3
\$\begingroup\$

So I was looking at the schematic of NPN and PNP transistors which basically involved layered blocks of doped material stacked.

enter image description here

My question is how do wires connect to these devices? I think this image is misleading as the actual width of the n segment would be on the order of 50-60ish nm as per:

https://en.wikipedia.org/wiki/65_nanometer

And it seems a molecular level of accuracy would be required to line up a copper wire on both ends. How on earth is the copper bonded to the the actual structure, without damaging it? Surely heat can't be used, but if it is just placed on then it could easily fall apart if it isn't bound chemically to the junction.

\$\endgroup\$
  • 2
    \$\begingroup\$ Maybe have a look at the first 180 google search results of "wire bonding" \$\endgroup\$ – PlasmaHH Oct 1 '15 at 20:09
  • 1
    \$\begingroup\$ Contacts are made on the device and then thin gold wire is "wire bonded" to the metallic contacts. \$\endgroup\$ – George Herold Oct 1 '15 at 20:09
  • 1
    \$\begingroup\$ @PlasmaHH wire bonding doesn't seem to give me hits on the microscopic connections that take place between the transistors. Instead it mostly focuses on connecting the finished chips to the rest of the board etc.. I'm curious about the actual connections that happen say inside of the CPU itself between individual transistors \$\endgroup\$ – frogeyedpeas Oct 1 '15 at 20:17
  • 3
    \$\begingroup\$ The others have pointed you in the right direction about the bonding, but you should also realise that no-one is sticking a wire on a 65nm feature. That sort of geometry is within a massive chip, not forming a discrete device that you'd buy in a three-legged package. \$\endgroup\$ – user1844 Oct 1 '15 at 20:17
  • 2
    \$\begingroup\$ Thank you, I'll read into it. And I appreciate your taking the time to actually help guide and answer my specific question. \$\endgroup\$ – frogeyedpeas Oct 1 '15 at 20:22
4
\$\begingroup\$

When connecting small features to the outside world you need a metalisation step.

The feature (sub μm or several μm doesn't really matter) is too small to connect to in any kind of way.

So quite often those features are etched such that they lay a bit higher than any other conducting element in the vicinity. The vicinity for normal or crude devices often being more than 25μm-range, but for expensive processes, sometimes metalisation is done at several μm, possibly even smaller (though I have no knowledge of it).

Then, a layer of non-conducting material is put on, for example, polyamide. This is then carefully etched/trimmed down until a certain parameter is reached, either time or some amount of etchant contamination or what have you, that means that the very tops of the structure lay bare again.

Then one of several metalisation methods is used to lay down metal tracks on the non-conducting area with tiny bits of feature sticking through it. Sputtering and then etching out is a tactic. I wouldn't be amazed if by now there's some company using bubble-jet printing to make 100μm metalisation tracks. If only just experimentally. There are likely many ways to getting it done. And there may be many more layers in between depending on technology, but I propose we leave all that out for now.

This metal, usually some manner of gold or gold alloy (EDIT: Turns out gold is just in the one specific process I'm working with, but eh... Same difference), then automatically connects to the features and is isolated from anything else of relevance by the extra insulating layer and it can then trace all the way across the top if it likes to.

For very electrically complicated devices this may even happen several times in intermediary layers.

These metalisation tracks then lead to golden pads, usually somewhere near an edge of the die, but sometimes not. Those pads are then wire-bonded onto with tiny gold wires (in some LEDs you can see those gold wires coming from the top) and onto the wire frame of the device, after which it's dipped or otherwise encased in whatever material seems best.

That, at least, is the method I know of. There may be many more in other industries than I have had contact with. But options aplenty.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.