The model you started with was the original written by Helmut Sennewald and posted on July 11th 2009. It became widespread because it was discussed in the open forums. The area of transistors Q1 and Q6 were both then set at 14, but that 14 in the data sheet is not actually the transistor area, it is a minimum specification for the ratio of I
set to I
bias.
Eventually the model was modified to fix the transistor area bug and posted on April 7th 2011, in the file archive of the yahoo group “LTspice”. (Hidden from google, you need to be a member),
https://groups.yahoo.com/neo/groups/LTspice/search/files?query=LM334
LM334.sub 1KB Transistor size changed from 14-14 to 16-12 helmutsennewald Apr 7, 2011
This is now the current version of the LM134 / LM334.sub
* Selfmade model of the LM334 from National Semiconductor or ST
* Helmut Sennewald
*
* Pin order: v+ v- R
.SUBCKT LM334 v+ v- R
Q4 N002 N001 v+ 0 pnp1
Q5 N003 N001 v+ 0 pnp1
Q6 R N001 v+ 0 pnp1 16
Q1 N003 N003 R 0 npn1 12
Q2 N002 N003 N004 0 npn1
Q3 N001 N002 R 0 npn1
C1 N002 N004 50p
R2 N004 v- 1µ
R8 v+ v- 1G
C2 v+ v- 10p
C3 N003 v- 1p
C4 R v- 1p
.model npn1 NPN(Is=1e-15 BF=200 TF=1e-9 Cjc=0.5e-12 Cje=1e-12 VAF=100 Rb=100 Re=5)
.model pnp1 PNP(Is=1e-15 BF=100 TF=1e-7 Cjc=0.5e-12 Cje=1e-12 VAF=100 Rb=100 Re=5)
.ENDS
I had reality calibration problems with Helmut Sennewald's old 14-14 model for the LM134 / LM334. When I looked at the LM334 internal design I noticed that the area of Q6 determined the ratio of I
set to I
Q2e, a ratio specified in the datasheet to be about 18. Then I needed to select a Q1 area to bring I
set close to the ideal 273.15uA with an Rset of 226 ohms, at a temp= 0°C to get 1uA / 1°K. I ended up with transistor areas of 15.75 and 11.75. Although by using areas of 15.75 and 11.75 it was possible to get a close to perfect 1uA / 1°K model with an Rset of 226 ohms, that was un-real because the 226 ohm resistor specified in the circuits was chosen from the real world E96 standard series of resistor values which has {…, 221, 226, 232, …}, not as the ideal value the circuit needs for Iset = 273.15uA. Also, the area of the transistors should be integer because the transistor mask pattern was originally simply duplicated to make several transistors in parallel. By using areas of 16 and 12 the model was very real and I then needed to find the ideal value of Rset to give 273.15uA at 0°C.
Rset is the magic number you need to find to a couple of decimal places. It is NOT exactly 226.00 ohms.
Once you have found that Rset closest to 226 I will know your model is working and you can then match the I
95°C with the zero TempCo current sink.
Note: All this assumes that the NPN1 and PNP1 transistor models being used are realistic. There are also two hidden FETs, Q7 and Q8 being used to start the circuit. They source between 20nA and 200nA. They are not modeled here as the numerical spice model starts OK without them and the start bias currents do not appear to be significant to the model.
