4MB RAM upgrade for C103414-001 Rev 1.1 (520/1040 STFM)

[jwd]

A good video upgrading exactly this type of memory chips



I have it bookmarked as I'll be doing this on mine at some point, i got the chips from a 4mb falcon memory board as the chips were marked the same as in my ST.

You really need a diag cart and see what the output is via serial probably one or more chips not right.

[MaximilianM]

ah thx nice video, and yeah i conntacted my local dealer, maybe he has a diagnostic card... without its pain in the ass...

[MaximilianM]

Quick update: got the Atari Bombs Away diagnostic cartridge and tried it. Instead of the expected diagnostic menu, I get a different error pattern on screen — static vertical color bars, not the usual half-black/half-noise I had before. Reset button does nothing while the cartridge is plugged in (works fine again once I pull it).

Does anyone know which RAM bank or which specific RAM IC is tied to the video/screen memory on the 1040 STFM? Trying to narrow down which chip to focus on. Thanks!

[DoG]

Connect with another computer via null-modem-cable to get output from diagnostic cartridge. RS232-USB to PC for example also work with a terminal program.9600 baud,8N1 (8 data bits, No parity, 1 stop bit). No flow control

https://www.atari-wiki.com/index.php?ti ... dge_basics

(Then the basic: clean port and cartridge with isopropanol, set 'bombs away' to (mega) ST 4.4)

[MaximilianM]

thx! it works, an endless error message...

Bildschirmfoto 2026-09-24 um 18.30.01.png (454.01 KiB) Viewed 99 times

[DoG]

Looks like bit 7 is wrong. Written a 1 and read as a 0. Since you have nothing on screen it's bank 0. So chip U4, DQ4 (pin 25).

But I don't have schematics for that board so check with a multimeter that it's actually bit 7 on bank 0 that is U4 before you desolder it.

[MaximilianM]

Thanks a lot! Turned out to be 2 solder bridges shorting a data line to VSS on two of the RAM ICs — fixed those and the RAM/data line errors are gone (diagnostic cartridge now runs through cleanly and reports "2M RAM Pass").

However, I'm still not getting any video output, and only 2MB of RAM is being detected instead of the full 4MB. I've since confirmed with a multimeter that the A9 daisy-chain mod itself is fine — all 8 RAM ICs show a clean ~33Ω on their A9 pin back to the mod resistor, so that wiring checks out completely.

Given that, is there possibly a second mod needed alongside the A9 line — something related to telling the MMU the actual size/density of the new chips per bank? Wondering if there's a separate resistor/jumper for that, since electrically everything on the A9 side looks correct, but only one bank (2MB) is being recognized at all.

[DoG]

Check CASL, CASH and RAS lines.

Given that, is there possibly a second mod needed alongside the A9 line — something related to telling the MMU the actual size/density of the new chips per bank? Wondering if there's a separate resistor/jumper for that, since electrically everything on the A9 side looks correct, but only one bank (2MB) is being recognized at all.

Nothing else needs to be done. It should work if everything is connected right, no shorts is found and no tracks are broken. Only the A9 needs to be routed using a wire since the track on pcb is non existed.

[MaximilianM]

thx... checked

Bank 1 appears not to be recognized.

* RAS1 at MMU pin 18: clean signal present
* CAS1H at MMU pin 21: no activity
* CAS1L at MMU pin 22: no activity
* CAS0H/CAS0L: normal activity
* MAD9 at MMU pin 64 and A9 at the RAMs: clean signal
* All connections from the MMU through R71/R72/R73 to the RAM bank have been continuity-tested: all connected
* The MMU is the original C025912-38, which worked correctly with the 1 MB configuration before the upgrade, and it is soldered directly to the board.

MMU config showed correct 4M mode (2M+2M), Bank 0 (0-1FFFFF) passed all tests, but Bank 1 (200000-3FFFFF) had a reproducible data corruption.

Instead of guessing, we used the cartridge's built-in memory read/write commands to isolate it bit by bit:

W 200000=0010 (write bit 4 only)
RW 200000 → 00F0

W 200000=0020 (write bit 5 only)
RW 200000 → 00F0

W 200000=0040 (write bit 6 only)
RW 200000 → 00F0

No matter which single bit we wrote, the readback was always 00F0 — bits 4-7 stuck permanently high, bits 0-3/8-15 correctly reflecting whatever was written. That pinned the fault down to exactly one nibble (D4-D7) — and since D0-D3 and D8-D15 read back correctly at the same address, RAS/CAS for that bank had to be working; the fault was isolated to a single physical RAM chip, not the memory controller.