Item | Value |
|---|---|
Clock-to-Q (min) | 0.35 ns |
Logic + wires (min) | 0.05 ns (almost a direct flop-to-flop connection) |
Capture clock skew | +0.38 ns (capture clock arrives later) |
Hold requirement | 0.06 ns |
Hold slack | 0.35 + 0.05 − 0.38 − 0.06 = −0.04 ns |
The new data takes 0.40 ns to arrive. The capture edge comes 0.38 ns after the launch edge, and the hold window lasts 0.06 ns more.
The data arrives 0.04 ns before the window closes. It disturbs the value the same edge must capture.
Read each row as the moment the next word reaches the capture flop. It must land to the right of the shaded window.
Before: the new word arrives at 0.40 ns, inside the window that ends at 0.44 ns, so the old word is overwritten too soon.
After: 0.20 ns of buffer delay moves arrival to 0.60 ns, clear of the window by 0.16 ns.
The clock period is not on this axis at all. The window belongs to the same edge, which is why slowing the clock cannot move it.
Direct flop-to-flop connections are the typical hold risk, since logic delay is near zero. Clock tree synthesis and automatic hold fixing with delay cells handle this in the flow.
Avoid intentionally skewed clocks, such as gated or divided clocks, between directly connected flops.
Only data-path delay or skew changes move hold slack. The clock period does not appear in the equation.
Option | Effect on hold slack | Works? |
|---|---|---|
Slow the clock | None (period is not in the equation) | No |
Insert 0.20 ns of delay on the data path (hold buffers) | −0.04 → +0.16 ns | Yes, and it reduces setup slack by 0.20 ns (large here: 1.86 → 1.66) |
Reduce skew (rebalance the clock tree) | Each 0.1 ns of skew removed adds 0.1 ns | Yes, but a clock-tree change affects many paths |
Put combinational logic that is already needed on the path | Adds its delay | Yes, if the function needs it |
Use a flip-flop with larger clk→Q or a slower cell | Adds a little | Rarely a good choice |
1.86 − 0.20 = +1.66 ns. Plenty of margin remains, which is why hold fixing is usually safe on short paths.