According to the linked post there are PQ algorithms that will fit this niche:
> This variety of different trade-offs gives developers a lot of flexibility. For an embedded device where speed and bandwidth are important but ROM space is cheap, McEliece might be a great option for key establishment. For server farms where processor time is cheap but saving a few bytes of network activity on each connection can add up to real savings, NTRUSign might be a good option for signatures. Some algorithms even provide multiple parameter sets to address different needs: SPHINCS+ includes parameter sets for “fast” signatures and “small” signatures at the same security level.
Embedded/IoT is typically slow and small which is not a space PQ fits into.
I also think the article is overly optimistic claiming that ECC is “hard” because of the need for careful curve selection (even though we have very good established curves), but I find it hard to believe that PQ algorithms are immune to parameter selection problems and implementation challenges.
There has been research on the intersection of IoT and PQ signatures specifically at least, e.g. see "Short hash-based signatures for wireless sensor networks" [0] [1]. Unlike SPHINCS+ which is mentioned in the article, if you're happy to keep some state around to remember the last used signature (i.e. you're not concerned about accidental re-use) then the scheme can potentially be _much_ simpler.
The state is enormous. Dedicating megabytes and megabytes to key state is painful. And so is tracking state across components and through distribution channels. If you’re not afraid of that then just use symmetric crypto and be done with it.
To be clear my comment is specifically only relating to signature schemes, not encryption.
> The state is enormous
The scheme I linked to points towards efficient "pebbling" and "hash chain traversal" algorithms which minimize the local state required in quite a fascinating way (e.g. see https://www.win.tue.nl/~berry/pebbling/).
> tracking state across components and through distribution channels
Assuming you have reliable ordering in those channels I don't see how the stateful nature of such schemes makes it hugely more complex than the essential hard problem of key distribution.
