Arrival Time, Required Time and Slack

Every line of a timing report comes down to one number: slack. Slack is the spare time. Imagine a train that arrives at the station at 8:50, when the connecting train leaves at 9:00. The spare time is 10 minutes. In a chip, the signal is the first train, the clock edge is the departure of the second one, and the spare time is the slack. STA finds the arrival time by walking forward through the design from where signals start, and the required time by walking backward from where they must be caught.

7 min read | Intermediate

What You'll Learn

• Compute arrival times by propagating delays forward through a path
• Compute required times by propagating backward from the endpoint
• Calculate setup and hold slack and interpret positive, zero and negative values
• Read the worst negative slack (WNS) and total negative slack (TNS) of a design

Arrival Time: Walking Forward

The arrival time at a pin is the time at which a signal transition reaches it, measured from the reference clock edge. STA starts at the startpoints and adds delays as it moves along the arcs:

• At an input port, the arrival time is the clock edge time plus the input delay you constrained.
• At the clock pin of a flip-flop, it is the time the clock edge reaches the pin. The clock-to-Q arc then adds the launch delay.
• At every other pin, it is the arrival time at the previous pin plus the delay of the arc between them.

When several paths merge at a pin, STA keeps the largest arrival time for a setup check, because the latest signal is the dangerous one. For a hold check it keeps the smallest. Rising and falling transitions have separate arrival times.

Required Time: Walking Backward

The required time is the latest time at which a signal may arrive at a pin and still meet the check. It starts at the endpoint:

• At the data pin of a capture flip-flop, the setup required time is the arrival time of the next capture clock edge minus the library setup time of the flip-flop.
• At an output port, it is the capture clock edge time minus the output delay you constrained.

STA then moves backward and subtracts the delay of each arc. At a pin that drives several paths, it keeps the smallest required time, because that is the most demanding one. The picture below follows one path from input c through the XOR gate to the data pin of r2, with the clock period at 2.0 ns.

The arrival time starts at 0.4000, the input delay, and grows by the 0.1176 ns delay of the XOR gate to 0.5176. The required time starts at the endpoint, 2.0000 - 0.1174 = 1.8826, and shrinks as it moves back to 1.7650 at the input.

Slack

Slack is the difference between the two: the time the signal is needed minus the time it arrives. For the hold check the sign is the other way round, so that a positive slack always means good news:

Setup slack = required time - arrival time
Hold slack  = arrival time - required time

For the path above, the setup slack is 1.8826 - 0.5176 = 1.3650 ns. It is the same at every pin along the path, as the picture shows, because moving a pin along a path adds the same delay to the arrival time and takes it from the required time.

Slack of zero or more means the check is met. Negative slack means a violation. A slack of exactly zero is met, with no margin left.

Shrinking the clock period to 0.5 ns does not change the arrival time. It moves the required time earlier, from 1.8826 to 0.3826, and the slack turns negative at -0.1350 ns.

Several Paths Into One Endpoint

Where paths converge, the pin takes the worst slack of all the paths through it. The data pin of r2 is reached from two startpoints, and each path has an earliest and a latest arrival.

The latest arrival (0.5176 ns, from c) feeds the setup check and the earliest arrival (0.3657 ns, from r1) feeds the hold check. One endpoint can therefore have different critical paths for setup and for hold.

WNS and TNS

Designs have thousands of endpoints, so reports summarize the situation with two numbers:

• WNS, worst negative slack: the most negative slack in the design. It tells you how far the worst path misses.
• TNS, total negative slack: the sum of the negative slack over all violating endpoints. It tells you how much work is left.

At a 0.5 ns period, the four endpoints of the demo design have a setup slack of -0.4674, -0.2956, -0.1361 and -0.1350 ns. The WNS is -0.4674 and the TNS is the sum, -1.0341. The report prints the worst slack for both check types:

worst slack max -0.47
worst slack min 0.42

Here max is the setup check and min is the hold check. The setup check fails while the hold check still has 0.42 ns of margin.

Common Mistakes