Interesting double-poly latches inside AMD's vintage LANCE Ethernet chip | Hacker News Reader

Interesting double-poly latches inside AMD's vintage LANCE Ethernet chip

righto.com

148 pointselpocko17 comments

Author here for any questions :-) This is a change of pace from my x86 stuff, just a quick look at a 1982 Ethernet chip that unexpectedly used a second layer of polysilicon for resistors.

Great article Ken. I love the chip/die pictures.

The Commodore A2065 ethernet card uses the Am7990, and I'm still using several of them in various Amiga systems, running both AmigaDOS and NetBSD.

While trying to figure out some occasional network stalls, I looked at NetBSD's man le(4) page, which had an interesting note:

Alas, the Am7990 chip is so old that AMD has "de-archived" the production information about it; pending a search elsewhere, we don't know how to identify the revision C chip from the date codes.

Any chance this information is publicly available elsewhere?

Always fascinating writeups, Ken! Thanks :)

Sorry, I don't know where you would find that information.

Not sure how much this helps in identifying the actual chips, but, according to these datasheet sources, Revision C is dated June 1990 and Revision D is dated January 1992:

https://archive.org/details/bitsavers_amdAm7990A_1770429 (Rev. C)

https://www.ardent-tool.com/datasheets/AMD_Am7990.pdf (Rev. D)

Happy New Year, Master Ken! Thank you for all the gifts you've given us this year, even on the last day! Cheers, from an island Brazil!

No questions, just want to say thanks for your amazing series of articles. I learned a lot from reading them.

I love your work. Unrelated to this piece, do you know of any resource that allows you to zoom in on a modern-process-node chip, from 1X down to the transistor level?

Some ASIC designs are open source. That won't be down to the transistor level, but they are laid out at gate level. For example: https://tinytapeout.com/runs/tt02/

While the tool creates the chip by placing at the gate level, the designs themselves can be studied at the transistor level if you want. You can study the actual chip gds file or look up the gds for each standard cell individually.

If I remember correctly, AMD LANCE is also one of those NICs that became commonly implemented virtually for VMs, along with NE2000, Intel's 8257x, and one of the popular Realtek models.

The double-inverter loop is like the electrical analog of a "snap action" physical mechanism[1]; it's bistable, refusing to stay in an intermediate position, and requires force to change its state.

[1] When https://en.wikipedia.org/wiki/Pop_it toys became popular recently, the number of people who thought "that's a bistable memory" upon seeing them was unfortunately not many.

qemu doesn't seem to emulate it unless it's called something else:

  $ qemu-system-x86_64 -device \? | grep -i lance
  $ qemu-system-x86_64 -device \? | grep -i 7990

Back in the day (early 90s) LANCE was extremely common in embedded systems. We had a whole variety of Microware OS-9 machines that had LANCE chips for ethernet. So if you wanted to emulate that era of software it would make sense to emulate it.

It's in hw/net/lance.c -- that device doesn't appear in the -device help output because it's not a user creatable device, it's a sysbus device. The PCI card device that wraps it we call 'pcnet', and we use the sysbus device directly in some of the sparc machine types.

It's called "pcnet". Technically a 79c970, but backwards compatible to the original 7990.

VMware emulates one too, and so does VirtualBox.

I'm always fascinated with die shots that look like artists drawings (black pen on purple/blue felt tip marker fields) as shown on a tube TV (slightly blurry/color fringe).

But no, it's not a drawing, it's an actual photo. Well, I guess it's slightly blurry because of the magnification, but it's such a surreal look. Love it!

What I find interesting is that the colors aren't "real", in the sense that the materials aren't colored. The colors are mostly from thin-film interference due to the thin layers of material. And the black lines aren't "real" either; although the layers look flat, they form three-dimensional structures as they overlap. So many of the black lines show up because they are edges in three dimensions, not because they are actually black. Another consequence is that different chips look completely different due to the manufacturing process that was used, even though the bulk color of the materials is the same.

Of course, the images would be sharper if I had a more expensive microscope, since the quality of the optics matters too :-)

One has to be careful because there are several sources of “color” and it’s hard to argue successfully that any of them are more “real” than the others.

Reply on news.ycombinator.com