Let’s imagine it’s 1980 again and you have a field full of custom des processors (just doing exhaustive search).
The kind trapdoor I’d want is to probably abandon a given key before having to
Complete all rounds, putting that key at the back of the queue. Given the cost of io in a cluster and the accounting, I want every key derived from that candidate by key schedule to similarly get filtered out of the early trials
So now imagine your field had a special precomputation area with electronics dedicated to the subkey scheduling. So what would that look like?
Look at today’s general purpose CPU pipelining for a good clue on how such custom vlsi was built then.
Also? Since io is the main impeder, think of clustering that is centered on the accounting and queue. One thinks of a giant ram cache… so how do multi cores today share gigs of ram? Probably sane back then…