How to fix Bandgap Iref?

[KGF KING]

Hi All,

I’m trying to design a sub-1V bandgap that generates constant Iref = 0.5uA and Vref = 0.5V. I have attached my circuit below. The problem I’m facing right now is I’m able to generate constant Vref approx. to 0.5V but my Iref is looking like below(waveform attached) at VDD=1.2V. Initially I tried to swap the op-amp with ideal because doubt of my input offset one but seems like the op-amp is not the case. I don’t know how to troubleshoot the root cause of this. Can anyone help me how to troubleshoot this? Please let me know if need more details of the circuit.

 

[dick_freebird]

Geometries can't be sanity-checked because
not shown. Your iref drift w/ temp appears
sane-ish (PTAT) but slope seems excessive
and nonlinear, and value rather low (sub-uA).
That's gonna want a lot of resistor.

Begin with verifying that the loop is really
closed, to small Vio. With no explicit startup
(the op amp that controls bias rack, is fed by
bias rack) this might be some other case.

 

[KGF KING]

Geometries can't be sanity-checked because
not shown. Your iref drift w/ temp appears
sane-ish (PTAT) but slope seems excessive
and nonlinear, and value rather low (sub-uA).
That's gonna want a lot of resistor.

Begin with verifying that the loop is really
closed, to small Vio. With no explicit startup
(the op amp that controls bias rack, is fed by
bias rack) this might be some other case.

Can you explain more briefly? Sorry couldn't catch up your explaination

 

[abab1394]

Resistor Maybe has large temp coeffcient, check the model of it for TC1.

 

[KGF KING]

Resistor Maybe has large temp coeffcient, check the model of it for TC1.

the highlighted is my resistor TC1. is it considered high? My R4 resistance value at -40C = 1.6M; 25C = 1.03M and 125C = 0.57M

R1 = 1.51M
R2 = 1.01M
R3 = 0.26M
R4 = 1.03M

 

[dick_freebird]

Those are large values for most resistor options in foundry flows. The TC seems high to me for a Pbase resistor. Lighter doped gives better "signal" but tends to be less linear. Getting a good positive TC term can be tougher with "better" resistors like SiCr. Half my last VREF chip area was SiCr. But way less bowing than the comp and matched the die outline so all good.