Transistors - die pictures

[RoGeorge]

Wanted to ask before and censored myself, but I couldn't stand the itch any more:  What's the green and what's the yellow stuff?  What material is that?
Are those the real colors or are they colored by software?   

[Noopy]

That are not the real colors an they are not colored by software. 

With "normal light" you can only see the metal layer:



With light coming "from the camera" resonances are forming in the thin layers of the chip. Different thicknesses are creating different colors:



Professional people use reflected-light microscopes. I use light coming from behind the die:



It seems the lens is reflecting enough light to do the job.

 

[magic]

Multiple supposedly professional sources claim it's due to iridescence. Off the top of my head:

https://www.quora.com/Why-are-microprocessor-wafers-so-colorful
http://www.designinganalogchips.com/

It basically means you see the thickness of the surface layer of glass. This thickness varies, because glass accumulates on "finished" areas while new masks are applied and selectively etched to expose other areas to subsequent processing steps.

[Noopy]

I'm no optic expert but in principle that is what I said, resonances in thin layers, isn't it?

[T3sl4co1l]

There's kind of three things going on, actually:
- Doping
- Layers
- Patterns

Doped semiconductors actually reflect light differently.  I can't find a reference for it (go figure, searching for basic physics only turns up current high-level articles ) but it's something about the light's interaction with the carrier type and density, causing a phase or polarization shift, and therefore causing interference with the incident light.  There's an optical hall effect which might be what I'm thinking of, which is also affected by ambient magnetic fields as the name suggests.

Layers of semiconductor, oxide and etc. transmit light at different velocities, and are used in varying thicknesses, leading to simple interference colors.

Patterns, when periodic and finely etched (comparable to the wavelength of light), create diffraction gratings -- light is reflected from each wire in a bus, say, which when evenly spaced, causes interference at different angles -- a rainbow is reflected.  We don't see much of this on small devices (few features to reflect light) or large pitch devices (features are widely spaced), but it's why CDs, EPROMs and etc. are so colorful.

Note that interference depends on angle as well, as a glancing angle travels farther through the material.  So you generally get a play of colors over the surface of e.g. a microprocessor, but the exact rate (spectrum and angles) at which the colors are reflected varies by region.

A microprocessor is a good example, containing all of the above: well, probably not much visible semiconductor unless it's a quite old one, but periodic structures such as mask ROM, register files or caches, and buses, tend to show off all sorts of diffraction patterns, while "random logic" regions are more chaotic and have a noisy or speckled appearance.

Tim

[magic]

Patters - true but not a thing at sufficient magnification, like in this thread.
Layers - as I said above, I think it's mostly about the layer of glass on the surface. Light penetration through actual silicon is very low (a few µm at best for red) and many of those structures are "relatively" deep.
Doping - not entirely sure. Anyone volunteers to treat a die to hydrofluoric acid to strip the glass and see if any color remains? Actually, people do such things (and also remove the metal layers which might obscure underlying silicon) and I think they end up needing to treat the die chemically to "stain" the doped areas. Look up "deprocessing" and "delayering".

[Noopy]

I definitely won´t work with hydrofloric. That´s some pretty nasty shit. 

It seems that deeper structures don´t show up in different colors. You can spot them only by the bumps they create at the surface.



Here the deep implant, the areas containing active elements and the lateral isolation have all the same color. That´s probably because of the low light penetration. No light no colors. All three structures are quite "deep".

[magic]

The bumps which seem to be caused by collector buried layer are easy enough to explain - they are bumps on the surface of silicon itself, not differences in glass thickness, so the color is uniform.

But I'm not sure if the same is true about the bumps on isolation diffusions or what those bumps actually are

Maybe doping does play some role.

[Noopy]

Hm... 

The glass has to play a role. In the picture above the metal layer is missing but the vias are already etched. In the areas where the glass is missing you can see no color. Whereas around the vias you can see the color of the underlying layer.

[Noopy]

I have taken some pictures of an old thyristor: ST103

https://www.richis-lab.de/Bipolar13.htm






Looks pretty rude.




A rough structure...




Rough structures gives you a lot of leackage so they etched some traces to get smooth junction edges.

[David Hess]

Rough structures gives you a lot of leackage so they etched some traces to get smooth junction edges.

Like high voltage diodes, I thought the outer ring structure in thyristors was to prevent high voltage breakdown.

[Noopy]

Rough structures gives you a lot of leackage so they etched some traces to get smooth junction edges.

Like high voltage diodes, I thought the outer ring structure in thyristors was to prevent high voltage breakdown.

I think we are talking about the same mechanism:
The amount of leakage current is depending of the voltage.
More voltage, more leakage, triggering the thyristor... 

[David Hess]

Rough structures gives you a lot of leackage so they etched some traces to get smooth junction edges.

Like high voltage diodes, I thought the outer ring structure in thyristors was to prevent high voltage breakdown.

I think we are talking about the same mechanism:
The amount of leakage current is depending of the voltage.
More voltage, more leakage, triggering the thyristor... 

Time dependent false triggering in a thyristor is suppressed with metalization between the base and emitter of each transistor.  "Sensitive gate" SCRs are sensitive because they lack this.  An external resistor is less effective because of the distributed nature of the spreading resistance; the thyristor could be triggered in an area where there is too much resistance to the gate connection which is a very bad situation.

I was referring to the guard rings applied to semiconductor junctions including diodes and transistors which increase breakdown voltage.

[T3sl4co1l]

I don't think he was talking about rate?  Lacking a bit of nuance I think is all.

Namely: semiconductor breakdown is weird around the edges.  Guard rings are used to smooth out the electric field there, making breakdown less likely.  I forget exactly how these work; something about spaced P-N junctions, and maybe field plates too, that happens to do the job.

A fractured and contaminated edge can have all sorts of issues, including errant or intermittent conduction (due to stray charges, surface states and other arcana).  Passivated edges are better (keeps contamination out), but the materials can trap charges, and still transmit ambient electric fields (think random MOSFETs around the edges).

I would guess, for the technology of the day, they did the best with what they had: they probably found that etching the sidewalls, rather than leaving them open on the primary surface, simply gave better results.  The device might not be very tolerant of high voltage or rate stresses (avalanche and pulse operation?), but also maybe it was low enough voltage that it worked out okay.

Similarly, for a long time it used to be that silicon rectifiers were more fragile than some of the alternatives.  I guess that's sometimes still true today...  Avalanche-capable diodes were developed, and the huge energy capacity (relatively speaking) of a TVS diode, or a suitably rated MOSFET for that matter, can (probably?) only be possible this way, by preventing edge breakdown.

Tim

[Noopy]

Time dependent false triggering in a thyristor is suppressed with metalization between the base and emitter of each transistor.  "Sensitive gate" SCRs are sensitive because they lack this.  An external resistor is less effective because of the distributed nature of the spreading resistance; the thyristor could be triggered in an area where there is too much resistance to the gate connection which is a very bad situation.

I was referring to the guard rings applied to semiconductor junctions including diodes and transistors which increase breakdown voltage.

Didn't know that. Thanks for this information.

You talk about this ring?



In my view that's for smooth edges of then pn junction where rough structures and contamination can lead to leakage...



Guard rings are normally built with pn-structures and metal for all I know.


I agree with T3sl4co1l.
"Dirty edges" lead to all kind of negative characteristics.
In this thyristor I assume the biggest problem would be leakage current because worst case it can trigger the thyristor.

[Noopy]

I decapped a RF-Power-Transistor! 






In the package we find very long and thin transistors for best high frequency performance.
The capacitors are important for matching the Input and the output to the rest of the circuit.




Nice! 








There are some protection structures at both ends of the transistor die, probably zener or supressor.
The small transistors are too complex to identify the elements. Probably they used two metal layers.




Hey they damaged the die!  But probably no bigger problem.


A lot more pictures on my website:

https://www.richis-lab.de/FET03.htm

 

[RoGeorge]

Wow, that's beautiful!   

Datasheet says 2GHz https://www.nxp.com/docs/en/data-sheet/MRF18060A.pdf
It's intriguing to see so many wire bonds inside of a 2GHz part, thinking here about parasitic inductance and stray capacitance. 

[exe]

Wow, rf beauty!

It's intriguing to see so many wire bonds inside of a 2GHz part, thinking here about parasitic inductance and stray capacitance. 

I thought the same. However, many wires in parallel will actually lower inductance.

[Noopy]

And most important is the impedance matching. To achieve this they even put capacitors in the package! 160pF gives you 0,5 @2GHz but if the impedance matching is ok... 

[ocw]

The MRF18060A pictures reminds me of what a SD2942 looks like "under the hood."  Attached are some pictures of it.

The SD2942 is a 500 watt dissipation 250 MHz MOSFET.  The pictures show one of the two matched MOSFET's in it which are used in a common source push-pull situation.  See:https://www.st.com/resource/en/datasheet/sd2942.pdf

[David Hess]

I would guess, for the technology of the day, they did the best with what they had: they probably found that etching the sidewalls, rather than leaving them open on the primary surface, simply gave better results.  The device might not be very tolerant of high voltage or rate stresses (avalanche and pulse operation?), but also maybe it was low enough voltage that it worked out okay.

The MESA structure apparently does the same thing as the guard rings which were used later.

[ocw]

While significantly lower than the normal operating current and voltage, the attachment shows the excellent match in Id/Vds measurements on the two different halves on a SD2942.  The same measurements between two different MOSFET's shows a poor to no match.

[magic]

I see the Vgs/Ids consistency being poor. If we adjust Vgs from 2.2V to 2.1V or 2V, the bottom three transistors have almost identical Vds/Ids behavior

[ocw]

Sorry for the lack of an explanation of the Id/Vds graphs.  They showed six different curves at six different Vgs voltages for two matched MOSFET's located in the same physical package.  The first attachment of this message labels the A and B halves of the one matched MOSFET package.

The second attachment shows similar Id/Vds graphs for one half of two different SD2942 MOSFET packages.  The added 1 - 2 lines are between the two different MOSFET's at the same Vgs.  If they were matched they would be much closer together as they were on the first graph.  The match at the higher Vgs is what is important.

The curves in a matched MOSFET pair is typically close, like that shown.  While a perfect match is uncommon, much more significant mismatches between two different MOSFET packages with the same part number are not unusual.

[magic]

You are right, I didn't understand what was shown on the plot.

I was just nitpicking that you have only demonstrated a mismatch of Id vs Vgs rather than Id vs Vds
Frankly, you still are talking about Vgs mismtach.

That being said, we can actually see that one of the different FETs tends to have a sharper knee near zero than the other, regardless of Vgs and Id. So I guess that counts as a true mismatch of their Vds / Ids characteristics. I'm no expert on FETs so no idea what's causing this and if it's normal to see such differences correlated with threshold voltage of individual unit.