Problems with certain CF disk in Rev1 Tiny68K

Certain brands of CF disks may not work on some Tiny68K rev 1 boards. The simplest way of testing that is to copy files from one CF drives to another with verify switch turned on, e.g.,

PIP C:=A:*.*[v]

If the operation is aborted prematurely during the copying operation, then the Tiny68K board is having problem with that particular CF disk.

Below is the description of the problems and solutions. This writeup was originally posted here: https://www.retrobrewcomputers.org/forum/index.php?t=msg&th=222&goto=4050&#msg_4050


After tried about 10 different brands of CF cards including one that computerdoc kindly loaned to me, I believe the problem is due to ground bounce of fast CF driving a 16-bit high-capacitance data bus. This is a CF read problem and is data dependent such that reading data with all 1's (0xFFFF) can cause a glitch on the CF read control line. Because CF data register is a FIFO, a glitch on the read line will flush the current data and output the next data. A faster CF is more sensitive to glitch. This is why data with value of 0xFFFF dropped out of the sector data stream. This is not a common problem, of the 10 or so brands of CF I tried, only 3-4 brand had this problem and one Tiny68K board is particularly susceptible to this problem. The newer, faster, higher density CF is more likely to have this problem than the older, slower, and lower density ones.

A number of factors contributed to this issue: 2-layer pcb without power/ground layers, direct connect of CF to 68000 data bus, and 5V system. I don't want to change the 5V supply or go to the costly multi-layer pcb, so the solution was focus on the interface between CF and CPU and test the fixes on the one Tiny68K that's particularly susceptible. There are 3 separate fixes, each incrementally reduces the magnitude of the ground bounce. They are presented from the most effective to the least effective.
1. Pull up half of the data bus with a SIP 10K resistor pack. With the pull-up, the data bus is precharged to half low and half high so a data pattern of 0xFFFF only drive 8 data lines from low to high instead of 16. The magnitude of ground bounce is reduced by half. This is also an easy fix, just solder a SIP resistor pack to the 68000DIP and tie the common pin to a 5V point nearby. (see 2nd picture)
2. Terminate the CF read strobe (nIORD) with 100 ohm series resistor and a 100pF capacitor. The added network has a RC time constant of 10nS that should suppress fast glitch without unduly slowing down the read access.
3. Isolate the CF data bus with 100 ohm resistors. This effectively slow down the CF bus driver thus reduce the magnitude of the ground bounce.
www.retrobrewcomputers.org_lib_plugins_ckgedit_fckeditor_userfiles_image_boards_sbc_tiny68k_tiny68k_rev1_fixes_comp_copy.jpg

www.retrobrewcomputers.org_lib_plugins_ckgedit_fckeditor_userfiles_image_boards_sbc_tiny68k_tiny68k_rev1_fixes_solder_copy.jpg
Having said all that, your Tiny68K (rev 1 pcb) may not have any issue with the CF. If you do, try an older, lower density CF card and solder a 10K SIP resistor pack to the back of the board should fix it. If the problem is still there, please let me know. We can arrange an exchange.

The 3 fixes are implemented in rev 2 of Tiny68K. (see third picture) I've built and tested 8 of them and they are completely trouble-free so far (knock on wood).

www.retrobrewcomputers.org_lib_plugins_ckgedit_fckeditor_userfiles_image_boards_sbc_tiny68k_tiny68k_rev2_cf_fixes_copy.jpg

boards/sbc/tiny68k/rev1_cf_problem.txt · Last modified: 2018/01/09 15:42 by plasmo
Driven by DokuWiki Recent changes RSS feed Valid CSS Valid XHTML 1.0