bias current post simulation difference

I don't know why.
as shown below:
pre sim result:the voltage

post sim result: the volage

the bias size:

layout1

layout 2
.

3Q very much~

Your layout is as asymmetric as your schematic ;-)

The bulk tap is much closer to M7 than to M6. Better move the bulk tap in between.

To achieve a really reliable current ratio, interleave the M6 fingers between those of M7 -- repeating the 2:1 ratio -- + dummies at both ends.

And for your relatively high currents, I'd recommend to add bulk taps at both sides.

as erikl has already mentioned your layout technique makes for some poor matching but poor layout symmetry mostly adds random not systematic mismatch, which is the only thing an RC extractor can provide. Among the things mentioned the only one that the extraction tool will be sensitive to is the added dummies...

Since you operate at 1.2V this was built in a tech below 130nm, there are two major effects that can completely mess up your (systematic) matching: one is well proximity the other is STI stress. The dummies will help with STI stress but not really with well proximity.
To nail down which is causing the problem, look at the extracted netlist and annotate manually some of the extracted parameters for one effect at a time in the schematic. Just do this on M6 since its Vgs drift is sufficient to justify your results. If you have no idea what STI stress and WPE are, you have to read your PDK manual to learn about how they are modeled and implemented in the CDFs

Let us know if we can help more

Thank you very much for your kindly help.

this is a scan from 0.1m to 10mA
and top pic is the bulk current. the line have symbol is post simulation results.

I think the problem in my case is the calibreview,some one said that the calibreview is no acurrace.
Someone also told me that erikl have some documents or threads about this problem. would you please show me?
thank

I fixed my previous post because some of the proposed mechanism made no sense (I have to stop replying before coffee)
calibreview if run correctly is as accurate as any extracted netlist... but there are many ways to get bad extracted views.

if you think this is a calibre problem you can try:
extract only CC-level
extract to a netlist instead of calibreview
post an extracted netlist

Thank you ,I have fixed the problem using spectre format.
also thank to erikl ,both of you are very kindly.