If ST's RTC chip (m41st84w or M41ST85W) has a failure mode to provide invalid RTC data, just like 2022(Year)-18(Month)-49(Date)? If have, what's the failure rate?

[Tesla DeLorean]

Try explaining your situation better, it didn't translate well.

Y​ou have an issue? This issue?

Could you be mixing BCD and Binary?

Is there any report of this?

Is it described in an errata or some peer reviewed report?​

[Wwang.11]

Hi Delorean, Thanks for your response.

m41st84w is used in our product, in one time lab tests, software read invalid date from the RTC (we don't know the real invalid date, but we know this issue as it triggered some fault log), but this issue did not happen before, and now cannot be reproduced.

[Wwang.11]

In our current design, software will reboot if read invalid RTC data, as we suppose the RTC IC should not provide invalid time data, we considered 2 failure modes (RTC data loss, RTC power loss) for RTC function in FMEA, now we would like to confirm if the RTC IC has the failure mode to provide invalid RTC data.

[Tesla DeLorean]

You'd probably want a complete dump of the registers to process this further.

The RTC could get into states not consistent with your expectation whilst still internally valid and consistent from a design/functional perspective.

For example BCD 18 -> 0x12 and 49 -> 0x31 typically BCD reporting would be used where the values can by passed directly to 7-Segment displays, for example, without any decimal math.

You might need to work on the robustness of your methods, decoding states and flags you might currently be ignoring.

[Wwang.11]

The software has no issue to decode the RTC register value to decimal format.

To the RTC internal design, it should have been designed not to provide invalid time data, application note AN1572 figure 3 describes how the data is read from counter to I2C bus. But if some counter’s output is latched (hardware failure), 

for example, if the counter for 10date (10date of June-09th is 0) is latched to 0, the date of June-10th will be read at June-00.

If this failure is existed, but probability is very very low. or only existed in theory?

[Tesla DeLorean]

You'll probably need to discuss this with your local rep or FAE. Hypothesizing on failure modes from incomplete/inconclusive observations is not something I want to do.

The RTC has multiple time domains, clocking at different rates.

You might take longer to read data, or span a ripple-thru, such that half the data you're working on comes from an earlier instant in time than the latter. This is frequently why you read the values twice in a row, and insure both reads report the same data.