@User yes there are many such questions. Ideally, the array would have all transistors present, that way, you could change the ROM programming with a single mask change, i.e. if you had to revise the silicon. But for this exercise (which is just me mucking around), it might be better just to leave the transistor out for unprogrammed bits.

Yeah, I wasn't really considering this since we wouldn't really really respin a single mask for programming. There are numerous options using metal, contacts, or deletion mode threshold implants to program. I don't think we have the later though.

Yes many things are possible. I think simple via is easiest right now. For my target design, I would like a reusable filter block, with ringbuffer (SRAM), coefficient LUT (ROM), and multiply-accumulate block. I can design the core so that a special ROM address contains configuration information about the filter, such as the number of coefficients etc. In this way, the block can be synthesized and verified once, (and taped out, but that will probably never happen), and reused with different programming, to build up a DSP pipeline (for interpolation / decimation).

It's not a bad idea and is definitely something that previous ROMs do. I suggest using the parameter to the ptx. Also, it is likely we can tape out!