why we don't use simple diodes to build BGR

[nus_lin]

dear guys, the bjt used in bandgap reference is actually diodes, so why don't we just use nwell-diodes, or n+p- diodes?

 

[dick_freebird]

Because BJTs diode-connected are log-linear over a much
wider current range, which is good since you operate
the diodes at discrepant current densities.

 

[nus_lin]

here is my friend's comments:

in Wiley Sansens design essentials (slide 165, page 457) :

"A bandgap reference voltage uses a bipolar transistor, connected as a diode. Its current-voltage
expression is then quite accurately given by the exponential. A real pn-junction may have a
coefficient of 1.05–1.1 in front of the kT/q; a bipolar-transistor connected as a diode does not"

---------- Post added at 02:51 ---------- Previous post was at 02:48 ----------

Because BJTs diode-connected are log-linear over a much
wider current range, which is good since you operate
the diodes at discrepant current densities.

hi, freebird, your answer does not satisfy me, can not explain why simple diode is inferior to bjt in terms of log-linear.

---------- Post added at 02:52 ---------- Previous post was at 02:51 ----------

also according to my friend's comment:
I don't know why the diode should have more variation than the diode connected bipolar, which is actually nothing else than a diode, too

---------- Post added at 03:17 ---------- Previous post was at 02:52 ----------

i figure out a possible explanation.

if we look at their voltage drop, you will see that Vbe(bjt)<Vbe(Nwell diode)<Vbe(N+p- diode), under the same current density. if you look at the serial resistance, bjt should be better than both nwell diode and N+p- diode.

i think beta helps a lot. R_eq_c=R_b/beta, where R_b is the base resistance, R_eq_C is an equivalent resistor in parallel with the resistance of the collector itself, which should be much smaller than R_b, due to the large area of collector.