This could be an interesting approach. I imagine it would be very difficult to compromise an FPGA processor implementation in hardware, for the same reason that it's tough to have a processor mess with arbitrary software. The analysis tools needed simply aren't possible to implement at the level you're working at.
You could probably insert a hardware trojan that scans for specific FPGA elements and backdoors them. But there's a potential that an unrelated recompile could alter your signature. An alert adversary would be an even bigger problem. You're trying to hit a moving target from a stationary platform.
Unfortunately, commercial FPGAs today are notoriously proprietary. So while this idea may have theoretical merit, it is not currently an improvement in practice.
> still needs manufacturer supplied toolchain for the rest of the steps
This still kills the desired properties of the system. You need open source tools end-to-end, all the way down to the place-and-route system. A backdoor can be inserted at any point otherwise.
Stepping away from the current state of things, competition in the FPGA space still relies heavily on patents and trade secrets. Until that changes, the proposed approach isn't viable.
I wonder if Xilinx or Altera will ever consider this market space interesting enough to pursue. Unfortunately, my gut says no.
You could probably insert a hardware trojan that scans for specific FPGA elements and backdoors them. But there's a potential that an unrelated recompile could alter your signature. An alert adversary would be an even bigger problem. You're trying to hit a moving target from a stationary platform.
Unfortunately, commercial FPGAs today are notoriously proprietary. So while this idea may have theoretical merit, it is not currently an improvement in practice.