Help me gut my CPU

[m00g00]

I have an old Athlon 64 CPU laying around that I feel like destroying. I already removed the heat spreader. Is there anyway to remove the metal top of the core, so I can see the shiny transistor city underneath?

 

[aigomorla]

u mean you want to crack the die?

Sand Paper?

 

[m00g00]

I want to see something like this

 

[fffblackmage]

I don't know how you're gonna do it without utterly ruining it - I mean take it apart in such fashion that you don't destroy the die you want to look at. But if you somehow manage to do it, please come back with some pictures. =D

 

[m00g00]

I'm guessing the metal top of the core is not some cap that can simply be popped off. I don't know much about IC packaging. Is the shiny metal top somehow melded into the wafer? If so maybe sandpaper would be a good idea. Maybe I can use my belt sander...

 

[CTho9305]

You can remove the metal heat spreader... google "decapping" or "delidding". People generally seem to do that using a very sharp knife around the edges. In my experience, the lid is usually not soldered to the die, although I've seen some Intel CPUs where it was. Don't bother trying to sand/grind through the heat spreader - it's very thick, and the die underneath it is very thin.

Getting a picture like what you posted is relatively difficult. With older wire-bond processors (e.g. 386, 486, Pentium, Pentium Pro), simply exposing the die showed you the "interesting" side of it (see this for an example). However, all modern processors are "flip chip" - instead of connecting to the external world through a ring of wires on the outer edge, the entire surface is covered in small solder bumps that connect to corresponding solder points on the package. If you remove the lid, you'll see the uninteresting side of the die...and I've never managed to get through it from the uninteresting side. Intel's TV commercials show the interesting side up, but that's just not the way the chips are really built. You'll need to remove the die from the package. I've managed to removed dies from packages using a blowtorch to melt the solder. I would imagine the fumes are toxic, so you probably don't want to breathe them.

Here is a photo I took of the underside of a Core 2 Quad after removing it from the package. After sanding it a bit, I got this result, and with some more sanding I finally got this result. Note that it's extremely difficult to sand evenly enough to get good results on modern process nodes. Some areas of the die are exposed down to a very low layer, while others are still covered by higher layers. Here is a photo of a P3. Here is a P4 Prescott (note the diagonal scratches from my coarse sand paper).

So, the steps I've used were:
1. Sharp knife to remove the lid
2. Lots of heat to melt solder and free the core(s) when the solder is molten
3. 600-2000 grit sandpaper

I have many more photos if you're interested. Member idontcare helped me significantly.

 

[wonderflu]

This may be a dumb question, but why is the interesting part of the die colored? I always assumed the coloring was for visualization purposes and that a fabricated die would be uncolored.

 

[CTho9305]

This may be a dumb question, but why is the interesting part of the die colored? I always assumed the coloring was for visualization purposes and that a fabricated die would be uncolored.

My guess is that it's mostly related to the structures being small compared to the wavelength of light. "Color" in the normal sense doesn't really mean much for a wire that's 130nm (0.13um) wide. You'll only see color if you have a lot of those wires next to each other, and then the groupings of those wires could appear different colors based on their patterns (for example, you could build a "red" 650nm structure from the smaller components, or a "blue" 450nm structure.

I've noticed that the color changes dramatically based on the angle of the lighting; fluorescent lighting seems to work best for producing the most impressive colors (possibly because of the more pronounced emission peaks).

 

[Dadofamunky]

LOL. You must have WAY too much time on your hands for that kind of project! No, just kidding. Neat photos.

You can remove the metal heat spreader... google "decapping" or "delidding". People generally seem to do that using a very sharp knife around the edges. In my experience, the lid is usually not soldered to the die, although I've seen some Intel CPUs where it was. Don't bother trying to sand/grind through the heat spreader - it's very thick, and the die underneath it is very thin.

Getting a picture like what you posted is relatively difficult. With older wire-bond processors (e.g. 386, 486, Pentium, Pentium Pro), simply exposing the die showed you the "interesting" side of it (see this for an example). However, all modern processors are "flip chip" - instead of connecting to the external world through a ring of wires on the outer edge, the entire surface is covered in small solder bumps that connect to corresponding solder points on the package. If you remove the lid, you'll see the uninteresting side of the die...and I've never managed to get through it from the uninteresting side. Intel's TV commercials show the interesting side up, but that's just not the way the chips are really built. You'll need to remove the die from the package. I've managed to removed dies from packages using a blowtorch to melt the solder. I would imagine the fumes are toxic, so you probably don't want to breathe them.

Here is a photo I took of the underside of a Core 2 Quad after removing it from the package. After sanding it a bit, I got this result, and with some more sanding I finally got this result. Note that it's extremely difficult to sand evenly enough to get good results on modern process nodes. Some areas of the die are exposed down to a very low layer, while others are still covered by higher layers. Here is a photo of a P3. Here is a P4 Prescott (note the diagonal scratches from my coarse sand paper).

So, the steps I've used were:
1. Sharp knife to remove the lid
2. Lots of heat to melt solder and free the core(s) when the solder is molten
3. 600-2000 grit sandpaper

I have many more photos if you're interested. Member idontcare helped me significantly.

 

[SSJ4Gogeta]

My guess is that it's mostly related to the structures being small compared to the wavelength of light. "Color" in the normal sense doesn't really mean much for a wire that's 130nm (0.13um) wide. You'll only see color if you have a lot of those wires next to each other, and then the groupings of those wires could appear different colors based on their patterns (for example, you could build a "red" 650nm structure from the smaller components, or a "blue" 450nm structure.

I've noticed that the color changes dramatically based on the angle of the lighting; fluorescent lighting seems to work best for producing the most impressive colors (possibly because of the more pronounced emission peaks).

Color, afaik, doesn't have anything to do with the size of the structure. It depends on which color wavelength the material absorbs/reflects. A 450 nm structure may or may not be blue. When light encounters an object the size of its wavelength or less, a lot of diffraction will occur. I guess that's what's occurring here. The small structures on the die act like a diffraction grating. We see the rainbow colors on an optical disk for the same reason. The tracks on the disk are very closely spaced and they diffract light.

 

[JACKDRUID]

You can remove the metal heat spreader... google "decapping" or "delidding". People generally seem to do that using a very sharp knife around the edges. In my experience, the lid is usually not soldered to the die, although I've seen some Intel CPUs where it was. Don't bother trying to sand/grind through the heat spreader - it's very thick, and the die underneath it is very thin.

Getting a picture like what you posted is relatively difficult. With older wire-bond processors (e.g. 386, 486, Pentium, Pentium Pro), simply exposing the die showed you the "interesting" side of it (see this for an example). However, all modern processors are "flip chip" - instead of connecting to the external world through a ring of wires on the outer edge, the entire surface is covered in small solder bumps that connect to corresponding solder points on the package. If you remove the lid, you'll see the uninteresting side of the die...and I've never managed to get through it from the uninteresting side. Intel's TV commercials show the interesting side up, but that's just not the way the chips are really built. You'll need to remove the die from the package. I've managed to removed dies from packages using a blowtorch to melt the solder. I would imagine the fumes are toxic, so you probably don't want to breathe them.

Here is a photo I took of the underside of a Core 2 Quad after removing it from the package. After sanding it a bit, I got this result, and with some more sanding I finally got this result. Note that it's extremely difficult to sand evenly enough to get good results on modern process nodes. Some areas of the die are exposed down to a very low layer, while others are still covered by higher layers. Here is a photo of a P3. Here is a P4 Prescott (note the diagonal scratches from my coarse sand paper).

So, the steps I've used were:
1. Sharp knife to remove the lid
2. Lots of heat to melt solder and free the core(s) when the solder is molten
3. 600-2000 grit sandpaper

I have many more photos if you're interested. Member idontcare helped me significantly.

nice article! awesome pix, bookmarked the site.

 

[m00g00]

Wow. Suddenly I have much less desire to destroy my CPU. Though I do remember that I have a 486 SX chip laying around somewhere, maybe I should try my hand at that.

CTho9305: do you have any idea why it is more difficult to sand on the area of the die that is exposed after removal of the heat spreader?

I'm also somewhat surprised that a blow torch would not do more damage to the die itself before the solder melted.

Either way, sweeeeeet pictures.

 

[fffblackmage]

I have many more photos if you're interested. Member idontcare helped me significantly.

Excellent photos. Thanks a lot for sharing!

 

[CTho9305]

CTho9305: do you have any idea why it is more difficult to sand on the area of the die that is exposed after removal of the heat spreader?

I'm also somewhat surprised that a blow torch would not do more damage to the die itself before the solder melted.

The interesting side is like glass with very many small metal wires running through it. That probably makes it significantly less resistant to sanding than the exposed side because to make progress you just have to work through many small, separate structures that don't support each other mechanically. The other side is more like a solid piece of glass. That's just an educated guess though.

The heat probably does a lot of damage from a functional standpoint (breaking/shorting wires; causing dopant atoms to move around) but you'd probably need an electron microscope to be able to see that. I think I worked out that the best resolution I could get was 3um/pixel (and that might have been per RGB subpixel)...well above the scale of wires and transistors. Even if wires were shorted / disconnected all over, I wouldn't be able to tell in the photos.

 

[nyker96]

why not frame it like a souvenir?

 

[CurseTheSky]

Suddenly I don't think the Pentium, Celeron, Pentium 3, and Pentium 4 Northwood I have laying around "just in case" are going to be intact much longer...

 

[Vipersfate]

Sounds like something I can do on a dead day at work. I do computer repair, and we have LOTS of old PII and PIII CPUs laying around. 333Mhz stuff.

 

[ther00kie16]

Got an AMD-6 550 that i'll have to take a look at. Too bad a couple of Pentiums were already thrown away.

 

[MrK6]

I have many more photos if you're interested. Member idontcare helped me significantly.

Wow, those are great photos. Thanks for sharing .