avlsi wrote on Sep 13th, 2006, 7:13pm:Hi
I have thought to use an optimum Cc. I am using now 5p F.I want to restrict myself to that,since it affects my area and noise.
Good to see that you choose the Cc based on noise.
Quote:I calculate the affect of pole and zero at second stage.
Second stage gm = 10 * first stage input gm.
This will affect my current consumption.
You should also take a look at the paper by Ahuja (JSSC, Dec. 83 I think). Since you mention that slew rate is not important to you (and I hope you have investigated this well), you can use this scheme. There are 2 drawbacks of this scheme:
1. An extra stage is added => more current, especially is slew rate matters (not in your case)
2. Systematic offset introduced as two current sources fight each other in the intermediate stage.
However, it should allow you to improve the stability of your opamp quite a bit for lower current and will also give a better opamp. If you are making a fully differential design, then the systematic offset issue may not be as severe. The big advantage here is the elimination of the RHP zero.
Quote:I have been seeing many people,who just think analog is like a trial and error procedure. They know the theory but really cannot tell you ,why they took this value.
A pity! How the pioneers would cringe. Let us remember that a couple of decades ago, there was only one kind of circuit design, and people would have laughed if you told them that this sort of design (now analog) could not be done systematically and optimally, even if it was not possible to automate it.
Quote:Can u tell how can we plot a graph like this - Log(Av) Vs Log(Ids)
I want to do this to find optimum value of tail current.
I know wht is the best L I can use to get some specifiec gain. I plotted gain for different L.
Thanks in advance
Depending on the tool you use, you can usually just switch axes to log scale, e.g. Spectre allows plotting of gain in dB, and if I remember well, so does HSPICE.
Just a word of caution, if you sweep L or any other device parameter using an automated sweep, you may get incorrect results. I have not done this for a while, but the last time I did it, the results were wrong because tools don't handle the model binning correctly when you sweep a device parameter, and surely, things like source/drain diffusion parasitics are not correctly scaled. So, you should physically change the device properties, re-netlist, and run simulations, or else create two separate devices with different L, and simulate them in parallel as separate instances.
Regards
Vivek