bandgap die by die deviation??

I use pmos+noms cascade(non-opamp structure)
+BJT 1:8 +high R poly(width 8um) with R1 l=32um*6, R2 l=32um *29
to implement bandgap.

mos, R and BJT matching consider

but die by die deviation is too large
test 35ea chip
avg of output voltage ~ 1.18V
but 2ea > +-7%
6ea > +-5%
14ea > +-3%

our target is < +-3%
**************************

The process deviation is mainly caused by resistances in bandgap reference circuits.Either you have chosen wrong type or your process is very sensitive to die by die deviation in resistances.

Why don't you adopt the opamp structure? and do you add a buffer before deliver the bandgap voltage forward?

Thanks for your reply

1.It's a 0.5um process. Many designers choose high R poly for bandgap, even only
use width=3um. I use 8um, it's very safe for matching. High R poly spice model
shows +-20% deviation in S and F corner, only affects <1.5% Vref in hspice sim.
Simple PTAT bangap formula also shows R2/R1 ratio may affects Vref value,
not value of R. I copy formula attached below,
Vref=VBE+ (R2/R1)*Vt* ln

2. OP structure needs ~4p cap for ~60 degree compensation in our design.
layout area over our spec. And our VDD min~ 2.4V.
4uA design in our bangap out path (PTAT+R2+VBE).
Thus, I measure Vref out directly


How about bandgap die by die deviation in your design?
I read some article metioned trim, but not talk about how to do it.
I supposed that they said trim R.
But Vref=VBE+ (R2/R1)*Vt* ln formula shows if we enlarge the ratio of R2/R1.
Yes, Vref increase, but, temperature compensation lost, too.

phononwang said:

I read some article metioned trim, but not talk about how to do it.
I supposed that they said trim R.
But Vref=VBE+ (R2/R1)*Vt* ln formula shows if we enlarge the ratio of R2/R1.
Yes, Vref increase, but, temperature compensation lost, too.

Click to expand...

for a bandgap design , i did some time back :
as bandgap dc output depends on r2/r1.
these resistors are highly prone to variations, so we connected extra poly res finger in series with r2 and then shorted its terminals at top level metals(m5,m6)
being shorted it was acting as zero ohm in sims,
after fabrication if it is found that r2 has decresed due to variations then using laser trimming of wafer we just melt m5/m6 and that resistors becomes part of r2
similarly we used a opened resistors that can be shorted later after fabrication to decrease value of R2 ..
this is what i know abt trimming in bandgap.

have u done monte carlo simulations to see effects of variations on bandgap output?