More voltage references - die pictures

[Mickle T.]

Thaler VRE305K magic:

[doktor pyta]

We were missing You, Михаил

[Noopy]

We had no reference for a long time... 




Let´s take a look into a SZY22, a reference built by the "Werk für Fernsehelektronik".
With a current of 5mA it generates a voltage of 8,4V +/-0,4V.




The tempco depends on the sorting:
SZY20: black dot: 100ppm/K
SZY21: yellow dot: 50ppm/K
SZY22: blue dot: 20ppm/K
SZY23: red dot: 10ppm/K




Datasheet states three diodes, one acting as zener and two acting as "normal" diodes. The diodes are chosen in a way that the positive tempco of the zener compensates the negative tempco of the "normal" diodes.






The diodes are potted in epoxy. In the epoxy you can spot a board carrying the diodes and a second board isolating the first board against the metal case.




400°C later => surprise: There are only two diodes. 
It seems WF has optimized the reference so the tempcos compensate each other with only two diodes.
On the diodes there is a sticky residue. Probably they were protected with a different substance.




After cleaning we find two glass diodes.






The diode is a n-doped silicon plate with a metal part alloyed onto it. I assume that is Al which gives a p-doping.




While decapping the tin got liquid. In the second diode the tin now is located in the active area but i looks like the second diode had a similar construction as the first one.


https://www.richis-lab.de/REF15.htm

 

[Gyro]

Datasheet states three diodes, one acting as zener and two acting as "normal" diodes. The diodes are chosen in a way that the positive tempco of the zener compensates the negative tempco of the "normal" diodes.

...

400°C later => surprise: There are only two diodes. 
It seems WF has optimized the reference so the tempcos compensate each other with only two diodes.
On the diodes there is a sticky residue. Probably they were protected with a different substance.

One of the diode packages is probably a temperature compensated reference zener - something like a 1N827-9 with an internal series compensating diode junction.

[Noopy]

One of the diode packages is probably a temperature compensated reference zener - something like a 1N827-9 with an internal series compensating diode junction.

But then you would not need a second diode. And both diodes look the same, you can't spot a special construction.

Update: Of course you would expect a second diode for 8,4V. => 8,4V zener usually needs two diodes for compensation of the tempco.
But nevertheless you can´t spot a special construction of the diodes.
And as I wrote later perhaps WF was able to compensate the tempcos with only one zener and one "normal" diode.

[iMo]

FYI - attached a 1997 pdf how the 1N829 diodes were made..
I have got here several TESLA KZZ82 (former Czechoslovakia) produced temperature compensated zeners.
For example KZZ81 claims "<10-7/degC Vref TC" within +/-1% of nominal Iz (20-100mA) in entire 0-50degC temp range..
http://teslakatalog.cz/KZZ82.html
Too big and heavy to be sent in an envelope to Noopy for an analysis..

[Noopy]

Some more words about the number and types of diodes:
Normally for 8,4V you use one zener and two "normal diodes" because the higher zener voltage comes with a higher tempco that needs two normal diodes for compensation.
But the tempco also varies with current. Perhaps the SZY worked just fine with one zener and one normal diode.

By the way: The SZY datasheet states that all three diodes are zener diodes.

[RoGeorge]

I wonder if the short circuited diode can be restored by melting the alloy again, but using a soldering gun, heating the terminal with the diode kept inside the U shaped hitting wire (because the solder is pushed by the strong EM field to the tip of the heating loop).   

[antintedo]

400°C later => surprise: There are only two diodes. 

Have you tried less brutal decapping methods? Soaking in DMSO at 60-150°C for 0.5-3h destroys most epoxies and many other coating/potting compounds. They can be cleanly picked apart with tweezers afterwards. Bond wires won't survive, but nothing inorganic should be melted or broken.

[Noopy]

Have you tried less brutal decapping methods? Soaking in DMSO at 60-150°C for 0.5-3h destroys most epoxies and many other coating/potting compounds. They can be cleanly picked apart with tweezers afterwards. Bond wires won't survive, but nothing inorganic should be melted or broken.

Thanks for the hint. Up to now I have kept distance to chemical methods because most of them are very unhealthy or you need very very special solvents which normal people can´t buy.
DMSO seems to be quite uncritical and you can buy it on ebay. Sounds good so far.

Nevertheless DMSO has to compete with my "ofen process" which is quite fast while with DMSO I will have to cook some hours.

[serg-el]

Sold in pharmacies.
Димексид, ДМСО, Brosorb, Damul, Deltan, Demasorb, Demavet, Dermasorb, Dimethylsulfoxid, DMSO, Dolicur, Dolocur, Dromisol, Durasorb, Hyadur, Mastan, Somipront, Syntexan .
https://pharmacycode.com/Somipront.html

[Noopy]

The AD1403 is a low cost 2,5V-reference. Supply voltage needs to be 4,5V and can go up to 40V. The better graded AD1403A guarantees 2,5V+/-10mV. Tempco is typically 10mV/°K. It can source 10mA.




The die is 1,5mm x 0,9mm.
There is a testpad connected to the bondpad in the left corner. Would be interesting why that testpad was necessary. It would have been possible to connect the potential at the bondpad. 
The big orange resistors can be laser trimmed.




In the upper right corner there are two testpads that are just connected to a orange square. You often see such structures on dies with tunable resistors.
You can´t see any tuning. That´s interesting. Normally the laser tracks are visible. The tolerance of semiconductor resistors are quite high. I´m pretty sure the AD1403 needed some trimming.




Even more interesting is this "residue" on two smaller resistors which are built with a different material. That looks quite like laser trimming. You can even see some trenches. But usually you only trim the resistors that are designed for trimming. And tunable resistors are usually bigger to make trimming easier. Strange...




Two test structures. The upper structure is a pinch resistor, the lower one is a npn-Transistor.
Interesting point here: On top of the p+ doped trench that isolates the active areas is a red layer. The red layer is the p doped base material. It seems like Analog has used the same technique as National used in the output transistor of the LM306 (https://www.richis-lab.de/Opamp09.htm). The less p doped base layer gives you a higher breakdown voltage between the active area and the substrate.




There seems to be an active element under the output bondpad. Perhaps some overvoltage protection...




Identifying the different components is no bigger problem but there are some interesting structures...




The cyan parts are the bias circuit. Q13/R14/R15 are taking the output voltage and generating a reference current. R13 is for proper start-up. R14/R15 are the small tuned resistors. That makes some sense. There is a current regulation around the bandgap cell but it´s probably a good thing to adjust the supply current to a value near the optimum.

The green area is the bandgap cell with the transistors Q6/Q7 having different areas. R5 and R6 are tunable to adjust the tempco to an optimum. With the help of the testpad 1 you can meassure the voltage while tuning the circuit.

Q2/Q3 ist a current mirror which wants equal currents in the two legs. If there are different currents Q9/Q10 are adjusting Q11 so the voltage at the bandgap-reference base goes to a level where the currents are equal and the bandgap-reference works as intended.
There are two interesting facts: The collector currents of Q9 and Q10 are directed into the bandgap-reference. And there are the two pnp-transistors Q4/Q5 which act like small amplifiers for the bandgap currents.
The whole circuit looks like they had put a lot more effort in than it looks like in the first place. I´m sure there are some smart tweaks hidden in the schematic.

The red part is the regulation loop and the output stage. R10/R11 gives you the voltage you like. Because of that R10 and R11 are tunable.
R10 adds some drift effects to the reference voltage. That drift is compensated with R3/R4. R3 is shorted with the metal layer. Perhaps there was an option to adjust R10/R11 to a different output voltage that would have needed R3.




Here you can see the bandgap transistors with an area of 8:1. The bigger transistor is built with two four-emitter transistors surrounding the small one-emitter transistor.
In the red p-doped areas connected to ground there are the two pnp-transistors Q4/Q5 interacting with the collector area of the bandgap transistors.




R13 is quite interesting. The collector area is shaped into a stripe building R13. 


https://www.richis-lab.de/REF16.htm

 

[magic]

R13 is quite interesting. The collector area is shaped into a stripe building R13. 

So it's an epi-FET.

It seems you did a pretty good job with the schematic but you guessed transistor numbers incorrectly
(And missed the current limiter and compensation cap.)

[Noopy]

R13 is quite interesting. The collector area is shaped into a stripe building R13. 

So it's an epi-FET.

 

It seems you did a pretty good job with the schematic but you guessed transistor numbers incorrectly
(And missed the current limiter and compensation cap.)

Thanks! 
Where did you find that schematic?
I'll have to do an update! 

[magic]

From the patent mentioned on the first page of the datasheet: US 3,887,863

It covers the Brokaw cell in general and also has the schematic and description of this exact imlementation.

BTW, I can't actually find any capacitor connected to the collector of this central bandgap transistor. Also, 100pF would be a rather huge cap for an IC
But the limiter obviously is there, near R9.

[Noopy]

 

Thanks!

I will take a closer look and do an update. 

[Noopy]

Big update for the AD1403:




The circuit has a hidden feature described in the patent the datasheet refers to.
The bias generator is tuned to generate a current equal to the current flowing through each leg of the bandgap reference. This current flows through the yellow path. Furthermore the yellow path is equal to "half" of the path Q4/Q13/Q14. Via Q11/Q12 that leads to a constant collector potential at the transistors Q2/Q1 which then act more constant independent of temperature and bias. Freaky circuit... 




I found the 100pF capacitor! It is built with the buried n+ collector connection of Q1. You can spot the buried layer at some places. The area is quite big to get the 100pF. The second electrode is the substrate.




A quite integrated overcurrent protection. A current flow over the base area can lead to a voltage drop at the base contact in the upper area where the emitter is placed. In an overcurrent event the transistor is activated conducting current through the collector area.


https://www.richis-lab.de/REF16.htm

 

[magic]

Hmm, I think you may be right abut this buried layer. I completely haven't though of such possibility. I wonder what happens if isolation diffusion is produced over a burried layer: will it destroy it, or produce a capacitor with two layers and twice capacitance.

The current limiter is pretty standard. Off the top of my head, LM358 uses the same trick. Except that the resistor/base is also simultaneously the emitter of a lateral PNP

[Noopy]

Hmm, I think you may be right abut this buried layer. I completely haven't though of such possibility. I wonder what happens if isolation diffusion is produced over a burried layer: will it destroy it, or produce a capacitor with two layers and twice capacitance.

In the books you see both, isolation diffusion on the top of the die and isolation diffusion reaching the substrate. The last one would probably destroy the capacitor.
Anyhow I assume you would get quite a low breakdown voltage because of the high doping of the isolation diffusion. 

[magic]

Doping is lighter at the bottom. I think the LTZ1000 uses similar tricks in Q1/D1 (or at least that's the only explanation I was able to come up with ) and its absolute maximum rating is 15 volts. This capacitor holds only ~2V so even a reverse biased BE junction would work.

The circuit has a hidden feature described in the patent the datasheet refers to.
The bias generator is tuned to generate a current equal to the current flowing through each leg of the bandgap reference. This current flows through the yellow path. Furthermore the yellow path is equal to "half" of the path Q4/Q13/Q14. Via Q11/Q12 that leads to a constant collector potential at the transistors Q2/Q1 which then act more constant independent of temperature and bias.

I was wondering about it too. SPICE thinks (and I know no reason to disagree) that such Brokaw cells are somewhat sensitive to variation in collector voltage. Sensitive even to equal change in both collector voltages, because those transistors obviously aren't matched. Later AD58x bandgap references bootstrap the collectors to regulated 2.5V-1Vbe, here it seems that Q11~Q15 track the base voltage of Q7 as it pushes current through R30.

[Noopy]

Doping is lighter at the bottom. I think the LTZ1000 uses similar tricks in Q1/D1 (or at least that's the only explanation I was able to come up with ) and its absolute maximum rating is 15 volts. This capacitor holds only ~2V so even a reverse biased BE junction would work.

You are right with the lower doping in the deeper areas. 
Here the voltage is definitely not critical. But the voltage could become a problem in general.

The circuit has a hidden feature described in the patent the datasheet refers to.
The bias generator is tuned to generate a current equal to the current flowing through each leg of the bandgap reference. This current flows through the yellow path. Furthermore the yellow path is equal to "half" of the path Q4/Q13/Q14. Via Q11/Q12 that leads to a constant collector potential at the transistors Q2/Q1 which then act more constant independent of temperature and bias.

I was wondering about it too. SPICE thinks (and I know no reason to disagree) that such Brokaw cells are somewhat sensitive to variation in collector voltage. Sensitive even to equal change in both collector voltages, because those transistors obviously aren't matched. Later AD58x bandgap references bootstrap the collectors to regulated 2.5V-1Vbe, here it seems that Q11~Q15 track the base voltage of Q7 as it pushes current through R30.

 

[Noopy]

400°C later => surprise: There are only two diodes. 

Have you tried less brutal decapping methods? Soaking in DMSO at 60-150°C for 0.5-3h destroys most epoxies and many other coating/potting compounds. They can be cleanly picked apart with tweezers afterwards. Bond wires won't survive, but nothing inorganic should be melted or broken.

Today I tried 45min DMSO at round about 110°C. I cooked a modern SO8 and an old DIL8 => nothing 
The package is still like new.
 

[magic]

This stuff is more likely to work on things like PCBs and maybe some potting compounds.

That being said, one of the Dynaloy products is supposedly based on DMSO and phenoxyisopropanol. They recommend using it at 150°C.
https://siliconpr0n.org/wiki/doku.php?id=decap:solvent

[Noopy]

I have done some "research": It seems like DMSO can do quite a lot. I have read something about DMSO dissolving polyimide...

I"m a little afraid of going to high with the temperature. DMSO seems to get unstable at high temperatures and some chemicals can accelerate this behaviour.

[BU508A]

Hei Noopy,

are you interested in some 3.3V references from Texas Instruments?
This one: REF3033AIDBZR (SOT23 case)

I can send you 5 or 10 pieces if you like, just let me know. (PM via Forum).

Thanks for your work, I'm really appreciating it.