3478A FRAM modification

[Miti]

Nicely done mod.  I have a simple change for new builds that will increase endurance 8x.  Might be overkill, but the meter does constantly read the SRAM.

Because the FRAM is organized into 8 byte rows, a read of any byte causes a destructive read/write of all 8 bytes in the row.  Reading 8 consecutive bytes causes 8 destructive read/writes of the same row.

Putting only 1 value on each row eliminates this issue.  Simply use address lines A[10:3] and connect A[2:0] to gnd.

Very interesting and valid point. But… See a screenshot from the FM16W08 data sheet.



Cheers,
Miti

[Dan N]

Building on the good work of wolfalex and miti, I designed my own pcb for a couple of reasons:  I want it as small as possible and I want the overhang on the left side keeping the right side free for any future EPROM replacement/upgrade adapter.  I also included my row mod (2 posts above) to put only 1 value on each row.

First day installed and testing:  read/write cal data worked, with 5V rail at 5.069V, SRAM VCC was 4.796V, FRAM VCC is 4.833V so drawing less current through the diode.

For the glue logic I used multi-function logic (74LVC1G57) to get the exact functions I wanted and squeeze everything between the DIP pins.  May have slightly less logic-H margin than an "official" TTL compatible input, but works for now and will see if further testing shows any issues.

As wolfalex mentions on his website, I installed a resistor in place of the battery (see pic) to keep a comparator input from floating.

[Miti]

Thanks Dan, looks good!

Only one concern. On the Reset line, you basically short circuit a 1uF capacitor, potentially charged to 5V. MCP120 data sheet doesn’t seem to specify a maximum output current but it could give a significant spike.
That’s the reason I used a transistor that can take few hundreds mA. Still not ideal, I should have used a current limiting resistor series.

Cheers,
Miti

[Dan N]

I was also concerned about the current sink limit.  Top priority is the ability to pull the reset down fast enough to prevent spurious writes on power-down, and even that is only an issue if the cal switch is enabled.

I looked at the current sink chart for the LM339 in the original reset circuit and it maxes out at 16mA before dropping out of saturation and then the voltage rapidly climbs past 1V.  The MCP120 chart shows it's below 600mV at 16mA so the MCP120 should pull reset down faster than the LM339.

I'm less certain of this calculation but peak current shouldn't be an issue.  From the same MCP120 output chart the equivalent resistance is 30-40ohm (increases at higher current).  Discharge time should be <150uS.

EDIT:  I scoped CE2 (reset) and it drops to 2v @ 55uS, 1V @ 80uS, 0V @ 146uS.  My adapter is soldered in so I have no way to compare with the old LM339 reset.

[Dan N]

All versions of the FRAM mod add a new reset circuit onto the FRAM VCC supply which comes through a diode from the regulated +5V rail while the original was directly on the unregulated 5V (really 13V).  I assume the regulated rail has more bulk capacitance and drops slower than the unregulated rail, and the diode can keep the FRAM VCC higher than the cpu +5V rail, so the original reset may trigger sooner at power-down.

I wanted to confirm that the new reset circuit shuts down the cpu before its supply falls below 4.5V.  The attached scope screenshot shows that it does.  Yellow is CE2 (reset), purple is ALE (cpu activity), blue is the cpu +5V rail.  The FRAM VCC drops enough for the MCP120 supervisor to trigger reset while the cpu +5v rail is at 4.5V and 89uS (35 instructions) later the cpu halts.  The +5V rail doesn't drop any measurable amount on this trace.

So all versions of this mod should be good.