I found this summary a bit hard to follow, so I read the 2008 Nature paper the author refers to [0]. (It's behind a paywall, but the supplemental materials in [

I found this summary a bit hard to follow, so I read the 2008 Nature paper the author refers to [0]. (It's behind a paywall, but the supplemental materials in [1] are available.) It seems like Silicium is using an approach that's similar in concept, but there's not a lot of technical detail in this blog post. Presumably that's because they are trying to commercialize their research. The topic is engineering more effective thermoelectric materials -- they generate an electric potential when subjected to a thermal gradient. Here you want a high electronic conductivity and a low thermal conductivity, which is hard to achieve because electrons transport heat very effectively. The approach outlined here is to use an array of doped silicon nanowires instead of a bulk material. They make the nanowires thinner than the phonon mean free path, which essentially confines the dominant heat transfer mechanism (lattice vibrations) in this system to one dimension. The nanowires are thick enough that the nanowire array still has electronic properties similar to bulk silicon, so you can get a lower thermal conductivity without sacrificing electrical conductivity. The programming::materials design analogy at the end of the blog post is interesting. This is actually one of the core philosophies of materials science -- (hierarchical) materials structure (and chemistry) dictates physical properties, and is highly dependent on processing and fabrication routes. You engineer materials with more desirable properties by manipulating their structure. The hard part is learning reliable structure-->properties and processing-->structure mappings. [0]: doi:10.1038/nature06458 [1]: http://www.wag.caltech.edu/home/jamil/Nature-SiNWs-TE-supplement.pdf http://www.wag.caltech.edu/home/jamil/Nature-SiNWs-TE-supple...