However for the most part, the most 'interesting and exciting field' is very subjective and varies from person to person. I think you won't be able to get a definate answer on this because everyone thinks their field is the most interesting (and for good reason too - otherwise they shouldn't be in it!).
* Large mathematical theorems were proved in Coq (Feit Thompson Theorem)
* An optimizing C compiler with correctness proof was created (http://compcert.inria.fr/)
If this stuff gets to mainstream and I think, it will get there in 5-10 years, we will have unprecedently reliable and secure systems.
I don't disagree that there are fewer ideas, but that's because we need new ideas, including new architectures. Having spent quality time on a number of novel supercomputer architectures that seem to have been forgotten in the rush to turn GPUs into short vector SIMD-ish successors to Cray XMP/YMPs, I don't think we've looked at all the viable alternatives, at least not in a long time. For example, I'm clear on why memory-based vectors like the Cyber-205 faded at the time, but it's not clear the same limitations fundamentally still exist (especially with L2/L3 caches bigger than the 205 main store now). I think that's interesting...YMMV.