how to save a bad bandgap after tape out?

as we konw, the bandgap is very important to many analog circuit. If we find the output of a bandgap after tape out differ largely from the output that we get through simulation, what are the main or the most possible reasons? what can we do to get a steady output from this bad bandgap in chip afer tape out? or how can we repair this bandgap after tape out?

thanks in advance.

As I know, the common bandgap circuit need trimming circuit.
The circuit offset, layout mismatch, process variation, these factors can affect the precision of bandgap voltage.

muldersun said:

As I know, the common bandgap circuit need trimming circuit.
The circuit offset, layout mismatch, process variation, these factors can affect the precision of bandgap voltage.

Click to expand...

would u please give me more details about trimming circuits?

thanks

Another possible solution is to have the option to bypass the bandgap and apply an external reference...

Hello:

More information might be necessary, like:

(1) does it start up correctly?

(2) is the output voltage stable? noisy?
what's the behavior if we change VDD?

(3) is the output voltage same for all samples?

nige said:

muldersun said:

As I know, the common bandgap circuit need trimming circuit.
The circuit offset, layout mismatch, process variation, these factors can affect the precision of bandgap voltage.

Click to expand...

would u please give me more details about trimming circuits?

thanks

Click to expand...

Attached file is simple principle circuit. It is only for trim-up.
Original value is lower than you want.
Example: The bandgap voltage is 1.25V, the original value maybe 1.15v.
Red rectangle is trim PAD. Red line is top-level metal we called burned-fuse.
Burning the fuse between 1 and 2, you can get one step value.
If you set one step value is 25mV. Now the bandgap voltage is 1.15v+25mV.
Following this method, you can get 2^3 case.
The maximum bandgap voltage is 1.15v+7*25mV.
The step value & trimming-case should be depend on your accuracy.
If you add another trim PAD, you can get 2^4=16 case. So, you can reduce the step value or increase your range.
This is the common use circuit. Of course, you can use trim-up & trim-down circuit.
Or the trim PAD is isolated from the resistors.

sorry,what's the meaning of your "tape out"?

The sources of fail of your bandgap can be different. Try to analize layout. Is there critical devices are perfectly matched? How is technology reliable to mismatches? Did you make a Monte Carlo analysis with process deviation and mismathces? If not, try to do and see is there OK or not. Maybe problem i start-up circuit. Try to increase VDD rise time on your measurement testbench to 10-100 us.
Good luck!!!

I am eager to know what's the answer to the third question of JCPU. my output voltages are not only the reference voltage I wanted ,but different from each other.For example, i designed a bandgap reference for 1.2V, and my simulated datas say it done well,but after tape out (does it mean fabricate?) ,some is 1.1v ,and some is 1.3v, some other is 1.4v,and so on .The outputs are very dispersed. So where is the fault ? Thanks.

JCPU has some excellent questions. I've designed voltage references. To get a tight reference, you must trim. One way was described above. This can also be done at the bottom of the loop, trimming the other way. You wouldn't want to do both.

I would say that in vipwl's case, you are seeing variability in fab. SPICE only will tell you what you ask it. It assumes everything matches EXACTLY. It assumes the temperature is EXACTLY the same for each element in the circuit. When I see that level of variability, I would look at the layout. If you are using minimum devices throughout the circuit, this is the most likely candidate. The circuit works BECAUSE of match. If the match is poor, the performance is poor. In this case, the repeatability.

If it was a temperature problem, you would have a shift in output, but it would be uniform, if the design were robust.

Tapeout is a term we use for when we took our GDS data off the Calma station and sent this to someone to cut RubyLith. It then became taking the GDSII data from Chipgraph/IC Station or Virtuoso and sending it to the mask maker to have the masks printed. Now we just ftp the file.

Thank u ,dear upstairs