[SOLVED] Monte Carlo models -doubt

Hi,

I am using Cadence 6.15 and see four different types of Monte Carlo models:

1. Local

2. Global

3. Local + Global

4. mcminmax

Would appreciate if somebody could give an idea what these models entail?

Gaurav

AFAIR:

1. Local model represents the mismatch parameter distribution for device-to-device (on chip) mismatch for cells (subcircuits) on the same chip

2. Global model represents the typical mismatch parameter distribution for batch-to-batch (process) variations

3. Local + Global describes the superposition of both (for yield estimation)

4. mcminmax contains the (not just mismatch) parameter distribution for a (specific?) min/max range re. PVT variations

Hi erikl,

Thanks for the reply...

But I did not quite understand your explanation for mcminmax

G

panchkapunch said:

... I did not quite understand your explanation for mcminmax

Click to expand...

Mismatch parameter variation distributions (for transistors, resistors, capacitors, ...) for min. and max. (i.e. fixed) Process, Voltage & Temperature parameter combinations ("corners").

But how is this different from local and global ... I am especially concerned because the spread is much larger for mcminmax than local , global or local +global

Is mcminmax the most realistic simulation for yield/spread estimation or is it just overkill?

panchkapunch said:

But how is this different from local and global ... I am especially concerned because the spread is much larger for mcminmax than local , global or local +global

Is mcminmax the most realistic simulation for yield/spread estimation or is it just overkill?

Click to expand...

Local, global or local+global just includes Process mismatch variations, no Voltage or Temperature variations.

mcminmax contains local PVT variations

For spread over 1 wafer you need mcminmax, if you're interested in getting good chips within the full local PVT min/max range.

For yield estimation for 1 single wafer you need the local set if you want to see how many of the chips fulfill the typ. conditions. For the yield of chips which fulfill the min/max (i.e. all) corners' conditions, use both the local+mcminmax models' sets.

For yield estimation for batch-to-batch production you need the local+global set if you want to see how many of the chips fulfill the typ. conditions. For the yield of chips which fulfill the min/max (i.e. all) corners' conditions, use all 3 local+global+mcminmax models' sets together.

Thank you for the detailed response....

So do you mean that in mcminmax, it also varies the set temperaure say set at 100 degrees between different nodes in the ckt. and the applied input voltages from the ideal sources?

panchkapunch said:

So do you mean that in mcminmax, it also varies the set temperaure say set at 100 degrees between different nodes in the ckt. and the applied input voltages from the ideal sources?

Click to expand...

No, this would be totally unrealistic. mcminmax model set allows for analysis parameter combinations between the (foundry defined) local mismatch variations, possible power supply variation (e.g. 1.8V±10%), and possible temperature variation (e.g. -40°C .. +105°C). V&T min/max ranges can be determined by the user.

The MC randomly selected PVT parameters are valid for all devices on the chip, resp. for all chips on the same (local) or all (global) wafer(s).

Ok..so if i am designing my chip to work at only a particular temperature and well-contolled supply/other inputs, local+global montecarlo should be sufficient, right?

panchkapunch said:

Ok..so if i am designing my chip to work at only a particular temperature and well-contolled supply/other inputs, local+global montecarlo should be sufficient, right?

Click to expand...

Yes, if you can guarantee a stable power supply and temperature.