There are some very fast reed relays, but I've never seen one that worked faster than the millisecond range used in microcontroller app's. They exist, for sure, but they have considerable latency, and none of them are latching relays. The problem that a latching relay would address is the problem of interfacing the relay array as a matrix, since the latching relay remains in the last state into which it was programmed. As for the required duration of the pulse to program the relay ... well, a one-shot per relay driver might do the trick but I'd rather handle it in firmware, myself. Frankly, I'm glad I don't have to do this homework assignment. It would require 64 one-shots, and 64 relay drivers ... <sigh> ... and while there might exist situations in which one actually requires 64 relays, I haven't seen one lately.

Maybe, rather than a latching relay, one could use a T-flipflop, e.g. a 74HC107 with J and K tied high, and a driver for each relay, using the 8x8 matrix to drive the clocks to the flipflops. Perhaps that would work with the microsecond timing.

RE