Transistors - die pictures

[magic]

patience

aka focus stacking

[Noopy]

patience

aka focus stacking

Yes:






37 pictures / 1GB for this one...

 

[SilverSolder]

Are you using Photoshop for the focus stacking?

[Noopy]

I use Photoshop but for focus stacking I use Helicon Focus: https://www.heliconsoft.com/heliconsoft-products/helicon-focus/
Works fine. 

[Noopy]

The Motorola ENI-1B seems to be a special transistor for the RF amp manufacturer ENI. It was found in a plasma generator but I can´t tell you more. 
Perhaps Motorola did some binning. Perhaps they just gave it a new name. 






A nice big transistor with a big heatspreader and a big die.




And a nice bondwire! 




It´s a MESA-transistor.




There are more structures between base and emitter than you would expect.
We have similar structures in the conterfeit BUX22 (https://www.richis-lab.de/Bipolar09.htm), in the conterfeit BUX66 (https://www.richis-lab.de/Bipolar42.htm) and in the 3DD15D (https://www.richis-lab.de/Bipolar05.htm). I still don´t know why they integrated these structures. 




Base emitter breakdown occurs at -11V and is quite uniform.
20mA




50mA




100mA




200mA




500mA


https://www.richis-lab.de/Bipolar51.htm

 

[Noopy]

I have added two transistors built by CDIL (Continental Device India Ltd.):




CDIL BC550C






With an edge length of 0,31mm the CDIL BC550C is a little bigger than the Philips BC550C (0,26mm x 0,27mm).




Yes, it´s a small transistor. 


https://www.richis-lab.de/Bipolar39.htm






CDIL BC560C






Although pnp transistors have often worse specifications than npn transistors the CDIL BC560C has the same edge length as the CDIL BC550C.




Yes, also a small transistor. 


https://www.richis-lab.de/Bipolar52.htm

 

[exe]

Although pnp transistors have often worse specifications than npn transistors the CDIL BC560C has the same edge length as the CDIL BC550C.

Yeah, I wonder about that, according to datasheets, their collector-base capacitances are identical. Their die geometries seem to be the same or very similar. Perhaps, they tried to make matching devices? If so, they succeeded. Or, may be, N-devices are only smaller when it comes to fets?

[Noopy]

Although pnp transistors have often worse specifications than npn transistors the CDIL BC560C has the same edge length as the CDIL BC550C.

Yeah, I wonder about that, according to datasheets, their collector-base capacitances are identical. Their die geometries seem to be the same or very similar. Perhaps, they tried to make matching devices? If so, they succeeded. Or, may be, N-devices are only smaller when it comes to fets?

Philips BC550 and BC560 were different in size... 

Edit: It looks like the BC560 has a smaller "base contact frame" and so more active area than the BC550.

[Noopy]

OC811, the first junction transistor built in the GDR.
20V / 15mA








The usual construction...










In some of the OC811 it seems like they had problems with the production quality. 


https://www.richis-lab.de/Bipolar53.htm

 

[exe]

Ah, hand soldering, good ol' days . In the USSR we had a rumor that only women were hired to do the final assembly (i.e., attaching bond wires). Also, I was told they were not allowed to open windows in the facility as that would half the yield due to impurities in the air and drafts. Do you know any anecdotes about semiconductor production back in the day?

[Noopy]

I have heard of erratic low yields in one of the old american semiconductor companies. Most of the time everything was perfect and sometimes the yield was really low. It took them some time to realise the root cause was a farmer who from time to time sprayed herbicides. His field was located besides the factory. The chemicals made their way through the air supply and killed the devices. 

[Noopy]

ON Semiconductor BC546B. The high voltage type of the BC5xx series (65V).
...well, at least it should be a ON Semiconductor BC546B. You can´t be 100% sure. These small packages and dies give you almost no hint who is the manufacturer.
The pins are not covered with tin completely. That´s seems a little odd to me. 






The die is 0,33mm x 0,33mm.





ON Semiconductor BC546C. Index C gives you the highest hfe (420 - 800).






Unsurprisingly the BC546B uses the same die as the BC546C.


https://www.richis-lab.de/Bipolar54.htm

 

[Noopy]

ON Semiconductor BC556B, the PNP type beside the B546.






Same size, same structure. 


https://www.richis-lab.de/Bipolar55.htm

 

[Noopy]

I have put some more information about the BUX42 here: https://www.eevblog.com/forum/projects/decapping-and-chip-documentation-howto/msg3565360/#msg3565360

[Noopy]

...



The structure is interesting. The left brown area is the p-doped emitter. The following green are is the n-doped base. In the next pictures we will see that this is the base-emitter-junction.
Before the base contact there is a small brown p-doped layer certainly above the base area. Why that? Looks like a pinch resistor to increase the base resistance. But why would they increase the base resistance? Perhaps the resistance is equalising the electric stress on the transistor area? 




That´s the base emitter junction! 
(13,5V / 20mA)

...

Do you remember the counterfeit BUX66?
If we remove the metal layer we get a better look at the strange structure around the base area. As already described in the n-type base area (cyan) there is a p-type ring (light green) just behind the pn-junction. I still don´t know why. 


https://www.richis-lab.de/Bipolar42.htm

 

[magic]

It seems a theoretical possibility that uniformly distributed base resistance reduces formation of current hot spots and improves SOA.
A local increase in current increases voltage loss across Rb and reduces Vbe.

[Noopy]

It seems a theoretical possibility that uniformly distributed base resistance reduces formation of current hot spots and improves SOA.
A local increase in current increases voltage loss across Rb and reduces Vbe.

Sounds plausible. 






Today we will take a look into a RCA 2N3375.
The 2N3375 is very similar to the 2N3553 (https://www.richis-lab.de/Bipolar22.htm) but can conduct more current and dissipate more heat  (0,5A/1,5A vs. 0,33A/1A and 11,6W vs. 7W).




The pins are embedded in a pink ceramic.




You have to grind carefully because the 2N3375 contains BeO. 




The pins were soldered to the ceramic which carries the transistor.




Here we have the BeO which isolates the transistor from the package.
The flat makes it easier to align the ceramic.




The die is the same as the die in the 2N3553. The higher power capability of the 2N3375 is just caused by the different package.
Here we have only one bondwire but the bondwires are welded to the carrier three times.
Perhaps the 2N3632 (1A/3A, 23W) which is on the same datasheet contains two of these dies. The datasheet states two times the base-collector-capacity...




Gold solder. 






More about the multi emitter design can be found here:
https://www.richis-lab.de/Bipolar22.htm
https://www.eevblog.com/forum/projects/transistors-die-pictures/msg3286140/#msg3286140


https://www.richis-lab.de/Bipolar56.htm

 

[Noopy]

That´s a dead Motorola MJ802 (90V/30A/2MHz/200W).






A big die and a big heatspreader as we would expect.
They used different bondwire diameters for emitter and base.
And yes, it´s dead. 




We can spot four dots due to the testing in the production line.




It´s a MESA transistor with a trench at the edges.






Now that looks bad. There is a lot of molten metal.




After removing the bondwire we can see the whole mess. In the upper area there is a discoloration over the pn-junction like we have seen it in the 2N2222 I killed (https://www.richis-lab.de/Bipolar04.htm). The first failure must have occurred in this area where most of the metal is molten. Some of the metal accumulated right of the bondwire. The die is cracked. 








The cracks go all the way down to the heatspreader.
The molten trench is quite deep. Normally the metal layer isn´t that deep. I assume in this small area some of the silicon is molten. Silicon needs 1400°C to melt but with a lot of energy in this small area it should be possible.




I tried to remove the metal layer to see more details of the failed structures but the die fell apart...


https://www.richis-lab.de/Bipolar57.htm

 

[T3sl4co1l]

Aluminum depresses the melting point of silicon, although it does seem suspicious that there'd be enough to do that.  Alternate explanation might be arc flash, and the aluminum carried more current, preferentially heating those areas.  Maybe both!

Ed: If some of the bond wire melted/sprayed into it, that would definitely be enough.  It's not clear from the side view how much if any did; there's a spot on the wire but I can't tell if it's a cavity and I'm seeing through it, or it's just marked by metal vapor.

Tim

[Noopy]

That are some interesting thoughts. 

I didn't see significant damage at the bondwires.
In my view there is just some metal build-up

[T3sl4co1l]

I have a special fondness for Al-Si alloys as I did some amateur foundry work years ago, along with studying some metallurgy, which I still remember a fair bit of.

Al-Si alloys have high strength and low ductility as-cast, and flow into molds better than any other alloy.  (Ductility doesn't help castings, as you can't get any strength into them by mechanical working!  Compare with wrought alloys like 6061, which achieve maximum strength with rolling/drawing/etc. so are only available as plate, bar, extrusion, etc.)

Typical compositions are 8-10% or 22-28% Si, a few % of other elements to improve strength (Mg, Cu, etc.) and balance Al.

The eutectic is around 13%, which freezes suddenly at 576°C so isn't good for casting: metal shrinks on freezing, causing it to suck in from anywhere hotter, leaving voids there; it's better that this happens more gradually, with a "mushy" off-eutectic alloy.

The higher concentration is hypereutectic (literally: past eutectic), so instead of growing primary aluminum dendrites on cooling, it's full of plates of Si crystals floating in a eutectic (fine grained) matrix.  The larger Si crystals make it abrasion resistant, and it's good for hot-forging as well, so is commonly used for engine parts such as pistons and connecting rods.

Needless to say, such crystals are very heavily P-doped. I don't know that I've ever seen the resistance probed, that'd be an interesting experiment I suppose... it should be measurably different, but still quite a low resistance over either grain (Si or Al).

Al-Si forms no intermetallic compound, it's a simple binary eutectic system.  Much like Pb-Sn, for a familiar example.   Which, for soldering, we prefer around the eutectic, because shrinkage is avoided by the low molten mass required, and because we want the joint to freeze suddenly.  (SAC305 is a bit past eutectic, so tends to leave a rough surface as crystals have enough time to grow to macroscopic size -- hence its tendency to look "cold".  Both silver and copper, by the way, form intermetallics -- brittle compounds that make for a stronger alloy.)


So, at the surface of that poor ex-sistor, it'll be some kind of hyperhypereutectic Al-Si alloy, I should guess.  The melting point will be close to that of pure Si, ~1400°C.

Tim

[Noopy]

Hey, we have a Al-Si-Expert here. 

Now it would be interesting how high the temperature on such a die can go. I assume discharging a big capacitor quite fast in some micrometer^3 will give you a lot of temperature. 

[dzseki]

These are the real transistor die pictures  (pun intended)

That´s a dead Motorola MJ802 (90V/30A/2MHz/200W).






A big die and a big heatspreader as we would expect.
They used different bondwire diameters for emitter and base.
And yes, it´s dead. 




We can spot four dots due to the testing in the production line.




It´s a MESA
transistor with a trench at the edges.






Now that looks bad. There is a lot of molten metal.




After removing the bondwire we can see the whole mess. In the upper area there is a discoloration over the pn-junction like we have seen it in the 2N2222 I killed (https://www.richis-lab.de/Bipolar04.htm). The first failure must have occurred in this area where most of the metal is molten. Some of the metal accumulated right of the bondwire. The die is cracked. 








The cracks go all the way down to the heatspreader.
The molten trench is quite deep. Normally the metal layer isn´t that deep. I assume in this small area some of the silicon is molten. Silicon needs 1400°C to melt but with a lot of energy in this small area it should be possible.




I tried to remove the metal layer to see more details of the failed structures but the die fell apart...


https://www.richis-lab.de/Bipolar57.htm

 

[T3sl4co1l]

I've had IGBTs fail, in turn crowbarring an industrial 480V circuit.  That's a few 10s of kA fault current, and it lasts for some milliseconds before the fuses open.  In that time, the bondwires and die surface evaporate into plasma, the buildup in pressure blowing apart the plastic casing like a shotgun shell.  In the next milliseconds, the expanding arc flash erodes anything nearby; the PCB showed erosion and melting of supply and output connections, around solder joints.  (Soldermask is an effective insulator at these voltages -- or ablatively under the heat?)

And that was with the smaller SOT-227 modules.  The bigger bus-bar mounted kind tend to launch shrapnel when this happens.  Installing guards is recommended, or operating with the equipment closed up.

Likely the above example could've happened in an audio or servo driver, just from electrolytic capacitor discharge.  There's really not that much material there, and an arc delivers heat quite efficiently.  Once the bondwire pops off, the rest is history.

Tim

[capt bullshot]

And that was with the smaller SOT-227 modules.  The bigger bus-bar mounted kind tend to launch shrapnel when this happens.  Installing guards is recommended, or operating with the equipment closed up.

This is why most larger IGBT modules are gel filled. The gel absorbs the energy quite efficiently and doesn't create shrapnels.