Depending on your excitation wavelength, aluminia or alumnium substrates with an negligible amout of aluminia (i.e. air oxidation or simple an aluminium plate/substrate with no further treatment) helped for 785nm, 633nm and 532.8 nm  excitation without too much pl background, depending on the excitation power, it holds also for SERS or other a little more complicated Raman probes with weak singnals), depending on what you mean with weak.

For 532.8nm and 633 nm, but mostly for 532.28nm, one have to deal with a really high background signal if ones works with glass or glass similar substrates at high powers, (usually for Raman compared to SERS, one can see this at around 5-10mW input power (before your optical system. if you use a comercial Raman microscope like Rainshaw or Horriba Aramis or similar) with a 50x objective and you don't have the exact feeling about your spot size. Usually for a home made SERS Microscope in order to avoid the background, and with the best SERS nanostructures substrate one needs only 20kW/cm2 (thin multilayer film of densely packed nanostars coated with MBA SAMs) and for a single particles/nanostars cca. 120-240 kW/cm² (roughly maximal 1.9 mW on the sample with 100x objective , assuming a 1 micron diameter spot to be modest). Laser powers for more than 3mW focused on a less than 1 micron diameter , always give an  pl background on glass.

For more NIR wavelength is hard to say,  depending on your lab conditions and so on, sometimes you have to deal with some water films,  i hope some other colleagues can help you for this spectral region

We use UV-quality CaF2 (fluorite) slides. Circular IR fluorite windows can be used as well, and they are a bit cheaper than the fluorite microscope slides. They can be pretty expensive (> 50-100 euros per piece), but can be easily cleaned. The performance is excellent, flat background (save for a sharp band at 320 cm-1, which can be used for calibration) with any excitation wavelength, including NIR (e.g. 785 nm). Moreover, they are slightly hydrophobic, so if you deposit a drop of aqueous solution, it wont spread like onto glass slides. We use them for aqueous samples (a drop), for thin sections of biological tissues, for biological cells, and for many other kinds of samples (nanoparticles, thin films, etc.).

Try Anolux III from this source: http://anomet.com/aluminum/

You may ask them for samples free of cost. From my experience Anolux III gives the lowest Raman background.

http://anomet.com/assets/docs/Anomet-Standard_Aluminum_Stock_Sheet_2014.pdf

Another option is a silver (visible excitation) or gold (for red/NIR excitation) surface mirror (note, this is a smooth reflective layer, not a roughened or colloidal one so does not promote SERS). This will reflect the laser back through the sample. Further gains can be made by using a dichroic filter above that will allow Raman shifted wavenlengths to escape but will reflect unchanged wavelengths back down again, so that each photon has several opportunities to interact with the sample. Don't know the references off hand but I think Pavel Matousek done some work in this area.

Dear John Brindle

Polished silicon wafer could show a large peak using 785 nm laser setup.

It is very interesting your feedback about liquid samples, It really does not work.