Fixing Setup & Hold Violations

Re: how to fix hold violations

how does adding buffers fixes hold time?

if negative values occurs, consider it as zero.

Re: how to fix hold violations

gourav.panda said:

how does adding buffers fixes hold time?

Click to expand...

For hold , we would like to delay the data path, so that the data doesn't reach too early. So we use delay buffers to fix hold.

Re: how to fix hold violations

I want to know how buffers are used to delay the signal.buffers are only used to restore the signal level as far as i know.

Re: how to fix hold violations

gourav.panda said:

I want to know how buffers are used to delay the signal.buffers are only used to restore the signal level as far as i know.

Click to expand...

Hi Gourav

Yes buffers are used to restore the signal, but when you use any cell, be it buffer or inverter, these cells have some delay associated with them. So when you add any cell to your path the path delay increases. Since buffer or 2 inverters are the simplest form of cells, buffers are used. For fixing hold, there are special delay buffers present in your library, they are used.

If you look at Hold time equation Tcq + Tcomb> Tskew + Thold. Hold violation occurs due to faster datapath. If we increase datapath delay. It will solve hold violation and adding buffer in datapath will increase datapath delay.

https://www.edaboard.com/threads/227915/#2

yadavvlsi said:

If you look at Hold time equation Tcq + Tcomb> Tskew + Thold. Hold violation occurs due to faster datapath. If we increase datapath delay. It will solve hold violation and adding buffer in datapath will increase datapath delay.

https://www.edaboard.com/threads/227915/#2

Click to expand...

By data path delay you mean Tcomb right? but consider the setup time equation Tclk + Tskew>Tcq + Tcomb + Tsu .So an increase in Tcomb to fix hold time will essentially make the circuit operate at lesser freq( because of setup time). So is there a better way to optimize both setup and hold time considerations?

You can increase datapath delay (insert buffer) only you have enough setup margin else you will end up having setup violation. If your path is having hold violations and setup critical also, than buffer location becomes very important.

but as far as i know only setup time violations are taken into cosideration while calculating the clock for the circuit.

Yes, because setup depends on frequency and hold is independent of frequency.

But i suppose in practice we always set a target clock frequency while designing the chip itself which means that Tclk is fixed. since Tcq and Tsu are also fixed at the synthesis stage. The only parameters that we control are Tcomb and Tskew. We would like to have high Tcomb for hold time check and high skew for higher running freq without violating setup time. But it is important to note that (Tcomb)max is limited by setup while (Tskew)max is limited by hold time. The interplay between these two has to be controlled carefully to run at the right freq( without violating setup) and ensure proper data is sampled(without violating hold).
please let me know if my understanding is right

I hope this is the right to place to ask this question .

In a reg to reg path , if i have setup violation then where should i insert buffer , at launching flop or at capturing flop ? and why ?

Setup time:-The minimumm amount of time taken by the data signal that should be held steady or stable before the arrival of active edge of clock
equation:-Tcq+Tcombi<=Tcp-Tsetup
Hold time:-The minimumm amount of time taken by the data signal that should be held steady or stable after the arrival of active edge of clock
equation:-Tcq+Tcombi>=Thold
If data arrival time(Tcq+Tcombi)is greater than data required time(Tcp-Tsetup) then setup is violated....,we can fix the setup violations can be fixed by following techniques
1.cell sizing increase(launch flop)
2.combi delay decrease
3.Net length decrease
4.Clock period increase
If data arrival time(Tcq+Tcombi)is less than data required time(Tcp-Tsetup) then hold is violated....,we can fix the hold violations can be fixed by following techniques
1.Cell down sizing(lunch flop)
2.net length increase
3.combi delay increase

jeet_asic said:

I hope this is the right to place to ask this question .

In a reg to reg path , if i have setup violation then where should i insert buffer , at launching flop or at capturing flop ? and why ?

Click to expand...

Buffers are generally used to correct hold violations. To correct setup violations: consider Tclk> Tcq + Tcomb + Tsu - Tskew

suppose ur Tclk is fixed:

method 1: Try to decrease your Tcomb by pipelining etc
method 2: You might want to increase Tskew (By adding buffers in the capture path. But then again make sure that the increase in Tskew does not result in a hold violation since Tskew < Tcq + Tcomb - Thold

But i would suggest you try method1 before attempting method2