Polarised microscopy images of infiltrated microfluidic chips. 2D material particles can be seen in the liquid as dark areas in b and f, and as a bright spot in the inverted image (e). c-d) SEM images of the infiltrated microfluidic channels. Flakes of 2D material can be seen as either surface distortions of the liquid or dark area, depending on their vertical position within the channel.
The microfluidic structures used were carefully designed to enhance the intensity of the Raman spectroscopy signal from the molybdenum disulfide particles. To achieve this enhancement, the desirable microfluidic geometry was determined from numerical analysis of the Raman light generated under laser excitation at 532nm, at normal incidence.
Raman scattering is a quantum mechanical process with a random spatial distribution of the photons involved, however the optical behaviour of the scattered light can be modelled using a classical electrodynamics approach. The backscattered Raman signal intensity was numerically determined for wavelengths corresponding to the scattering from the Raman active bands using the scattering matrix method. The scattering matrix method (SMM) is a powerful tool for numerical determination of the near- and far-field light distribution for structures which can be split into uniform layers in at least one direction.
The main principle of this method is the decomposition of the electric and magnetic fields into Fourier series in each layer and the connection of the Fourier components within adjacent layers in accordance with the boundary conditions of Maxwell’s equations.
The molybdenum disulfide flakes within the nanocomposite were modelled as a system of chaotically-oriented oscillating electrical dipoles within the microfluidic channel, with the dipole emission as the origin of the Raman signal. They were assumed to have widths of 1µm while being negligibly thick. In this work, we use a previously proposed and demonstrated3 silicon-on-insulator (SOI) based optofluidic waveguide channel design for in-situ micro-Raman detection and monitoring of the integrated nanocomposite, where enhancement is achieved due to tuning of the Fabry-Pérot type resonances within the geometry of the microfluidic structure.
The channels are bounded by: air above, two silicon walls, and a silicon dioxide buffer layer below. Below the buffer oxide layer is the silicon substrate. We optimised the optofluidic waveguide design to significantly enhance the back-scattered Raman signal from the incorporated molybdenum disulfide nanoplatelets. Channels were designed to be 15µm deep and 3.7µm across, and the buffer oxide layer to be 2µm thick (Figure 3).
Figure 3. Schematic of the cavity design showing dimensions and materials used.
Having utilised careful design of the geometry to enhance the Raman signal of a particle, we then desire a method to accurately track the position of the liquid dispersed particles in three dimensions simultaneously. 
We have developed a method combining numerical determination of the dependence of the Raman signal intensity on a particle's position with experimental spectral analysis. Again, using a scattering matrix method, we first establish the effect of a particle's position on the intensity, here for the two characteristic bands of molybdenum disulfide. 
The effect of the flake position on the Raman signal intensity was modelled by varying the position of the oscillating dipoles within the optofluidic waveguide channel both laterally (x) and vertically (y), with the third dimension (z) invariant in this case and therefore displacement in the z-direction has no effect on the Raman signal.
However, the only barrier to modelling with a variant z is the increasing computational time required.
Very little difference is immediately observable owing to the very small difference in wavelength between the Raman photons scattered after interaction of the laser light with the two molybdenum disulfide Raman bands respectively (Figure 4a-b). 
The scattered radiation is expected to have energies corresponding to wavelengths of 543.15nm and 543.69nm respectively. However, we can then analyse the difference in the predicted intensities, alongside the ratio between them, at which point a stronger inter-band variance is observable (Figure 4c-d).
Figure 4. a-b) The predicted intensity for the Raman bands of MoS2 with emission corresponding to wavelengths of 543.15nm and 543.69nm respectively under excitation by a 532 nm laser. Inset: schematic of the cavity showing nanoparticle at position (x,y) in the cavity. c) The difference between the Raman intensity for the two bands and d) the ratio between the intensities of the bands. By comparing experimental spectra to these numerical results, one can determine the nanoparticle position accurately.
We can then take the experimental Raman spectrum measured for a particle at an unknown position within the microfluidic channel and extract the peak intensities for each of the Raman bands.
One advantage of this methodology is that the Raman spectra can be gathered at very high speed, owing to the fact that measurements are only required at a small number of wavelengths – in essence, the repetition rate of scanning and hence also of positional determination is limited only by the mechanical properties of the Raman spectrometer itself. It should be noted that the procedure below is significantly easier to apply for two or more dimensions simultaneously, than that for a single tracking dimension.3
These peak intensities are then normalised against the spectrum for a particle of the same size and shape taken in a microfluidic reservoir where there is negligible influence of the microfluidic geometry on the signal intensity.
Normalisation is used to account for the fact that the intensities of different Raman bands are not expected to be identical under standard conditions -a fact that the scattering matrix method used doesn’t consider. The normalisation process also removes any enhancement observed due to the substrate rather than the microfluidic structure.
There are two requirements for the spectrum used for normalisation: firstly, the aforementioned lack of microfluidic geometry influence on the signal, and secondly, that the liquid depth is equivalent to that in the channels where particle tracking is taking place. 
From the normalised data, the ratios of the peaks for the two Raman bands of molybdenum disulfide are calculated. The absolute difference in intensity is also calculated.
We then compare the normalised peak intensity differences and ratios to those established from the numerical calculations (Figure 5). Initially, a broad range either side of the exact peak difference or ratio is used to establish the possible positions of the particle (Figure 5 insets).
A range of values either side of the precise experimental values is used in order to account for experimental errors in the spectral acquisitions. By narrowing down the range, a decreasing number of possibilities are observed. By comparing the possibilities for the position using each of the peak difference and peak ratios separately, a single point of coalescence can be found corresponding to the particle's precise position within the channel (Figure 5).
Figure 5. Determination of particle positions by combining numerical analysis and experimental data for differences (blue) and ratios (red) of the two Raman bands of molybdenum disulfide. The differences have been slightly offset so that the overlapping region is clear. Insets: The possible positions using a wider range round the precise experimental values.
By taking multiple spectra at fixed time intervals, particles can be tracked spatially within the cavity, as their positions change with time. Temporal resolution is limited by the available sampling rate of the Raman spectrometer used.
It should be noted that the cavities used were symmetric in the x-direction, hence two possible positions are determined with no way to distinguish them. This would easily be rectified by the use of an asymmetric cavity, or by an off-centre alignment of the exciting Raman laser. There are two potential limiting factors to the accuracy of the positional determination achieved. Firstly, the resolution of the numerically determined data set. However, this can readily be mitigated given greater preparatory computational time. Secondly, the accuracy of the experimental spectra. This can also be improved by a number of different means.
Further optimisation of the cavities used to give greater signal enhancement would reduce the percentage errors involved. Alternatively, spectra could be acquired for longer, sacrificing potential temporal resolution in exchange for spatial accuracy.
We demonstrate a novel technique for the monitoring of the spatial dynamics of nanoparticle systems.
The technique is based on combining numerical analysis and experimental Raman spectra for nanoparticles confined within microfluidic cavities. Through application of the technique above, we can successfully monitor the temporal variance of the spatial distribution of fluid dispersed nanoparticles of any type.
Hence, we believe that this could represent a powerful technique for the monitoring of self-assembly processes in nanoparticular systems and composites, allowing understanding of potential modes of device operation.
1. B. T. Hogan, E. Kovalska, M. F. Craciun, and A. Baldycheva, "2D material liquid crystals for optoelectronics and photonics," J. Mater. Chem. C, vol. 5, no. 43, pp. 11185-11195, 2017.
2. C. Backes et al., "Guidelines for exfoliation, characterization and processing of layered materials produced by liquid exfoliation," Chemistry of Materials, vol. 29, no. 1. 2017.
3. B. T. Hogan et al., "Dynamic in-situ sensing of fluid-dispersed 2D materials integrated on microfluidic Si chip," Sci. Rep., vol. 7, p. 42120, Feb. 2017.
Website developed by S8080 Digital Media