[N8VEM: 20241] KISS-68030 timing / F195/F257 vs. -A suffix parts

[John Coffman]

All,

In the hopes of reproducing the memory failures with TEST3.BIN I have
finally received 74F257 chips to replace my 74F257A chips. Both of the
boards I built up use vanilla 74F195 chips.

Tests done on my first board: 25Mhz CPU osc., 64Mhz DRAM osc. 2 dual
sided 32Mb off-brand 60ns memory SIMMs.

Original memory chip set: Signetics 74F195N, 3 x Motorola MC74F257A
-- no errors

Last night's test: same Signetics 74F195N, 3 x Fairchild 74F257PC --
ran TEST3 all night with no errors (330 passes)

I had hoped to show that by reducing the DRAM oscillator frequency to
50Mhz, the memory errors would cease. I am unable to introduce memory
errors by going to vanilla 257 chips, which are slower than the 257A chips.

The 22V10 GALs use in the DRAM memory access circuitry are USED Lattice
-7 ns parts, reprogrammed.

Would others kindly contribute their comments on how the memory error
problems were solved. My own way was more electrolytic decoupling, and
a week-long burn-in.

--John

[William R Sowerbutts]

I am using 32MHz CPU oscillator, 64MHz DRAM oscillator, 3 x 74AS257 as the
DRAM muxes.

As high levels of decoupling were recommended I am using:
- 47uF low ESR electrolytic for C501, C930 (Panasonic ECE-A1AKG470)
- 470uF low ESR electrolytic for C901, C928, C929 (Nichicon RNE1C471MDN1PX)
- 47uF tantalum for C926, C950, C904 (AVX TAP476M020CCS)
- 0.47uF MLCC everywhere else (TDK FK28X7R1E474K)

I am also using used Lattice 22V10 7ns GALs purchased on Ebay from the vendor
you suggested.

The board had memory errors in every pass; nearly 5 days of burn-in made no
difference as far as I could observe.

My memory errors were solved by:
- Replacing a faulty GAL with a different part (from the same batch)
- Replacing 74F195 at U403 with 74F195A

Not sure what was wrong with the GAL. They were second hand and having
ordered 10 I received 12 which, in retrospect, does rather suggest that the
vendor expected one or two to be dodgy!

If I were building this board again I would buy new Atmel 22V10 GALs instead
of second-hand chips. I would also use lower value decoupling caps -- I don't
really know a great deal about this but I believe that lower values perform
better at higher frequencies? I would use MLCCs in place of the tantalums if
possible as I believe tantalum is a conflict resource.

We've just moved house so my life is in boxes. I hope to unpack the '030
board and make some more progress with debugging the Linux port in the next
few weeks. It's tantalisingly close to usable.

Best wishes

Will

Post by John Coffman
All,
In the hopes of reproducing the memory failures with TEST3.BIN I have
finally received 74F257 chips to replace my 74F257A chips. Both of the
boards I built up use vanilla 74F195 chips.
Tests done on my first board: 25Mhz CPU osc., 64Mhz DRAM osc. 2 dual
sided 32Mb off-brand 60ns memory SIMMs.
Original memory chip set: Signetics 74F195N, 3 x Motorola MC74F257A
-- no errors
Last night's test: same Signetics 74F195N, 3 x Fairchild 74F257PC --
ran TEST3 all night with no errors (330 passes)
I had hoped to show that by reducing the DRAM oscillator frequency to
50Mhz, the memory errors would cease. I am unable to introduce memory
errors by going to vanilla 257 chips, which are slower than the 257A chips.
The 22V10 GALs use in the DRAM memory access circuitry are USED Lattice
-7 ns parts, reprogrammed.
Would others kindly contribute their comments on how the memory error
problems were solved. My own way was more electrolytic decoupling, and
a week-long burn-in.
--John

_________________________________________________________________________
William R Sowerbutts ***@sowerbutts.com
"Carpe post meridiem" http://sowerbutts.com
main(){char*s=">#=0> ^#X@#@^7=",c=0,m;for(;c<15;c++)for
(m=-1;m<7;putchar(m++/6&c%3/2?10:s[c]-31&1<<m?42:32));}

[John Coffman]

Will,

I think you are suggesting that all GALs are not created equal. [no pun
intended]

On the decoupling issue, as far as I can tell, the 0.1uF caps work just
as well as the 0.47uF caps. However, the board needs larger
electrolytic caps than originally called for (22uF). (2 x 100uF + 2 x
47uF) are on the current board.

Wayne obtained new electrolytic caps as I recall, and saw no burn-in
time. I used electrolytic caps from a mass kit purchase of several
years back, and experienced a burn-in phenomenon on both boards.

If anyone is experiencing memory errors from TEST3.bin or TEST4.bin, and
would like to try the upgrade from F257 chips to F257A, I have many of
the -A parts. I would especially like to swap for F257 parts from a
board that is having problems.

I want to nail down this memory error issue. I can't do anything until
I can reproduce the error condition.

--John

Post by William R Sowerbutts
I am using 32MHz CPU oscillator, 64MHz DRAM oscillator, 3 x 74AS257 as the
DRAM muxes.
- 47uF low ESR electrolytic for C501, C930 (Panasonic ECE-A1AKG470)
- 470uF low ESR electrolytic for C901, C928, C929 (Nichicon RNE1C471MDN1PX)
- 47uF tantalum for C926, C950, C904 (AVX TAP476M020CCS)
- 0.47uF MLCC everywhere else (TDK FK28X7R1E474K)
I am also using used Lattice 22V10 7ns GALs purchased on Ebay from the vendor
you suggested.
The board had memory errors in every pass; nearly 5 days of burn-in made no
difference as far as I could observe.
- Replacing a faulty GAL with a different part (from the same batch)
- Replacing 74F195 at U403 with 74F195A
Not sure what was wrong with the GAL. They were second hand and having
ordered 10 I received 12 which, in retrospect, does rather suggest that the
vendor expected one or two to be dodgy!
If I were building this board again I would buy new Atmel 22V10 GALs instead
of second-hand chips. I would also use lower value decoupling caps -- I don't
really know a great deal about this but I believe that lower values perform
better at higher frequencies? I would use MLCCs in place of the tantalums if
possible as I believe tantalum is a conflict resource.
We've just moved house so my life is in boxes. I hope to unpack the '030
board and make some more progress with debugging the Linux port in the next
few weeks. It's tantalisingly close to usable.
Best wishes
Will

Post by John Coffman
All,
In the hopes of reproducing the memory failures with TEST3.BIN I have
finally received 74F257 chips to replace my 74F257A chips. Both of the
boards I built up use vanilla 74F195 chips.
Tests done on my first board: 25Mhz CPU osc., 64Mhz DRAM osc. 2 dual
sided 32Mb off-brand 60ns memory SIMMs.
Original memory chip set: Signetics 74F195N, 3 x Motorola MC74F257A
-- no errors
Last night's test: same Signetics 74F195N, 3 x Fairchild 74F257PC --
ran TEST3 all night with no errors (330 passes)
I had hoped to show that by reducing the DRAM oscillator frequency to
50Mhz, the memory errors would cease. I am unable to introduce memory
errors by going to vanilla 257 chips, which are slower than the 257A chips.
The 22V10 GALs use in the DRAM memory access circuitry are USED Lattice
-7 ns parts, reprogrammed.
Would others kindly contribute their comments on how the memory error
problems were solved. My own way was more electrolytic decoupling, and
a week-long burn-in.
--John

_________________________________________________________________________
"Carpe post meridiem" http://sowerbutts.com
(m=-1;m<7;putchar(m++/6&c%3/2?10:s[c]-31&1<<m?42:32));}

[Wayne Warthen]

As you mentioned John, I never saw any memory issues, so my input is
probably of limited value.

For what it is worth here are the salient aspects of my build:

- Decoupling caps are 0.47uF ceramic (new), except:
- C926, C950 caps are 47uF tantalum (new)
- C904 cap is 10uF tantalum (new)
- Power smoothing caps are 470uF electrolyitic (new)
- TI SN74AS257N (3, new)
- Phillips 74F195AN (2, NOS)
- Atmel ATF22V10C-7PX (2, new)
- Atmel ATF16V8BQL-15PU (1, new)
- Motorola MC68030RC-40 (used)

Currently running with no errors w/ CPU @ 32MHz and refresh @ 64MHz

--Wayne

[John Coffman]

Post by Wayne Warthen
As you mentioned John, I never saw any memory issues, so my input is
probably of limited value.
o C926, C950 caps are 47uF tantalum (new)
o C904 cap is 10uF tantalum (new)
* Power smoothing caps are 470uF electrolyitic (new)
* TI SN74AS257N (3, new)
* Phillips 74F195AN (2, NOS)
* AtmelĀ ATF22V10C-7PX (2, new)
* Atmel ATF16V8BQL-15PU (1, new)
* Motorola MC68030RC-40 (used)
--Wayne

I thought I remembered that you had no memory error issues. AFA I can
tell, electrolytic cap decoupling was the major issue for me. My old
caps had to form up by being powered for a couple of days. Both boards
had this issue, and both are perfectly error free now.

--John

[John Coffman]

You mean DRAM [RAS/CAS] timing at 64Mhz. The refresh is fixed at 15.625
usec. (64000Hz).

--John

[Wayne Warthen]

Sorry, yes, RAS/CAS timing is 64MHz.

--Wayne

Post by John Coffman
You mean DRAM [RAS/CAS] timing at 64Mhz. The refresh is fixed at 15.625
usec. (64000Hz).
--John