Experimental Details:

1)	The Raman spectra of crystal violet were acquired using an inVia Raman Microscope
2)	The incident laser was a 50 milli Watt continuous wave laser at 532 nm
3)	The grating had 1800 lines per mm and was blased for 532 nm
4)	The objective was an infinity corrected N-plan 50x short working distance objective (numerical aperture of 0.75) manufactured by Leica.
5)	Spectra were typically recording by averaging a square grid of 25 acquisition points on the sample. These were separated by 1 micrometre in the SERS sample and 5 micrometres on the Silicon substrate.
6)	The integration time for each point was 10 seconds.
7)	The spectrometer slit width was 60 micrometers wide.
8)	The spectral resolution is 0.3 wave numbers.
9)	The fluorescence background was removed using a polynomial fit algorithm in the Renishaw WiRE software and is a proprietary process.
10)	The spectra were exported to txt file and plotted in python for the published article.

SERS Substrate fabrication details:
1)	The SERS substrate was fabricated on a Silicon substrate with a thermally grown 200 nm silicon dioxide surface layer. 
2)	A double polymethyle metacrylate-methyl metacrylate resist layer was used to cover the substrate. 
3)	The G-shaped nanostructures were then created by electron beam lithography. 
4)	A 25 nm layer of Gold was deposited by evaporation using a DC sputtering system. 
5)	The resist was then removed by a ‘lift-off’ procedure. 
6)	The unit cell comprised of four such motifs in a four-fold tetrad. 
7)	The array of Gold nanostructures covered an area of 2.5 mm × 2.5 mm on the Silicon substrate. Hence, the enhancement of the Raman signal from CV on the SERS part of the sample was compared to measurements of CV on the clean Silicon area of the sample.
8)	A 0.3 milli-molar solution of crystal violet in ethanol was drop-casted onto the SERS subsrate and allowed to evaporate in air. Because crystal violet has three amino groups, it bonds to the Gold surface and the effect of liquid tension during the evaporation of ethanol is minimized. 

Simulations Details:
1)	Simulations of the SERS substrate were performed in in ANSYS Lumerical.
2)	Periodic boundary conditions were imposed around the edges of the unit cell to simulate an infinite array of unit cells. 
3)	The Silicon substrate model was a 500 nm thick slab of Silicon based on Palik material model (details in a later section). 
4)	The G-shaped unit cell was modelled in Autodesk Inventor and imported into Lumerical with a CRC [handbook of chemistry and physics] material model (details in a later section). 
5)	A plane wave source was introduced from 1 micrometre above the surface with a wavelength range of 0.2 to 1.8 micrometres. 
6)	The void above the plane wave source extended by an additional 1 micrometres. 
7)	Perfectly matched layer boundary conditions were applied to the top and bottom plane of the simulation domain. Hence, the simulation perceived the Silicon substrate to be infinite in extent. 

Palik material model of Silicon:
1)	The fit parameters for the material model data were in the wavelength range of 0.2 to 1.8 micrometres
2)	The ‘fit tolerance’ was set to 0.1
3)	The ‘max coefficient’ was set to 10
4)	The ‘imaginary weight’ was set to 1
5)	The ‘improve stability’ option was checked
6)	The ‘make fit passive’ option was checked

CRC [handbook of chemistry and physics] material model of gold:
7)	The fit parameters for the material model data were in the wavelength range of 0.2 to 1.8 micrometres
8)	The ‘fit tolerance’ was set to 0.001
9)	The ‘max coefficient’ was set to 12
10)	The ‘imaginary weight’ was set to 10
11)	The ‘improve stability’ option was checked
12)	The ‘make fit passive’ option was checked