Contamination delay (tcd, also called tmin) is the shortest time after an input changes before an output can start to change. In sequential circuit timing it decides whether the hold time of the capturing flip-flop is met: the fastest path from the launching flip-flop, its minimum clock-to-Q delay plus the minimum delay of the logic in between, must be no shorter than the hold time of the capturing flip-flop. Propagation delay (tpd, or tmax) sets the clock frequency; contamination delay sets hold safety, and no clock change will fix a hold failure.
Key takeaways
- Contamination delay is the minimum delay; propagation delay is the maximum delay. Data sheets often list them as tmin and tmax.
- Setup checks use the slowest path (tpd). Hold checks use the fastest path (tcd).
- Hold condition, zero skew: tcd(launch flop, clock to Q) + tcd(logic) >= thold(capture flop).
- Hold does not depend on the clock period, so a hold violation on silicon cannot be fixed by slowing the clock.
- Positive clock skew (capture clock arrives late) makes hold harder and must be added to the hold requirement.
The original question
Q: What is the significance of contamination delay in sequential circuit timing?
Fact: 70-80% of designers who deal with timing closure daily are unaware of this fact.
What is contamination delay?
So, what is contamination delay anyway? Look at the figure below. tcd is the contamination delay.
When an input of a gate or a block of logic changes, the output does not jump straight from the old value to the new one. For a while it holds the old value. Then, after the contamination delay, the output may start to change and is no longer guaranteed to be the old value: it is "contaminated". After the propagation delay it has settled at the new value. Between tcd and tpd the output is unknown and may glitch.
| Time after input change | Output state | Which delay marks it |
|---|---|---|
| 0 to tcd | Old value, guaranteed stable | Contamination delay ends this window |
| tcd to tpd | Changing, may glitch, not valid | Window between min and max delay |
| After tpd | New value, guaranteed stable | Propagation delay starts this window |
For a block of combinational logic, the contamination delay is the delay of the fastest path from any input to the output, and the propagation delay is the delay of the slowest path. For a flip-flop, the equivalent figures are the minimum and maximum clock-to-Q delay (tccq or tcq,min, and tpcq or tcq,max).
Worked example: tcd and tpd of a small logic block
Suppose a block has two paths from its inputs to its output Y. The gate delays below are illustrative values.
| Path | Gates on the path | Sum of tcd | Sum of tpd |
|---|---|---|---|
| A to Y | NAND2 (tcd 20 ps, tpd 35 ps), then NOR2 (25 ps, 45 ps), then INV (10 ps, 20 ps) | 55 ps | 100 ps |
| B to Y | INV (10 ps, 20 ps) only | 10 ps | 20 ps |
The block's propagation delay is the largest path sum, 100 ps (path A). Its contamination delay is the smallest path sum, 10 ps (path B). The short path B is harmless for setup but it is the one that matters for hold.
Why contamination delay matters for hold time
Without understanding contamination delay you "should not" complete timing estimation of any sequential circuit.
What do you mean by that?
Contamination delay tells you if you meet the hold time of a flip flop. To understand this better please look at the sequential circuit below.
The contamination delay of the data path in a sequential circuit is critical for the hold time at the flip flop where it is exiting, in this case R2.
Mathematically,
th(R2) <= tcd(R1) + tcd(CL2)
Contamination delay is also called tmin and Propagation delay is also called tmax in many data sheets.
Why the fast path is the danger
On a clock edge, R2 captures the value on its D input. At the same edge, R1 launches a new value. That new value races through CL2 towards R2. R2 needs its D input to stay at the old value for thold after the edge. If the new value can arrive sooner than that, R2 may capture the wrong data or go metastable. The earliest the new value can arrive is tcd(R1) + tcd(CL2), so that sum must be at least thold(R2).
Adding clock skew: the full hold check
The formula above assumes both flip-flops see the clock at the same instant. Real clock trees have skew. In this post the sign convention is:
tskew = clock arrival time at the capturing flop (R2) minus clock arrival time at the launching flop (R1). Positive skew means the capture clock arrives later.
With that convention the hold check becomes:
tcq,min(R1) + tcd(CL2) >= thold(R2) + tskew
hold slack = [tcq,min(R1) + tcd(CL2)] - [thold(R2) + tskew] (must be >= 0)
Positive skew delays the moment R2 samples, which gives the racing new data more time to arrive too early. Negative skew helps hold, but hurts setup.
Worked example: hold check with numbers
| Quantity | Value |
|---|---|
| tcq,min of R1 (contamination clock-to-Q) | 80 ps |
| tcd of CL2 (fastest path) | 40 ps |
| thold of R2 | 60 ps |
- Zero skew. Earliest arrival = 80 + 40 = 120 ps. Requirement = 60 + 0 = 60 ps. Hold slack = +60 ps. Passes.
- Skew of +100 ps (R2 clock is 100 ps late). Requirement = 60 + 100 = 160 ps. Hold slack = 120 - 160 = -40 ps. Fails.
- Fix. Insert a delay buffer in the R1 to R2 path whose contamination delay is at least 40 ps, for example a cell with tcd 50 ps. Earliest arrival = 170 ps, hold slack = +10 ps. Check that the buffer's propagation delay does not break setup on the same path.
Notice that the clock period appears nowhere in these numbers. That is why a hold failure found on silicon cannot be cured by running the chip slower.
Contamination delay versus propagation delay
| Contamination delay (tcd, tmin) | Propagation delay (tpd, tmax) | |
|---|---|---|
| Meaning | Earliest time the output may start to change | Latest time the output is settled |
| Path used | Fastest path | Slowest (critical) path |
| Timing check | Hold | Setup |
| Depends on clock period | No (for same-edge paths) | Yes |
| Typical analysis corner | Fast process, high voltage, and the temperature that makes cells fastest | Slow process, low voltage, and the temperature that makes cells slowest |
| Usual fix when violated | Add delay (buffers) on the short path | Shorten logic, upsize cells, retime, or lower the frequency |
Common mistakes to avoid
- Using typical or maximum delays for hold. Hold must be checked with minimum delays at the fast corner. A design that passes hold at the slow corner can still fail at the fast one.
- Ignoring skew sign. Positive skew, as defined above, adds directly to the hold requirement.
- Assuming a shift register is safe. Back-to-back flip-flops with no logic between them have tcd(logic) close to zero, so the whole margin is tcq,min minus thold minus skew.
- Fixing hold before the clock tree exists. Real skew is only known after clock tree synthesis, which is why hold fixing is normally done late in the flow.
Related reading
- Setup time and hold time: checks, examples and fixes
- What a negative hold time means in a library flip-flop
- Slack in static timing analysis, with setup and hold examples
- Transport, inertial and delta delays in VHDL and Verilog
- The Semiconductor Hub glossary entries for hold time and static timing analysis (STA)
Frequently asked questions
What is the difference between contamination delay and propagation delay?
Contamination delay is the minimum time from an input change until the output may begin to change. Propagation delay is the maximum time until the output has settled at its new value. One is a lower bound, the other an upper bound.
Is contamination delay used for setup or hold analysis?
Hold analysis. Setup analysis uses the maximum (propagation) delays of the launch flop and logic; hold analysis uses the minimum (contamination) delays.
Can contamination delay be zero?
For a plain wire it is close to zero, which is why a direct flop-to-flop connection is the worst case for hold. For a real gate or flip-flop it is always some positive value, though it can be small.
Why can a hold violation not be fixed by lowering the clock frequency?
In the hold check for a same-edge path, the launch and capture happen on the same clock edge, so the clock period cancels out. Only the minimum path delay, the hold time and the skew matter.
How do you increase contamination delay to fix hold?
Insert delay cells or buffers on the short path, use slower (for example higher threshold voltage) cells on that path, or reduce the skew between the two clocks. Each change must be rechecked against setup.




Your comments will be moderated before it appears here.