Sciency stuff

21Sep/10179

P-III autopsy

0000047

For teaching a course I needed to take a closer look at a CPU. I asked around and got my hands on an old P-III Coppermine that was about to get thrown out. I’ll start with a disclaimer: I know virtually nothing about CPUs, so if I claim something to be true, it probably isn’t.

The first challenge is to get the actual silicon processor chip off of the plastic bonding board. In the picture below, the blue thing you see is  the back side of the processor chip. When the processor is finished, it is turned upside down and bonded to the green circuit board. This allows the metal pads on the silicon chip and the pads on the circuit board to join, creating a connection (this is one of those claims...). I believe that the CPU at that stage is heated up in order to melt the joints and thereby solder them together.

Click images for full-size. Especially the scanning electron microscope images below could be interesting to view in all its splendor.

The blue part in the middle is the actual Si chip. I needed to remove it in order to further inspect the CPU.

Back side of the circuit board containing the CPU. Each pin you see (give or take) should be connected to a pad on the silicon ship.

I figured I should be able to remove the chip by heating it. I first tried using a heat gun, but that just made some bad smelling fumes. I instead turned to brute force and used a power-saw to cut out the part containing the actual chip. Using pliers I managed to get a few pieces off of the board (cracking the chip in the process, but I was planning to do that anyway) and got the rest off by using a scalpel.

A saw comes in handy sometimes...

Below you can see the result. On the bottom side of the piece that came off you can see all the connector pads that were previously connected to the pins on the backside of the circuit board.

A piece of the processor chip came off.

This is the piece that came off. It's been flipped so that the side you see was originally facing towards the circuit board. Each little dot contains a metal pad that connects the interior of the chip to the leads on the board.

Now the interesting part begins. I looked at the piece above in an optical microscope. The picture below shows an enhanced version of the little dots in the above picture.

This is a piece of the processor. The side you're seeing was once facing towards the plastic circuit board. Each little hole is a metal pad connecting the interior of the chip to a macroscopic lead on the circuit board.

Looking closer, we start seeing some structure inside the holes.

A processor contains many layers of metal leads in order to connect the transistors at the surface of the silicon chip into useful units. The metal layers are clearly visible through the small holes in the chip.

Furthermore, by changing the focus of the microscope, we can see multiple layers within each hole

Focus is on upper layer.

Focus is on middle layer.

Focus is on bottom layer.

In a CPU there are multiple layers of sandwiched metal leads going down to the transistors at the bottom (at the surface of the silicon wafer). I believe what we're seeing is simply those different layers.

Since optical microscopy doesn't show very much detail, I decided to load the chip into a scanning electron microscope (SEM).

What I did was to cleave the chip into smaller pieces. This way I can peek from the side of the chip and get some cross-sectional images. Below is a series of images that shows a zoom-in on the surface of the Si chip. For some reason I had a lot of trouble getting a good focus. I'm not sure what kind of Si is used for these CPUs, but if it's a non-conducting silicon, the electron beam in the SEM can charge the material making it difficult to focus. Another possibility is that all the plastic encapsulation material caused charging effects.

You're looking at the processor chip from the side. In the top part of the image you see the metal pads that were once connected to leads on the circuit board when the chip was bonded face down.

We still don't see a lot. The light stuff between the metal pads is probably some kind of polymer used to fill up the space.

We start to get more detail at the surface of the silicon chip (bottom part of the image. The Si starts somewhere at the bottom of the large metal blob). The texture in the polymer filler (next to the metal blob) could be due to something being mixed into the polymer to increase its thermal conductivity.

We start seeing something below the metal blob. The vertical lines that are barely visible would be the multiple layers of metal leads. To the right you see the same structure but from another angle since the cut changes direction.

At this point they are clearly visible. I count about six layers of metal leads visible in the image.

The feature size of the lowest metal layer is around 200-250nm. Since the P-III started out at 250nm process but developed into 180nm (according to wikipedia), the transistor layer must be fairly close to the lowest visible metal layer in the image.

Just a nice overview.

This is not a cross-sectional image but is taken from the top (along the sample normal for those of you who speak science). I accidentally chipped the processor and this is how it looks. Several of the metal layers are visible and as we go down in the image, we go down through the metal layers. I would guess that the bright spots you see are vias connecting leads lying in different layers.

I believe that in order to see the actual transistors at the bottom (at the surface of the silicon wafer) I will need to remove all the layers. I'll let you know if I figure out a good way of doing this.

Comments (179) Trackbacks (13)
  1. This is incredible. I’ve been an IT professional for years and have never actually seen photographs of the chips I deal with. Thanks so much for putting these online!

    • I second that motion :) Thanks man, great post.

    • You should stop calling yourself a professional if that is the case.

    • Well, it looks like you still haven’t. These are just the leads that go to the silicon chip. Still incredible though.

    • Computer engineers deal with chips, not IT professionals

      • IT Professionals deal with the entire computer, Computer Engineers help make computers more efficent, faster, and all together better, IT fixes the problems that happen

        • I have a B.S. in Computer Engineer and I’ve been doing embedded S/W for 23 years. I do black box stuff like a digitally controlled radio or an image scanner or video tracker. It’s all DSP’s and micro-controllers. SOIC kind of stuff. I rarely deal with a PC other than as a development tool.

          A good computer engineer can design and debug both H/W and S/W at the board level and, grudgingly, do some GUI and Network crap too.

          As far as I’m concerned, IT and Professional is an oxymoron.

          • Hey Heffie, do you think you won the d*ck measuring contest? Because you didn’t.

            Your another one of these old men who think that because they deal with hardware that, that makes you the smartest person in the room. How much did you pull down last year? 100K? If your lucky. Guess what? I started out as a IT Pro and went into management, now with bonuses I make 250K a year.

            Have fun with the hardware as someone in their thirty’s fires you because your knowledge is old and outdated and we can get kids right out of college who know just as much as you do and more and are willing to do the job for 60K a year. And are willing to work more hours because they have the energy and dont have any kids basketballs games to go to. Engineers are a dime a dozen. We can even pull them from India, the Middle East and Eastern Europe for 30K on H1B’s.

            As far as were concerned in management, your nothing but a mechanic.

            O, and as far as the gui and network crap goes, good luck with that. Its a bit more complicated now than it was 20 years ago.

          • Steve, looking at the quality of recent engeneering gradutes, it seems to me that the young ones are “a dime a dozen” because most of them don’t have the right stuff in them. There are some remarkable exceptions, of course, but these fine minds are so used to look down on people, that they quickly turn down engeneering and blissfully ascend to the management cast…

          • @Steve
            Wow, what a serious asshole. Do you believe you deserve a 1/4million income? The hell you do. Not everyone who is intelligent or capable is born from parents that allow their children the opportunity to partake in a public school system. For those children, life is more difficult, and it’s a sad fate that’s hard to overcome in particular situations.
            I’m speaking for myself, of course.

            I don’t take the IT guy seriously, but you can go drown in a mud puddle.

      • Right. Computer engineers do their best to avoid dealing with IT professionals. *duck*

        Cool pics, love it!

        • Wow, why don’t you all get off your F’n high horses… Cool that you can do hardware, and cool that you can create a secure network, and you both need each other because both professions are technical as all get up… thanks for your work, but get the F over yourselves…

    • Absolutely stunning ! Handled lots but never seen one up close and personal.. A well appreciated article! Thank you…

      Regards

      Gav

  2. Nice work. I especially like the bit where you had to hit it with the saw.

    • Brute force is almost always necessary in some area of a dissection, and I’ve learned to see the grace in it! Great post, great pictures.

  3. Amazing. Thank you very much!

  4. Amazing! Thank you for doing this!!

  5. The “professional” way is to use acid. You can find some cool pics of reverse engineers who’ve opened up NAND memories to find counterfeits and such. Check this blog out: http://www.bunniestudios.com/blog/?p=918

    Here’s a how-to:
    http://www.instructables.com/id/How-to-expose-the-die-contents-of-an-old-CPU/

  6. When I first started looking at the pictures I thought it was gonna be something silly, but wow, these are fantastic. Crazy how tiny those things are, I guess now I understand what they meant by micro processor.

  7. Great work. a layman reveals the secrets of chipmaking!

  8. It’s nice seeing such pioneering research in action.

    • One it is not in action and two that is not pioneering even when it was new. Intel stole the Pentium chip design from AMD via corporate espionage. They acquired the PII from a little know company named Intergraph finding out about its existence via corp. espionage and then buying the fab it was being developed in thereby acquiring all tech within. The P3 which is in the pictures came ultimately in 6 different layers, cache on board, off board etc all in an attempt to hide the real fact that the P3 was nothing more than an overclocked P2. P4 was good but it was complimented by RD memory and later DDR memories. Too soon to tell who Intel stole that from or if they finally invested in an R&D department.

      • 6 different flavors not layers (sorry typo). So uhhh yea Intel wasn’t a pioneer back then just a thief. It is still hard to tell how much of a pioneer intel is (for example after trying for nearly 10 years to produce a decent and cost effective x64 CPU they ultimately decided they suck and to this day pay their main competitor AMD for rights to use AMD’s 64-bit architecture.

        • amd-64 sucks. AMD just extended a convoluted and very outdated architecture to 64 bits. This is why ARM is eating Intel and AMD in the cell phone marked. x-86 has a lot of baggage from years of CISC and segments and what really is required is a small low powered multicore architecture that is efficient and scalable which is Arm.

          • AMD-64 is pretty much the same as Intel-32, just added NX and some other enhancements, and made it 64-bit, and simplified segmentation. Some ARM processors support memory segmentation also (http://en.wikipedia.org/wiki/Memory_management_unit). Memory segments allow more memory and add security to systems. ARM has been around for decades just like x86, and was designed for embedded from the start. ARM is not eating x86 market share, when x86 was invented there was no mobile market, smartphones are relatively new and are a completely new market. There are ZERO ARM desktops or notebooks in the world. Don’t get me wrong, I like the ARM architecture, and use it in robotics designs, but every processor design out there has a purpose for which it was designed, and to act like one is always better is pointless.

          • Joe is absolutely right. ARM relies on RISC where Intel & AMD relies on CISC. For mobile computation Intel & AMD powers are too great, moreover like a overkill. But like spiderman said “with great power comes great power consumption”, ARM RISC processor was used in embedded. Now today, due to smartphone & tablet market it’s look like ARM is eating away Intel & AMD.

            Now, imaging you are rending a scene in the ARM based tablet … :D

        • I don’t know about the Pentiums, but Core 2′s and Xeons (on the same architecture) are clearly better than AMD’s designs from the same generation – a quad core Xeon matches an octa-core Opteron in performance.

      • The P3 is not a P2 it has a new version of SSE which was developed by Intel. AMD wouldn’t have been able to produce an x-86 clone if it weren’t for them buying NexGen!

        The picture counteracts what your saying. Semiconductors are not exactly a recipe a sous chef can cook. AMD is just an Intel Knockoff. Intel has always been very innovative in technology. They developed Hi-k semiconductors and the first DRAM and 3D transistors and they produce their own sub micron semiconductor processes. Intel also developed the cartridge edge SECC slot for the Pentium which although it is no longer used was innovative. The Pentium architecture was also a step up from the 486 and no one but Intel would have the muscle to extend the x-86 as far for 20 years with all the CISC baggage.

        Intel may have used some of the patented features of the Intergraph but Intergraph no longer makes chips. Intel settled that lawsuit a long time ago. AMD would have been smart to take advantage of the Intergraph lawsuit. Intergraphs semiconductor process was probably so inferior to Intel’s Intel would have no need for it. If Intergraphs tech was so good you think they’d still be making chips! If anyone needs espionage it’s the other way since everyone is trailing Intel and Intel’s 8+ billion capex expenditures. The Dec Alpha Engineers were impressed by the Intel semiconductor yields which they couldn’t believe were as high as they were. This was at the time when DEC had a good fab and the Apha. Intel was also one of the first to move to 300mm wafers. Everyone else by now besides Intel is going fabless including AMD. Spansion anyone? At least Numonyx is in business. Micron courtesy of Intel is a major RAM maker.

        The P4 was a power sucking disaster.

        • AMD was in the PC processor buisness long before buying NexGen.

          But saying that Intel is not innovative is false. They developped some very good processor pieces, starting with the 4004 to the current processors.

          They succeeded in the market, because IBM selected their (fairly bad) processor over some more innovative competitors for some reason that I didn’t research. I’ve started working with microprocessors around the time of the 8080. That was the “CPM” processor and it was fairly popular at the time.

          IBM and the success of the IBM PC created the Intel power it has today. At some stage, “innovation” moved from IBM to Intel and his pair Microsoft.

          A lot of effort from Intel went into RD, just to get rid of competition by implementing some new patented features that the cloners weren’t allowed to copy. All of them went out of business respectively were swallowed by some other companies, except AMD who really managed to go it’s own way. There, Intel tried to use anticompetitive behavior to get the competitor down.

          The DEC Alpha processor was the most advanced of his kind at the time of its announcement, but delays devoured much of the advance. One of the incredible unbelevable features of the first Alpha processors were a clock speed of unrivaled 100 MHz. Intel acquired the Alpha processor at a later stage, and they did to acquire the know-how behind this beast.

          BTW: The PC market is by far not the most important one for microprocessor companies, but it is the most visible. Only reacently with the hype of mobile phones, one very innovative company could rival Intels popularity: ARM. That company evolved from a small computer manufactorer (the very successfull British BBC Computer) to a first grade IP company.

      • > Intel stole the Pentium chip design from AMD via corporate espionage.

        Actually, this is incorrect.

        Intel made the Pentium and AMD couldn’t make a Pentium so they bought NextGen for it’s x-86 core so they would be able to compete with Intel. AMD is an Intel knockoff.

        • It’s not a knockoff when it’s licensed. Listen to you all argue. AMD and Intel cross-license MANY things. You all just want to bicker about what you believe (or think) is true.

          If either of them were genuinely a knockoff of each other, injunctions and law suits would still be occurring. They’re competitors that license each others technologies and when it comes right down to it, Intel should be the innovator considering how much bigger the company is than AMD and how much larger their R&D budget is and how many fabrication facilities they own. All things considered AMD has done pretty well for themselves.

          Those who hate on AMD should think about what their Intel chips would cost if there was no competition at all.

          To the OP. Thanks for the pictures. I’ve never had the opportunity to play with an SEM.

          • AMD started the x86 business as a licensee of Intel, because IBM wanted a second source. Since then, they were very successful in that market, so that at a later stage Intel (cross) licensed AMD technology.

  9. Fascinating. I’d love to see an I7 under a scanning electron microscope. A+ exploratory work.

    • Buy him an i7, I’m sure he’ll consider it ;)

      Better yet, buy him two. That way he can use one and put one under the microscope!

      • Naah, simply buy overclocked and burnt ones, that oughtta be quite cheap.
        Bonus points if you then actually manage to figure out where exactly they’ve been burnt beyond operational ;)

        Bastard points for shredding such a nice P3/800 (fortunately it wasn’t one of the even higher-clocked 900/1000/1100 ones ;) . Should have taken one of the ghastly P4/1700 (or some such) crapware things instead.

      • Lol, yeah definitely have to do two ;) I know I would hate to destroy an I7, even in the name of science, when my computer is a Core 2 Duo.

        Awesome pics too, I have never seen a processor up close like that!

  10. Very nice images, and an interesting break down. One thing to consider for future work is shooting the SEM images at even mags such as 2000X instead of 1820X. Its something that alot of engineer types prefer (we’re anal, sorry bout it) and makes things a little easier for calculation and looks a little more professional. Again, very nice images, I’ve never thrown a processor in an SEM before.

  11. Amazing. I’d love to see the process that a chip is made through an optical microscope. In slow motion.

  12. That’s really interesting. I’ve opened older cpu’s several times, but they all came in ceramic packages and could be removed by heating on a gas stove to get the lid off and then remove the chip. Those had the connections around the edge so there was more circuitry to see. I’m really surprised that they now use connections across the full area of the chip, but it makes sense. I’m going to have a look at a cpu like yours soon, but use acid to strip the polymer away.
    Thanks for the excellent microphotography!

  13. Incredible images. there is a majesty in scale whether minute or astronomical

  14. I used to work for a company that made equipment to remove the back side (blue side in your images) of the silicone chip but leave the chip functional (I think it was done for heat mapping of the chip while running, among other things.) It was very interesting going through the various encapsulation layers, of course we never went through the actual semiconductor layers, just thinned out the silicone over the top of the transistors – the idea was to mill a small window through which to view the chip while it ran so you could not touch the actual circuit.

    If you want to know the company e-mail me, they might work with you if you can interest them in it ;-)

  15. actually this is not what i needed when i searched on google which routed me here.but actually i needed more information on this topic for my neice as she had asked.

  16. Amazing find! An insight into a hugely complex microscopic world of microprocessors. :P

  17. It’s the Tron’s city. It’s real!!

  18. Fantastic trip. Never saw one of those images. Imagine a new gen. processor like an i7. Thanks for the share.

  19. Oh, man. I’m going to pull my Phenom II X6 out now and get the Dremel.

  20. You can get silicon wafers that are rejects from manufacturing.

  21. I am happy to say that I deal with this awesome stuff every day.
    Yes, I am a product engineer and I deal with making the product come to life at silicon level.

    Also, it is appalling to see so many “IT” professionals dont know whats going on inside the silicon.
    OP, let me know if you need more information about the stuff you are seeing. I help you out.

    • There is no need to know anymore how the processor works or even the circuitry around the processor. Thats the way of life. The computer operator, database programmer, systems manager etc don’t need to know the internals, but sometimes it helps, if you where, at the time, looking at the spikes.

  22. woww, really a nice images, thnxs

  23. Wow ! This is amazing ! Thank you so much for doing this !

  24. Wow, an amazing sight. Is it even possible to photograph if you get a ‘deeper look’ – All this in a P3… Wonder what an i5 looks like, heh heh.

  25. Nice pics!

    You can use an HF based solution and mechanical polishing to go down some layers on the CPU.

    If you have access to a SEM, I guess having access to a polisher won’t be that difficult :) And I guess not very far from the SEM you’ll have a chemical lab where you can work with acids hehe

    I’m looking forward to see new pics!

  26. I recommend for the removal of the top layers the so called “selective chemical etching”, the ideal case would be to have a diagram with the different layers and what semiconductor is each epilayer. Once you have that you can find the chemicals that will remove only certain layer leaving untouched the others. (i spent my whole PhD doing these sort of things). Gimme a shout in my email and i can give u references for selective etching chemicals for each layer.

    PS: great post!

  27. PS2: i would think its the plastic causing the focussing problem (as u mentioned by being non conductive and creating a chargin up effect), i did some test in the SEM and didnt like plastics much :P a not so nice solution is reducing the current of the electron source for minimising this effect.

  28. Truly am grateful what you’re doing here!

  29. You might want to look for 3D reconstruction from optical microscopy depth focal series. There is free software out there to do it.

  30. There’s a guy with an afro visible in the center of the second to last photo. How odd.

  31. Oooh, so that’s what you guys are doing over at the nano lab. Great stuff. :)

  32. Very cool pictures!

    I work at an Institute for Quantum Computing and we have a clean-room where they fabricate silicon chips. They don’t create incredibly detailed CPU’s but some of the circutry is down to 20-30nm.

    FYI, the CPU you scanned was a Pentium III EB. It was on a 180nm process and known as the ‘Coppermine’ family of chips.

  33. A silicon chip is literally the equivalent of putting a street map of Los Angeles on the head of a pin, complete with three-dimensional freeway overpasses. But semi-conductors are extremely vulnerable to heat, so I doubt they heat the chip itself to connect it to the board. If they do apply heat, it’s likely extremely localized and brief, and just enough to join the solder. Possibly they heat the board connectors instead? Anyone know, or is that proprietary?

    It’s amazing how clunky and crude the connectors look compared to the chip itself. Wonderful pictures, especially the next to last.

    • I’ve been wrong many times before, but I believe you will find that the circuit connections are made by using microwaves as a heat source – process developed by a German company in the late 80s early 90s, this allowed PCBs to be layered. Because of the layering, it made the modules created impossible to repair under field conditions. – there is no way to get a soldering iron between the PCB layers.

      Have to agree, great post!

  34. i have never imagined that a cpu is so complicated!!. Front page of reddit, sir.

  35. greg – the mags are irrelevant and, speaking as a scientist, shouldn’t be used (its only relevant if the screen/format you are viewing them on is exactly the same size as that of the screen on the instrument which in most cases, it isn’t. For instance, 200x here on the web page isn’t 200x on the SEM screen ). The only thing that really matters is the scale bar – more often than not I remove mag from the SEM display.
    If you do measurements using only the mag you will more often than not get the wrong number…..

    Nice pictures though and nice blog!

    As you mentioned, theres a bit of charging in some pics – if you ever do this again I would drop the accelerating voltage a little, and or drop the beam current and run under Variable Pressure mode (which usually negates the need for gold or carbon coating).

  36. that is pretty neat. I’ve never seen a chip like that before.

  37. Wonderful pictures mate.
    I remember visiting the Intel site in Ireland, and they gave tours showing the CPUs being built.
    Apparently the old CPUs (like this one) if it had its picture taken of its circuitry, could fill a large warehouse floor, and still be required to use a magnifiying glass to see to the lowest layer.
    The Sample Normal pic above looks almost like a photograph of earth from an airplane (city lights).
    Brilliant work – let us know if you do any more work like this :)

  38. I swear, I thought I saw Tron in one of the pictures.

  39. Well done! I enjoyed these photos.

  40. I’d love to see a Socket A Athlon. Or any chip, really. And pictures that compare it with something ancient like a 486 so we can see just how much feature size has changed since then.

  41. Really cool! I have assembled pcs and dismantled them but never disassembled a processor. Looks awesome!

  42. Thank you so much for this, I have tried to find this forever! Finally, I know what an inside of a CPU looks like.

  43. Hi –

    There are various ways to etch silicon, silica, and the other materials that make up the chip

    I would start by using fuming acid (a mixture of nitric and sulfuric acid) to dissolve the epoxy.
    Then you can use HF to etch the silica intralayer dielectric, which should show the damascene copper layers

    Your SEM images can be improved by either operating in low vac mode or by sputtering a conductor to reduce charging. Or you can reduce the beam energy; silica is a light element and your interaction volume will be quite large inside of it, reducing your potential resolution with beam energies as high as 15kV

  44. As an IT Pro who has actually soldered resistors into pre-drilled circuit cards (1968), and closely followed the science since then, I am very grateful for the work you did here. This a very solid contribution to the History of Computing and I hope that this work is included in that body of work.
    Thanks again,
    Cliff

  45. On first looking at it I was struck for an instant as to how much it looked like and overhead shot of the 1942 Dieppe Raid that failed then in looking as the slid show was amazed and how they got all this. It makes me pause and think that this is no grand accident here, not by far no. Their is or was another hand at work in this that is plain to see…………..we truely live in an amazing world

  46. You were looking at one of the 7 metal layers and their interlayer dielectric (ILD) insulator that keeps them from shorting. This is probably why you cant focus on it well. There is very little Si until the very bottom…

    The easiest way to see the underlying structure is to etch in acids or bases to remove the ILD or metal.

    HCl + H2O2 at 80C will work to remove the ILd but could damage the metal. HF would also be a good choice. Si is unaffected by most acids…there are companies that do this for every Intel process called chip works. Perhaps you can find something they use for decapping and cross sectioning

    • I was planning to further strip the ILD layers. I guess HCL+H2O2 could work but I would prefer using HF. My first choice would be to remove any polymers using e.g. an oxygen plasma in a RIE system. Thanks for the info!

  47. wow! very nice experiment! =) it’s my first time to see what’s inside!! ^^

  48. Very Good – But Why a P III, (about 2001-2 vintage)? Why Not a 2010 Intel chip?

    PS: I used to manage a failure analysis laboratory where we looked at chips to determine why they failed.

    John

    • Well, it was the chip I could find on short notice when I needed something to look at. Principle is the same, and I doubt the layman would spot any difference between a 2001 and a 2010 chip.

      • The difference would be that the 2010 processor cross sectional features (metal runners, contacts, would be around 10x smaller than the 2001 processor. Serious advancements have been made in the shrinking technology.

        I work in a failure analysis lab, if you want a much much cleaner cross section, you can polish the chips interface on a grinding wheel. Very effective. Also, etching the polished interface with a buffered HF solution would make for a much nicer image as the dielectric films would be slightly recessed.

  49. An eye opener, amazing voyage into the processor. We read a lot and saw lots of schematic of microprocessor, this shows, how they really look like inside. Great job.

  50. You sir, are a god among men. Electron microscopes are the kind of thing that are well beyond that practical use of most people, but if I had one I would be breaking apart everything I can find just to see how it ticks.

  51. This is really good! Thank you.

  52. Awesome. Reminded me of TRON.

  53. Incredible! Now lets do a similar exercise on brain cells and see how the Almighty’s chip is created.

  54. Thankyou so much fro this post. I’ve learnt a lot and has really inspired me to do more of my own revers-engineering. Thankyou again.

  55. Great job, man. Thanks for the photos!

  56. I don’t know much about microscopes but this is impressive thianks for the work. Warm regards, Frank

  57. Welcome to the Matrix..

  58. I recomend you and old Pentium, you can remove the cover with a plumber torch and see the surface like this.

    Much easier than taking apart the silicon from a PIII ;)

  59. Muy interesante.
    Gracias.

  60. Cryomicrotoming the processor will expose the bottom layer, but unless this is a field emission SEM, you won’t get good enough resolution. You could also use HNO3 to etch away the metal, but be sure to wash first if you do that. Your best bet is to thin section it with a cryomicrotome and put it in a TEM layer-by-layer to get all the detail

  61. Inspired bit of sleuthing. Love the SEM images. Perhaps you can follow up with other computer components sliced and diced.

  62. Awesome job the pictures look great, not quite what i expected things to look like though

  63. That is really cool. Thanks for sharing!

  64. Well, it is intriguing.

    You may “speak science”, but you have Fists of Ham…

  65. Now all we need is a google processor maps…

  66. I am taking a CMOS/VLSI design class in college. It is interesting to see what my designs actually become once they are fabricated.

  67. thanks for sharing the photos.

  68. Have a look on the search page of Sparkfun, perhaps you can get some more ideas on how to get down under with the chip? http://www.sparkfun.com/search/results?term=atmega328&what=news

    There are three separate articles ,”Revisiting the Counterfeit ATMega328s”, “Dear ON Semiconductor” and the final one “Fake ICs Identified”.

    An ATmega is still a lot larger than an Intel CPU.. but still, it should work for it as well.
    Hope to see some progress on this! Very nice pictures so far!

  69. I cant believe this shit man! Thats the inside of all our shit?

  70. delightful.
    this is a treat.thanks .

  71. Nice to see what happens inside a little chip

  72. Do a Geforce next chap, i bet they are busy inside

  73. great job, i was thinking that there are only Si particles inside, never expected this much of complexity …

  74. Please update this. I need to see farther down the rabbit hole.

  75. Sweet pictures! Do you teach at a college? I assume this is how you have access to the Scanning electron microscope.

  76. How did anyone every come up with this technology, im such a monkey!

  77. How on earth are they making it? I can’t believe how small everything is inside of it.

  78. Awesome… This shows how powerful the human mind is!

  79. I think if you uesed phosphoric acid that would allow you to get to the lower levels. The amount of time would proably require some trail and error.

  80. WONDERFUL!!!
    Even i have tried that but i had to stop as i had no optical or SEM But you my Friend have Made my wish come true!!
    I’m Very grateful to you!
    Thanks a million!

  81. Awesome! But there is no continuity between the picture that says “Focus is on the bottom layer” and the ones after that. Can you tie them together by showing us a zoomed out picture to orient us? I know it’s a cross section (whereas the ones before were not) but the surface at the top looks nothing like anything that you showed from the top view (with the holes), so I’m confused.

  82. This is simply spectacular.
    Thanks so much for putting the effort in to show us this!

  83. Excellent stuff; much appreciated.

  84. an alternative to etching could be to try polishing gently through the layers (using a rotary polisher). You’ll certainly need to make sure your sample is either earthed in the chamber (sputter coating) or use a low-vacuum mode to get to the nm scale. I’d opt for the latter if available, sputter coating might obscure detail.

  85. Awe freaking some!!!! Maybe you some how eat away at the lead? And then you would know where the transistors would be????

  86. Very beautiful, alone I have seen with a maximum of 40X using an optical microscope, personally I use fire to liberate the chip of the encased one without damaging it.

  87. Great post =) and way much informative!

  88. Years ago… probably 1986ish, I *believe* in Science Digest, I saw a great article by the guy who was in charge (a pun you’ll understand in a minute) of quality control at Texas Instruments. They had created a state of the art plant for making calculator chips, with a testing rig for each wafer, and they would reject the whole wafer if there were any bad tests on it… so they had 100% good chips going into their calculators, but then were having about 30% failure rate after purchase — incredibly bad and bad for business.

    Long story short, they were going to great lengths to make sure the testing rig was at zero potential relative to the wafers on the way down… but after running the tests, capacitances all over the wafer would micro-spark as they lifted the rig off — I think they said it was essentially static electricity of about 1/5th of a volt, whereas static we can feel is about 200,000 volts and up… but anyway, the article contained beautiful electron microscopy of the chip/wafer surfaces, and it looked like World War III — “giant” hunks of molten metal sprayed across moon-cratered blasted surface of the chips… creating unplanned for electron tunnels between layers which would oxidize over time and create new pathways leading either to complete failure, or, more disturbing: 1+1 = 3

    So, my challenge to you all is — can you find the article I’m talking about? I’ve been trying since digitized journals starting showing up on the internet, and I’ve never had any luck. If someone finds it, post a link here, it will be well worth it!

  89. Believe it or not, there are still others who are using pentium 3 computers :D

  90. Great work.
    Thank you for the insight.

  91. This takes me back to 1982. While in the last year of high school I dissected LEDs, 74XXs and CD4000s with a grinding stone at the metallurgy lab. I could borrow an optical microscope and look at the structures… what a wonderful feeling when I could identify the aluminium connections and the long, skinny wells of the CMOS process depicted in text books. I could not get a camera attached to the microscope, so I can only rely on my memory. Thanks for sharing your work.

  92. Hello. You can try to cut slices from your processor using the chemistry. There are some solutions that permits the corrosion of Si at decent temperatures, like NaOH in water. I dunno exactly the concentration but you can search on google. There is more than Si in that structure, there are metals, SiO2 and other materials. You can try to find a layer of cleavage on that and try to remove all what is up that layer. Sorry for my english, I’m not a native speaker.

  93. Dude this article is awesome!

  94. Beware of programmers carrying Dremel tools… that’s a nice electron microscope. Can I have it?

  95. As a computer engineer and as a gamer, I’m astonished. Never took the time to know the layers that communicate the different levels of transistors. I’m familiar with wafer dies and schematic blueprints for integrated circuits, but not with the manufacturing process for processors.

    A cool topic indeed.

  96. Zoom! Enhance!

    So I take it you are the guy from CSI?

  97. You need to use an acid to etch away the top layers. Hydrochloric can eat the metal but will take a lot longer on the silicon. Hydrofloric acid will eat the silicon easily, but is VERY HAZARDOUS. To dissolve the plastic, try chloroform or some other hydrocarbon solvent (acetone or equivalent will NOT work)

  98. To dissolve the plastic, try chloroform or some other hydrocarbon solvent (acetone or equivalent will NOT work). You need to use an acid to etch away the other layers. Hydrochloric can eat the metal but will take a lot longer on the silicon. Hydrofloric acid will eat the silicon easily, but is VERY HAZARDOUS.

  99. Wow!! Will need to repost this to my german readers – thank you for sharing, never had such a deep view into a chip!

  100. Great work and cool post! Tnks!

  101. I learned about microprocessor architecture & fabrication in varsity as a course. I also saw many many picture of how Intel chips looks like ( everybody did :D ). But nothing like this !!!

    This is one hell of a view of the actual processor.

    Kudos for your great work, keep it up.

  102. Great job man.
    Now I wish I had such a SEM to viewing my old chips a little closer :-D
    Please let us know if you got more pictures from the transistors.

    Have a nice day,

    f0rk

  103. Very nice pictures to see and an awesome job.
    If you need some cpu’s to dissect just drop me an email and I’ll send you a few.

  104. The chip you broke apart was built with the 858 180nm process. The interconnect wiring is AlCu alloy with Tungsten via plugs. The Dielectric (insulation is Flourinenated Silicon Glass and regular SiO2. The large blobs you found when peeling the PCB from the Die are called bumps. Basically they are blobs of lead/tin solder that are formed outside the fab in a special area due to the hazards of lead contamination. Originally they are made flat for e-testing then heated to “reflow” and then they form a round balls that easily bond to the pads on the circuit board this referred to as a “ball grid array”. Many chips today are sold as “chip scale packages” that are prepped like this and bonded directly to circuit boards such as in our smart phones today.

    • Thanks! This is very useful information! At some point I will post a follow-up with the dielectrics removed. With all the information I have from the comments, it should be fairly easy to etch away all the insulating material.

  105. It “caught my eye” again. Would be interesting maybe to look at a DVD at nano scale.

  106. Waw, awesome..
    i had many times see this, but only on CG..
    i never see the chip in real photo, this is cool..

    how about trying to other chip? Like a Core i-series ??
    we can see the different about those chip, at different manufacturing process
    that’s will be more interesting :D

  107. please hold… the black helicopters will be arriving soon.

  108. Amazing. I wish to be able to see a transistor image one day, hopefully you’ll make my wish come true.

    I always had a question, how did manufacturers worked at such a low microscopic level, I never could comprehend it.

  109. Amazing, everything looks so incredible close up, the inside of these chips are unreal, its hard to get your head round that something so tiny is made by humans!

  110. seriously awesome seeing all the stuff “going on” inside a chip – and an old chip at that. i can just imagine whats a core i7.

  111. Thanks for the pictures, it was super cool!

  112. You want invoation, go to IBM or Motorola. Both companies that threw bigger bucks into experemental chip technology than intel could even dream of back when it was all happening. The old 68K was one of the most amazing processors ever made, IBM had their stale period with the power architecture but in the end is still pushing the envelope by challenging the absolute edges of sanity with its partnership with sony/toshiba producing cell. and again producing a competely seperate but inovative processor for the Xbox 360 not to mention massive leaps and bounds in the realms of super and near-super computing. not a perfect company by any means, but way out there in terms of development.

    In comparison companies like AMD and Intel are just twiddling their thumbs and squeezing incrimental gains from maturing process technologies and optiomisations in a venerable architecture foundation dating back now more than 30 years

    im actualy impressed to see AMD recently sticking its balls out in the cold and trying some different concepts mixing up the tech they acquired from ATI. APU’s, microcores and the IP to actualy challenge the validity of X86-n as an architcecture

  113. You blew my mind. This is fantastic work. I very much appreciate it. But we have to go deeper !


Leave a comment

(required)