Inside our failed Intel 3242 chip (Patreon)

After our last repair session of the HP 9825, Ken Shirriff went home with our failed Intel 3242 DRAM controller. He has since decapped the chip and is tweeting about his findings as he discovers things. You can follow him on Twitter at https://twitter.com/kenshirriff . This is mostly from cutting and pasting from his thread.

Chip from outside, with both Intel D3242 and HP 1820-2076 markings:

Chip from inside:

Here's a die photo of the chip after removing most of the metal and oxide, showing the transistors and resistors underneath. The chip uses bipolar NPN transistors. They are Schottky transistors for high speed. (If you've used 74LSxx chips, that's what the "S" is.)

At the time, 16-kilobit dynamic RAM chips were popular. They needed 14 address bits but to save on pins, they used 7 pins, splitting the address into two parts. Also, dynamic RAM required refresh: you needed to periodically re-read data to prevent it from fading away.

This Intel chip simplified things. It split the 14-bit address into two parts for you. It also had a 7-bit counter to refresh memory for you. So you could replace a handful of memory support chips with this memory controller chip.

Here's a simplified diagram of the chip. In the center, 7 blocks form the 7-bit refresh counter. Around the edge are the 14 address inputs. Diode-transistor-logic gates select the row address, column address, or refresh address. At the left are the seven output drivers.

The chip's logic is a mixture of TTL, DTL, and weird bipolar latches.

This closeup shows one of the TTL transistors with four emitters (the four dark spots). This computes a carry for the high-speed adder. Base connection at top, collector at bottom.

Here are some zig-zagging resistors. Unlike CMOS, bipolar logic requires many transistors, which take up a lot of space on the chip. Resistors are used for pull-ups, pull-downs, or controlling base current.

Closeup of the Intel copyright on the chip.

Ken hasn't found any obvious visible damage to the chip, although it is definitely completely absolutely confirmedly dead!

More on Ken Shirriff's Twitter account: https://twitter.com/kenshirriff

Marc