Required time
Required time is the deadline — the latest moment a signal is allowed to arrive at a node and still be captured correctly, like the last possible second you can mail a tax form before the post office closes. While arrival time is computed by walking forward from startpoints, required time is computed by walking backward from endpoints: the tool starts at a capturing flip-flop, subtracts the setup time and any clock uncertainty from the arriving clock edge, and pushes that deadline back through the logic.
Required and arrival times meet head-on at every node, and their difference is slack. For a setup (max-delay) check the required time is essentially the clock period minus setup time; for a hold (min-delay) check it flips to a minimum — the data must arrive no earlier than a floor, so the old data isn't overwritten before it's latched. Constraints like input/output delays and false-path or multicycle exceptions are precisely the knobs that reshape these required times.
A subtle trap: required time depends on the capture clock, not the launch clock — for paths that cross between two clock domains or two clock edges, getting the right capturing edge is half the battle of getting timing right.