Hi Van Han,

Are you comparing devices of the same dimensions? (e.g.  not comparing short channel vs long channel)? I believe you are doing a one-a-one comparison.

I think it is normal. I have worked with a TSMC technology (0.18 um), and the ratio was about the same you found.

Best regards,

Hi Dalton,

Absolutely, you're right. I compared one-to-one comparison between P and N (with the same dimension).

Thanks for your kind reply.

Right now, I have another issue. It has been recommended that in analog design, the length of designed devices should be > 2 times of minimum dimension (channel length). However, it causes a reduction in speed, especially of opamp. Have you ever designed opamp with minimum channel length?

Also, for hand-caculation, the lamda (I mean channel length factor) varies according to the chosen channel length. I checked it with many ratios of W/L (under the condition of same length for all cases). The Lamda varies in a little wide range.

So that, the problem here is that, in the opamp design, we can not determine the channel length modulation factor correctly (at least within a narrow range of variation). Do you have any opinions?

Thank you!

Hi Van Han,

One of the reasons for using length > 2 times of minimum dimensions is the process variability. You known that transistor variability (e.g. VTH standard deviation) decreases with transistor area.  Transistor noise (flicker) and other effects (short channel effects) are also worse when using the minimal dimensions.  Moreover, you know that amplifier gain is also proportional to the transistor length. Therefore, in order to achieve a reasonable gain, you cannot use minimal dimensions.

Thus, I have not designed any  opamp using the minimum transistor length so far (basically because the offset voltage / voltage gain). However,  I believe you could use it when you are designing a voltage comparator that does not need precision and charge sharing/store in transistor capacitance is critical. 

Regarding lambda for hand-calculation... you are absolutely right: it varies depending on the transistor length.  You use an approximate value for all values of L in order to have a feeling about it. It is the maximum  you can do.  This situation is similar to VTH (threshold voltage), it also varies according to L. (OK… for large values  of L (above 5 um), VTH is almost constant).  In summary…. when doing hand-analysis… you assume an average value and that 's it.

Best regards,

Dear

too, impacts body effect coefficient