All of academia has a replication crisis at the moment however this is less theoretical than most and easily passes the sniff test.

You know how bismuth crystals have all sorts of different colors? It’s essentially growing a “bismuth crystals” on top of a cmos camera, except the “bismuth crystal” is much more random and the specific wavelength of light it lets through is dependent on some physics fuckery.

Will it ever be commercially produced? I doubt it, but hope I’m wrong:

• the lenses will not perfectly overlap each sensor resulting in many having ‘leakage’ from other frequencies resulting in a high signal to noise ratio
• there doesn’t seem to be a way to guarantee a consistent number of sensors per frequency resulting in highly variable sensitivity per frequency.
• Relying on randomness and only releasing the ones that are “good enough” is a fairly common practice but the yields are abysmal which causes the price to skyrocket.
• The use of a spectrogram is primarily as a scientific instrument, and an instrument which has wildly variable sensitivity/selectivity per sensor is a cause for concern.

I however do see potential uses for a cheap handheld machine that can do a quick and dirty material composition check. Contaminant tester (drugs, assembly lines, chemical stocks, etc.), hobbyist labs, chemical reaction monitor, etc.