The Ultimate AMD 486 Die & Packaging Guide

[xsecret]

After weeks spent to test A LOT of AMD 486 with the Universal Chip Analyzer, messing with a gas torch to decap some of them and speaking with a former AMD engineer that worked on them back in the 90s, I’m happy to publish here all the information I was able to get! Here it is:

The Ultimate AMD 486 Die & Packaging Guide

PS: If you find something wrong or have any more information about AMD 486 Die and packaging, please leave a comment. Thanks!

[Mixeur]

Amazing article !

[frag_]

Love this type of articles!

I have some questions and comments.

- What is the source of date ranges for different dies/packages?
- What about 25220 A with small lid? It can't be 700 nm, right?
http://www.cpu-world.com/forum/viewtopic.php?p=210820#210820
- Is 24361 B2 really ES? It's not that rare and nothing on the package indicates it's sample.
- 350nm 16K is interesting in that current rises with bus clock, 4-3-2 multi, 8.33-11.11-16.66MHz bus gives 189-191-197 mA.
What would happen running DX2-66 with multiplier x3, probably close to 191?
- Picture with dies have different die sizes than in description (e.q. first one 90mm^2 on the picture but 70 in the text).
Thanks!

[xsecret]

[cvandijk]

Very interesting article, so now we can check better if we have fake ones in the collection

[Vegeta]

Very nice!

[debs3759]

I've made this sticky as it's a very detailed article that I can see being very useful.

[CPUShack]

In that case it should probably be in the Modern CPUs subsection

[frag_]

Yes, looks like that small lid 25220 A was fake. But why did they do it, small lid ones are at least 66 MHz already.
Need to check 25220 A ones.

Non-es B2 is here:
http://www.cpu-world.com/forum/viewtopic.php?p=292600#292600

[xsecret]

It looks perfectly genuine.

I don't have an assembly start date, but I know that AMD started to ship Am486s in WW18'93 and I only have a first assembly-date for C-Step in WW24'93, so there is a gap between WW18 and WW24 where they must have shipped retain B2 Step.

I also have a B2 ES built in WW17'93, so it seems legit that retail B2 shipped for some weeks just after launch. I will edit the guide right now.

[Neon]

Good job putting together this guide - also very good that you were able to speak with a former AMD engineer, so that some of the information could be verified.

I have compiled date code ranges for these packages and die steppings, using examples from my collection and others viewable online. These date codes are my empirical observations on actual samples, and are not based on knowledge of actual AMD production run dates. These dates mostly agree with yours, with a few minor extensions to some of the timelines:
1. earliest 700nm die B2 stepping date 9302 (January)
2. earliest 700nm die D stepping date 9338
3. earliest 500nm die A stepping date 9445
4. earliest 500nm die C stepping date 9530
5. earliest 350nm die B stepping date 9711
6. earliest 350nm die C stepping date 9550
7. latest package 24361 date 9505
8. package 25255 was produced for at least 9 weeks, dates 9519 to 9528
9. package 25498 was produced at least until date 9642 (October)
10. package 25498 A stepping was produced at least until date 0033
__

Usually, it seems that die steppings are an indicator of progressive improvement in the wafer manufacture 'recipe'. Thus, dies made with stepping B would have new features or better average performance than dies made with stepping A. Then, C better than B, and continuing on so.

What about the B and C stepping versions of package 25372 then? The stepping C / 8 KB have dates 9644 to 9646 - a year later than the B stepping / 16KB variants - indeed, they appear to be the latest 500nm dies (except for 25498 stepping A).
One possibility is that when production stopped of 16KB on 500nm, some wafers were unfinished, put aside for a year, and then half cache disabled down to 8KB, to get acceptable die yields. In this case, it would seem that the C stepping designation does not represent a change in wafer manufacturing recipe from the B stepping, but just a way to differentiate them from the 16KB versions.
A second possibility is that 8KB / 500 nm / stepping C proceeded during dates 9530 to 9642, and then there was a brief last attempt to produce a second batch of 16KB wafers on the mature 500nm stepping C process during dates 9644 to 9646. In this case, the C stepping designation would be a newer manufacturing 'recipe' than B stepping, so it makes sense. However, at that time 350nm production was already ongoing, so why bother with a second batch of 500nm 16KB wafers?

[xsecret]

Thank you very much for your insights as they helped me to double check everything and to find some errors and typos.

Here are some comments about the date code first:

Agree (Fully or within 2 weeks)

2. Fully Agree
3. Fully Agree
4. Fully Agree

5. For 25544/B, I have 9709, but it's almost the same. Very close
6. For 25544/C, I have 9548 as earliest retail. Very close

Now Agree (mistake on my side now corrected).

8. Agree. I got the same range lately and forgot to update it.
9. Agree. A second batch as been made lately. Last production date should be 9644

Conflict

1. May I ask for some info (pics, markings, etc.) about any non-ES 24361/B produced before 9313 as they clearly conflict with the data I have (that's earlier than the initial tape-out) ?

7. Any infos on 24361 produced in 1995 are also welcome as I have a last production date in November 94.

10. I don't have data from AMD for this "special" stepping. Date range was grabbed from pictures online. I already saw two 25498/A with a date code in 2000. Both were 0033 (never seen any 00xx not 0033) and one of them looked fake as hell.

---------------

I asked about the elusive 500nm/16KB cache die in 25372 package but I only got very few information. Actually, my source never heard about them. That's extremely strange because he worked on the 350nm/8KB die that never reached the market after tape-out. But this later die was designed as a general market (embedded industrial) die.

My guess is the 500nm/16KB die was a custom order for a big customer. Only one batch as been produced but yield was very poor and only a few were tested/packaged as 16KB die. I don't know if they fused-down later the already-produced wafer but it makes sense.

PS : Your stepping table is very clear. Can I use the same layout and publish it on my website?

[Neon]

OK, for the 24361 /B2 I do not have photos, so the 9302 one I must have seen somebody else's photo online. Since you have actual tape-out date later than this, maybe this one was ES.

The 24361 9505 is from my collection. When I get home next week, I will find it and post the photo here.

25498 /A yes I also got data from online - you may be right that the 0033 are fake.

Certainly you can use my format and data. Thanks also for your power consumption work, it helps explain what is going on.

P.S. I also have photo of 25253 A0 ES - only one that I have seen.

[Neon]

Here is A80486DX2-66N E6 9505EPK 24361 (the "1" is there, but did not scan properly)

A80486DXL4-100 A0 25253 ES

[Andreycpu]

For 24361 also Am486SX2-50, Am486SX2-66
For 25220 also Am486SX2-50, Am486SX2-66