[Milkymist-devel] The dungeons of NORia: CE0/FLASH_CE_N rework

[Werner Almesberger]

Executive summary:
- adding a pull-up to CE0 has no significant effect on NOR corruption
- next: reset chip rework


The details
-----------

This time, I also pulled up the CE0 (chip enable zero) line:

http://downloads.qi-hardware.com/people/werner/m1/nor/d7/nor-ce0.jpg

This is again a 4.7 kOhm resistor to 3.3 V. Soldered "wirelessly"
directly to neighbouring pads.

As expected, the results look familiar:

http://downloads.qi-hardware.com/people/werner/m1/nor/d7/out.png

NOR corruption still occurs and happens quite often. The observed
frequency is a bit lower than before, but I think this is caused
by having only a small number of samples. Had I given it a few
more days, the results would probably look exactly like in the
previous runs.

Once again, I ran into an uncommanded locking of the first block,
this time after, estimated, 6300-8500 cycles.

The corruption pattern also looks familiar:

00000 ____________________ | 00000000 00010000 | d7/1376-corrupt.bin
                           | 00000000 00000000 | d7/2892-corrupt.bin
                           | 00000000 00010000 | d7/5633-corrupt.bin
0000c ________________11__ | 00000100 01000000 | d7/6295-corrupt.bin
0003c ______________1111__ | _____0__ ________ | d7/6062-corrupt.bin 1/2
00082 ____________1_____1_ | _0__00__ 0_____00 | d7/1376-corrupt.bin 1/1
                           | _0__00__ 1_____00 | d7/6062-corrupt.bin 2/2
0008c ____________1___11__ | 00_00___ _0___000 | d7/5633-corrupt.bin 1/1
00410 _________1_____1____ | ________ _100____ | d7/1635-corrupt.bin 1/1
00480 _________1__1_______ | ________ 0_0_0___ | d7/329-corrupt.bin 1/1
007fe _________1111111111_ | ________ __0_____ | d7/3803-corrupt.bin 1/1
00800 ________1___________ | ________ ____0___ | d7/3473-corrupt.bin 1/1
00850 ________1____1_1____ | ________ 0_0_0___ | d7/3634-corrupt.bin 1/1
00fb0 ________11111_11____ | ________ _001____ | d7/2892-corrupt.bin 1/1
01100 _______1___1________ | ________ 0_0_____ | d7/6295-corrupt.bin 1/1


Conclusion
----------

The two pull-ups (WE# and CE0) Xilinx recommend in their manual
don't have an effect on our NOR corruption, but they also don't
cause any ill side-effects.

In general, I think it still is a good idea to add the pull-ups
in a future board revision.


What's next
-----------

After checking those two "low-hanging fruit" reworks, the path is
now clear for the experiment that we hope will remove the root
cause of all this for good: connecting the reset chip to the 5 V
supply instead of the 3.3 V rail, and using a reset chip with a
higher threshold voltage.

I've already done the rework and a test is running now.

- Werner