Chip development costs really vary. Like you said, if they just did a process geometry shrink, then effort is small. If they are designing a lot of new stuff, the effort is big.
The big difference is the tape out costs. 28nm is sub $2M, where as 16nm is around $8M now.
The other issue with HF seems to be whether they paid their design contractors/vendors. Other than Simon, I am not sure they had any chip designers on staff. I had heard that they outsourced much of this. So maybe the IP is locked with different contractors.
I do not know what HF did or didn't do or what they had planned to do.
I can say that test chips are typically used to prove out sections of a new chip/new technology. It is a small fraction ($100K range vs $Ms) of what a full tape-out costs and is used typically to mitigate risk. It isn't typically viewed as an overrun, but more as money spent to reduce risk.
The cost overruns in 16 NM come from the test wafers?
From a business or bank side when a developer asks for x dollars and you pass it then ask for more cash was my thought process.
I was reading Bitfury scraped 16 NM because they ran into problems.
Its about 80% more to develop 16NM compared to 28 but doesn't include what a developer can import from a previous chip?? It depends on how far along HF got.