MCP23017: same part number, two fewer inputs

[NiHaoMike]

https://hackaday.com/2023/02/03/mcp23017-went-through-shortage-hell-lost-two-inputs/
Looks like Microchip messed up a batch (or even an update to a newer process) and decided to change the datasheet rather than fix the problem. Hope it doesn't end up in some product where those missing inputs are important...

[djsb]

Maybe it's shrinkflation?

[ataradov]

More likely scenario is that some marginal behaviour was discovered and those pins can't fully meet the specification and instead of providing a full new set of characteristics, they were descoped.

It is likely that if it worked so far in your design, it will continue to work. The change is to prevent new designs from into running into potential issues.

In MCU world this happens occasionally and accompanied by a note in the errata document, but I'm not sure if there is even errata for basic ICs like this.

It would be interesting to see if those pins actually work just as they used to and the change is only to the datasheet.

[gamalot]

I'm guessing there is nothing has changed on the chip itself, it's just that those two pins were found to work incorrectly when used for input.

[newbrain]

More likely scenario is that some marginal behaviour was discovered and those pins can't fully meet the specification and instead of providing a full new set of characteristics, they were descoped.
[...]
It would be interesting to see if those pins actually work just as they used to and the change is only to the datasheet.

I think you are 100% correct; from the change notification page:

NOTE: Please be advised that this is a change to the document only the product has not been changed.

EtA: Oh $DEITY - why did I start reading the comments on HaD, it's the YouTube of tech sites.

[Ian.M]

It was a known silicon bug (circa 2014) that never made it into the errata!  Basically a rapidly toggling signal on bit 7 of either port intermittently causes the other bits of the port to read as '1' irrespective of their input level.

See https://www.microchip.com/forums/m831656.aspx
and
https://forums.raspberrypi.com/viewtopic.php?f=44&t=91209

Edit: I am the same Ian.M as reply #2 in the Microchip topic, and have been wary of this issue whenever I've used a MCP23017 ever since.

[thm_w]

The weird thing is in that 2014 Microchip thread, ticket was filed with support, and support confirmed the problem. You'd expect it to be added to errata within a reasonable time period.
Makes you wonder how many other reported issues exist and have never made it to the datasheet.

EtA: Oh $DEITY - why did I start reading the comments on HaD, it's the YouTube of tech sites.

If you can filter through most of the posts of trolls and noobs who have no idea what they are talking about, the remaining 20% are good. All of the info posted in this thread was already covered there.

[ataradov]

Makes you wonder how many other reported issues exist and have never made it to the datasheet.

A lot. In fact only a fraction of the issues makes it into the errata or the datasheet. There are many issues that only show up in marginal use cases, those will almost never get published. This is especially true for very old parts.

In this case to me it seems like response could be faster, but it is hard to tell what was going on. It is possible that it affected some big customer and they needed an official reason for a design change on their side or something like this.

[vk6zgo]

Back in the day, we had an "in house" built TV transmitter remote control unit (one of two) in for a modification for one of the programme inputs.

In the "guts" we used a 4000 series monostable IC (too long ago, so I don't remember the number), which was used to obtain a relatively long time delay.

For some reason, whilst working on the equipment, I removed this IC & carefully placed it to one side, where it promptly fell on the floor, where try as I might, I couldn't find it.

"Bummer!" said I, but consoling myself with the fact that we had plenty in stock, I grabbed one from the parts store & continued with the "mod".

When I finished & tested the unit, it didn't work correctly, as the mono always timed out early.
Assuming it was one "rogue" device, I grabbed some more, all of which had the same problem, while another mono of the same type from another part of the unit worked correctly.

After much digging through data books (No Internet back then), it turned out that people had been having problems with erratic operation when using short time delays, so the problem was fixed by optimising the device for such intervals, to the detriment of long ones.
The problem was that they kept the original number for these modified ones & allocated the original device a new number.

We bought a bunch of the new number, which worked as before, changed the remote control's documentation to reflect this, then when the second unit was taken out of service for the original "mod", changed the ICs in that one too.

OK for us, with two units---imagine if we had thousands of them in the field & in production!

[dmills]

I remember getting bitten by that one.
Daft thing is the 'long duration' variant got re released under a different part number.

[Smokey]

Holy crap!  I almost just got bit by this!  I was looking at an older version of the datasheet during the whole design.  I just happened to google the part number as I was setting resistor values on the BOM consolidation phase and saw the updated datasheet with that note. 

Wow.  This is so messed up!

https://microchip.my.site.com/s/article/GPA7---GPB7-Cannot-Be-Used-as-Inputs-In-MCP23017
"Answer:
In the datasheet Revision D (June 2022), GPA7 & GPB7 are mentioned as outputs only for MCP23017. But, the register bits do allow for the direction of these I/O pins to be changed to inputs. However, the SDA signal can be corrupted during a reading of these bits if the pin voltage changes during the transmission of the bit. It has also been reported that this SDA corruption can cause some bus hosts to malfunction. In conclusion, GPA7 & GPB7 should only be used as outputs to avoid future issues."

I totally came here to make a post about this, but of course it already exists....