Hello,
here a status report, I was investigating about a dozen diodes.
In brief, there is quite some tension to obtain a reasonably low TC0 point and low noise at the same time.
There are two sets of diodes which differ a lot, but are fairly consistent within each set:
Set 1: 2DW232 from Group Buy
Set 2: 2DW233 from technix
Set 3: fake 2DW232 with internal DO-style diodes
Here a list of typical values I measured.
"nominal" means fixed standard I and T values for comparison,
"TC0" means current for a desired TC0 point around 30+/-10C
Tempco mV/C was measured between 22 and 28C, and current for mV/mA between 9 and 11mA,
both were extremely linear in this range.
mA T Vz Noise uV_pp tempco mV/C mV/mA
Single diode Set1
nominal current and T 10 25 5.4 0.5 +1 +7
TC0 point 2.5 35 5.2 4
Series diodes Set1
nominal current and T 10 25 6.2 0.5 -0.5 +9.5
TC0 point 26 25 6.2 0.3
Single diode Set2
nominal current and T 10 25 5.3 0.7 +1.5 +2
TC0 point 0.5 40 5.2 15
Series diodes Set2
nominal current and T 10 25 6.1 4 +0.04 +5
TC0 point 7.5 40 6.2 5
Single diode Set3
nominal current and T 10 25 5.6 1
TC0 point none found
Series diodes Set3
nominal current and T 10 25 6.4 5
TC0 point none found
Comments:
1) Surprising is the high noise for series configuration of Set 2. Also the low noise for the single diode configuration of the fake diodes.
2) The only combination with low noise and TC0 near room temp is series configuration of Set 1, however at around 30mA. Which is a bit high for practical purposes.
3) This means that for a more reasonable current of 10mA and low noise, temperature stabilization appears necessary.
4) The tempco of the forward diode with -1.5mV/C was lower than what is usually assumed
5) I was trying to see whether one can cancel the tempco at 10mA/25C by an extra LM234 current regulator, along the lines of page 8/fig 15 in
http://www.ti.com/lit/ds/symlink/lm334.pdf by using the forward diode of the 2DW23x.
However it turns out that given the measured values of tempco mV/C and mV/mA, this is physically not possible (it would require negative resistances of R1 or R2).
6) It would be interesting to compare more diodes from different batches.
Edit:
7) BTW for the fake diodes the polarity was reversed, that may explain the discrepancy of data sheets.