Thanks to all the people who offered suggestions.  In the end I
never did figure out the cause of the problem - it proved to be more
expedient to just use a fast cache SRAM (access time around 35ns)
together with a slow latch (74HC373 - tpd around 35ns!) to get a 70ns
memory system.

7C199 35ns 32Kx8 SRAMs are available from Jameco, qty 1, for less
than $2 each.  That's the same or less than what a 62256 costs.  The
7C199 is a power hog compared to the 62256 but that's not an issue for
this application.

Anyway, thanks again,
Bob Armstrong

If you look at the outputs of the 373 on a high speed scope (with a very short scope
probe ground lead, please), I bet you see ringing.
If so, you need to use some form of line flattening. The easiest is probably series 100 
ohm resistors right at the outputs.  This only works if all your loads are close together
on the far end of the line.

Paul Probert

bob@jfcl.com wrote:

> Hi,
> 
> I'm building an embedded controller that uses the Dallas 89C420 ultra
> high speed micro with an external 32K SRAM.  There's no external
> program memory - I'm using the internal flash for that.
> 
> Right now it works just fine at 15Mhz with a 74HC373 address latch
> and a 70ns SRAM, however I want to run in clock doubled mode at 30Mhz.
> For this I figure I'll need to use a 74F373 latch instead of the HC
> part.
> 
> The trouble is that it doesn't work, even at 15Mhz 1X (i.e. not
> doubled) with the 74F part.  Running a little memory diagnostic that I
> wrote, I get random SRAM errors that _don't_ occur with the 74HC
> address latch.
> 
> Are there any circuit changes that need to be made when using the
> 74F373 instead of the HC part?  Am I going to need series damping
> resistors or pull up resistors on the F373s outputs?
> 
> There are lots of Dallas appnotes for the 320, 550 and 420 family
> that talk about replacing the 74HC latch with a 74F part, but I
> couldn't find one actual schematic where they did it!
> Thanks much,
> Bob Armstrong
> 

bob@jfcl.com wrote:
>>I would suspect  ground bounce as a potential problem.
> 
> 
> Can't see any ground bounce on the scope, but that doesn't prove that
> it isn't there :-)  I did try tacking a 1uF tantalum directly across
> the F373's power pins and that didn't help.

It wouldn't - the minimum impedance will have been at around 1Mhz or 
lower. You need to bypass the 1uF with a 10n ceramic - minimum impedance 
~100Mhz.

Steve

> I would suspect  ground bounce as a potential problem.

Can't see any ground bounce on the scope, but that doesn't prove that
it isn't there :-)  I did try tacking a 1uF tantalum directly across
the F373's power pins and that didn't help.

To answer some other questions people have posed - it is a PC board
(not wire wrap) although only a two layer one so there is some
inductance in the power supply leads depending on how good the routing
is.

The SRAM chip in question is a 62256 (more exactly, a SEC
KM62256CLP-7L).

It works fine with the HC373 latch, but not with a F373 or ACT373
part.  I made no other changes except swapping the latch - the clock
speed is still 15Mhz.  I would have liked to try an AC373, but I don't
have one handy.

Thanks again,
Bob

In article <1105859858.842573.117870@f14g2000cwb.googlegroups.com>, 
bob@jfcl.com says...
> Hi,
> 
> I'm building an embedded controller that uses the Dallas 89C420 ultra
> high speed micro with an external 32K SRAM.  There's no external
> program memory - I'm using the internal flash for that.
> 
> Right now it works just fine at 15Mhz with a 74HC373 address latch
> and a 70ns SRAM, however I want to run in clock doubled mode at 30Mhz.
> For this I figure I'll need to use a 74F373 latch instead of the HC
> part.
> 
> The trouble is that it doesn't work, even at 15Mhz 1X (i.e. not
> doubled) with the 74F part.  Running a little memory diagnostic that I
> wrote, I get random SRAM errors that _don't_ occur with the 74HC
> address latch.
> 
> Are there any circuit changes that need to be made when using the
> 74F373 instead of the HC part?  Am I going to need series damping
> resistors or pull up resistors on the F373s outputs?
> 

I would suspect  ground bounce as a potential problem.   I haven't
experienced it personally, but have had reports of such problems
from another developer who was using similar latches on an IDE
interface.  In his case I think the solution involved some changes
in PCB layout.

> There are lots of Dallas appnotes for the 320, 550 and 420 family
> that talk about replacing the 74HC latch with a 74F part, but I
> couldn't find one actual schematic where they did it!
> Thanks much,
> Bob Armstrong
> 
> 
Mark Borgerson

bob@jfcl.com wrote:
> Thanks for the quick answers, guys.
> 
> 
>>The F373 does have TTL levels, whilst the HC373 has CMOS thresholds ?
> 
> 
> Absolutely.  Correct me if I'm wrong, but this shouldn't be a concern
> for CMOS outputs driving a TTL input (i.e. the 89C420 driving the
> 74F373) because the CMOS voltage swings are much wider than TTL
> requires.  

It matters for noise margin.

> It _is_ an issue for TTL outputs driving CMOS inputs (i.e.
> the 74F373 driving the 62256 SRAM) because the TTL logic 1 level is
> pretty low by CMOS standards.  BTW, Eric - that's the logic for adding
> pullups to the F373 outputs - to drag the TTL high levels up a little
> higher.  I tried adding 4.7K pullups to the outputs of F373, and it did
> improve the output levels but it didn't fix the problem.
> 
> 
>>You could try the newer Dallas 89C430/440/450
> 
> 
> Sorry - I must have missed something - how would that help?  I'd
> still need a faster address latch, right?

They may have different noise characteristics. Newer devices and all...

Most 80C51 variants have a cut/paste warning from older devices, that
ALE may have noise during multiple address bit transistions.
I have never seen this in a real design, but on loaded devices it
could happen.

You could try a series resistor PAD in the ALE to the F373, to slow the
edge slightly, and shift the threshold - but I would avoid the F373....

-jg

<bob@jfcl.com> wrote in message

> I'm building an embedded controller that uses the Dallas 89C420 ultra
> high speed micro with an external 32K SRAM ... 74F373 latch ...
> The trouble is that it doesn't work, even at 15Mhz 1X (i.e. not
> doubled) with the 74F part.  Running a little memory diagnostic that I
> wrote, I get random SRAM errors that _don't_ occur with the 74HC
> address latch.

What do you see with a 'scope: does the part meet setup/hold
at the specified voltage levels?

Early 74F parts were notoriously noisy: you may need .01 || 0.1
uF decoupling caps just for that IC.

You don't mention what ram you are trying to drive with the F373.

-- 
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer:  Electronics; Informatics; Photonics.
Remove spaces etc. to reply: n o lindan at net com dot com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/

Thanks for the quick answers, guys.

>The F373 does have TTL levels, whilst the HC373 has CMOS thresholds ?

Absolutely.  Correct me if I'm wrong, but this shouldn't be a concern
for CMOS outputs driving a TTL input (i.e. the 89C420 driving the
74F373) because the CMOS voltage swings are much wider than TTL
requires.  It _is_ an issue for TTL outputs driving CMOS inputs (i.e.
the 74F373 driving the 62256 SRAM) because the TTL logic 1 level is
pretty low by CMOS standards.  BTW, Eric - that's the logic for adding
pullups to the F373 outputs - to drag the TTL high levels up a little
higher.  I tried adding 4.7K pullups to the outputs of F373, and it did
improve the output levels but it didn't fix the problem.

>You could try the newer Dallas 89C430/440/450

Sorry - I must have missed something - how would that help?  I'd
still need a faster address latch, right?

Thanks again,
Bob Armstrong

Thanks for the quick answers, guys.

>The F373 does have TTL levels, whilst the HC373 has CMOS thresholds ?

Absolutely.  Correct me if I'm wrong, but this shouldn't be a concern
for CMOS outputs driving a TTL input (i.e. the 89C420 driving the
74F373) because the CMOS voltage swings are much wider than TTL
requires.  It _is_ an issue for TTL outputs driving CMOS inputs (i.e.
the 74F373 driving the 62256 SRAM) because the TTL logic 1 level is
pretty low by CMOS standards.  BTW, Eric - that's the logic for adding
pullups to the F373 outputs - to drag the TTL high levels up a little
higher.  I tried adding 4.7K pullups to the outputs of F373, and it did
improve the output levels but it didn't fix the problem.

>You could try the newer Dallas 89C430/440/450

Sorry - I must have missed something - how would that help?  I'd
still need a faster address latch, right?

Thanks again,
Bob Armstrong

bob@jfcl.com wrote:

> Hi,
> 
> I'm building an embedded controller that uses the Dallas 89C420 ultra
> high speed micro with an external 32K SRAM.  There's no external
> program memory - I'm using the internal flash for that.
> 
> Right now it works just fine at 15Mhz with a 74HC373 address latch
> and a 70ns SRAM, however I want to run in clock doubled mode at 30Mhz.
> For this I figure I'll need to use a 74F373 latch instead of the HC
> part.
> 
> The trouble is that it doesn't work, even at 15Mhz 1X (i.e. not
> doubled) with the 74F part.  Running a little memory diagnostic that I
> wrote, I get random SRAM errors that _don't_ occur with the 74HC
> address latch.
> 
> Are there any circuit changes that need to be made when using the
> 74F373 instead of the HC part?  Am I going to need series damping
> resistors or pull up resistors on the F373s outputs?
> 
> There are lots of Dallas appnotes for the 320, 550 and 420 family
> that talk about replacing the 74HC latch with a 74F part, but I
> couldn't find one actual schematic where they did it!

The F373 does have TTL levels, whilst the HC373 has CMOS thresholds ?

You could try the newer Dallas 89C430/440/450, and also try a
CMOS version of the F373 (as that is really a dinosaur), like the
P74LVC573 from Philips (etc) ?
-jg