Let's crack on. We are exploring an LLM mined dataset of ~several thousand thermoelectric materials. We have been looking at the distribution of the figure of merit of the materials and taking a closer look at a couple of the best performing theoretical and experimental alloys.
First thing to say is that in my post yesterday I made a mistake! There is an experimental material claiming a higher ZT than the SnSe materials we looked at yesterday. It can be found at https://doi.org/10.1016/j.phpro.2015.05.072 and is a multi-layer system of gold and silver nanoparticles embedded in a silica matrix. This was achieved using a process which blasts the material with ionising radiation - in this case Ar or Si ions. It looks like the author, Daryush Ila, has been working on high energy beam modification of materials since the 90s and is now achieving ZT of 3.2 - 3.6 in these Au/Ag systems. Fascinating stuff, and maybe these techniques might be useful for Dyson spheres in centuries but probably not the scalable, low cost material we're looking for. Perhaps there are some lessons here though, and the type of structures he creates are possible with different routes so we'll mark this for potential deep dive later and carry on with our wide ranging exploration for now.
Another promising material in the dataset is Sn doped ZnO, all low cost elements, readily available, potentially scalable - but its another red herring and the LLM has picked up a paper which doesn't even mention thermoelectrics or ZT values! The material in question is a thin film transparent conducting oxide, so it is a semi-conductor, but with a bandgap ~3.3 eV which is too high for thermoelectrics.
One that is legit is SnS. Turns out this has the same structure as the SnSe discussed in Pt. 1. DFT modelling shows
Figure from https://doi.org/10.1016/j.mtcomm.2020.101167
Another paper (10.1021/acsami.1c24028), again based on DFT modelling, predicts a ZTmax of 1.68 for p-doped SnS at 850K. The authors suggest Na, K, Tl, and Ag as dopants, and provide some guidelines for effective synthesis. Although I haven't done a thorough search yet, the papers I found which have actually synthesised doped SnS don't reach the levels of the predictions with ZT = 0.3 (at 573K) and ZT = 0.38 (at 525K) from https://doi.org/10.1016/j.jallcom.2025.179124 and https://doi.org/10.1016/j.jallcom.2024.176803.
Overall, I feel I am gaining more and more familiarity with the thermoelectric field. I'm seeing more clearly the strengths and weaknesses of the LLM mining approach (I'm still wrestling to clean up the rest of the dataset to have a look at some of the other properties!). Basically I treat it like a person has sat down and read the entire literature then regurgitated it, you get a good broad overview of the shape of a field and capture the salient large scale features but will definitely have some details missing here and there. SnS is definitely one I want to explore more. Oh and I need to learn what Boltzmann transport modelling is🤔.