One of the problems with cryptography is, you have timing channel vectors that are so minute and hard to measure that they don't look exploitable; the measurements you'd need to take are drowned by latency and jitter. But someone can come up with a novel application of signal processing --- an advance in a totally different discipline from crypto --- and suddenly an attack that was impractical can be modeled and made practical.
One of the problems with cryptography is, you have timing channel vectors that are so minute and hard to measure that they don't look exploitable; the measurements you'd need to take are drowned by latency and jitter. But someone can come up with a novel application of signal processing --- an advance in a totally different discipline from crypto --- and suddenly an attack that was impractical can be modeled and made practical.
This stuff is just awesome to think about.