Email or username:

Password:

Forgot your password?
Top-level
Ken Shirriff

The 386's logic circuitry is powered separately from the I/O pins. Changing an output takes a lot of current, resulting in power fluctuations. These could cause unreliability if logic circuits used the same power. So I/O has separate pins and wiring on the die. 6/13

17 comments
Ken Shirriff

This X-ray shows the layers inside the 386 package. Thick traces go from the signal pins to the die. Power/ground have metal planes, perforated so the ceramic sticks together. Thin tungsten(?) wires to the package edge are used to electroplate the pins. 7/13

Ken Shirriff

This closeup shows how tiny bond wires connect the die to the package. The package has two tiers of golden pads that are connected to the tiny bond pads on the silicon die. 8/13

Ken Shirriff

The power and ground connections use multiple wires in parallel. This allows more current and also reduces inductance (and thus noise). Intel's package was much better than other 132-pin packages with 1/4 the inductance. Resulting noise was under 0.5 volts. 9/13

Ken Shirriff

Intel's cost to produce 386 dies rapidly dropped. Eventually, the package cost more than the die itself. Intel redesigned the 386 so it could fit in a cheaper plastic package. 10/13

Ken Shirriff

Packaging is underappreciated, but it is essential to making chips small and reliable. In particular, getting power to a chip and then distributing power on the chip is harder than you might think. Nowadays, packaging, power, and removing heat is an even bigger issue. 11/13

Ken Shirriff replied to Ken

Credits and links:
Thanks to twitter.com/johndmcmaster for taking the X-rays.
Package cross section from "High Performance Technology, Circuits and Packaging for the 80386", ICCD 1986.
16-pin quotes from Federico Faggin archive.computerhistory.org/re
4004 photo by Thomas Nguyen (CC BY-SA 4.0 DEED) commons.wikimedia.org/wiki/Fil
386SX photo by FDominec commons.wikimedia.org/wiki/Fil 12/13

Credits and links:
Thanks to twitter.com/johndmcmaster for taking the X-rays.
Package cross section from "High Performance Technology, Circuits and Packaging for the 80386", ICCD 1986.
16-pin quotes from Federico Faggin archive.computerhistory.org/re
4004 photo by Thomas Nguyen (CC BY-SA 4.0 DEED) commons.wikimedia.org/wiki/Fil
386SX photo by FDominec ...

StarTracker replied to Ken
Dave Warnock replied to Ken

@kenshirriff thanks for this thread. Brought back fun memories.
See amastodon.uk/@Dave42W/11184795

Oh I also remember when I was able to upgrade the office NetWare server from a 286 to 386 again a real game changer.

Anthony 🦘🐨🪃🕯🧙‍♂️🔆🇦🇺 replied to Ken

@kenshirriff wow! An interesting tour through a bit of cpu history. 😃

Peter Kovář replied to Ken

@kenshirriff Chips. Memories. Thank you so much.

mariuz

@kenshirriff I remember this type of cheap package :)

ES Michelson

@kenshirriff I think this is a 386SX chip. A little sibling chip with, I think, 16 bit external data path. Maximum supported RAM was 4MB and I bought such a PC with 1MB RAM as a 'smart purchase.' The PC ran MS-DOS, then Windows 3.1 It had a 40MB HD which, the seller said, would never fill up. With the Windows upgrade the RAM and HD had to be upgraded. The next PC after that was a PowerPC chipped Macintosh. That was an excellent computer in its day but hobbled by increasingly terrible OSs.

tTh

@kenshirriff This is a SX 386, who have an only 16bits address bus and a smaller data bus than the vanilla 386.

Lee Cremeans

@kenshirriff AMD made a PQFP 386DX-40 in a package like this one, and it showed up on a bunch of late-model (circa 1993-1995) cheapie motherboards.

Steve Loughran

@kenshirriff 386SX was 16 bit memory bus, wasn’t it. So needed fewer pins at the cost of memory access times. And the double sigma on some of the packages means “gets 32 MUL eax right” so safe for 32 bit apps….

Go Up