Electronics > Microcontrollers

Is a revision A of Silicon reliable, if there is a B?

<< < (3/4) > >>

Siwastaja:

--- Quote from: westfw on June 18, 2021, 01:31:47 am ---Note that SOME VENDORS (cough Atmel) have been tacking on "B" suffixes where there used to be an "A", for essentially new parts with dramatically different functionality.  :-((ATmega88 - ok.  ATmega88P: improved low power features (supposedly) ATmega88A, ATmega88PA: die shrink/revision?, ATmega88PB: oh, we changed some of the power pins to IO pins, added some features to the UART, connected some former ADC-only pins to digital GPIO...  And the 328PB - that gets additional UART, SPI, and I2C ports, two more timers, and a touch controller.  Grr.)

--- End quote ---

I think you are mixing up part number suffixes and silicon revision codes.

ATMega328PB is a different part with a different part number than ATMega328P or ATMega328. Even in ideal world, they are not supposed to be compatible! Normal level of carefulness is enough when ordering. Every letter, number and symbol in part number matters. Always. Copy-paste the full part number. Do not assume anything else. Easy enough!

This thread discusses revision codes which are not visible in the part number and usually are not something you can order at least through normal distributors like Digikey. Here non-compatibility is a huge issue, and some manufacturers play dirtier tricks than others. In ideal world, the parts should be 100% compatible (or the manufacturer should change the part number instead, like ATMega328P->ATMega328PB) but in reality you need to be careful, for example STM32H743 had this change of ADC clock divisor messing up both binary and source level compatibility going to REV V. In some very rare/unlucky cases, it couldn't be worked around in software at all, rendering parts inoperable. But you can't order the old or new revision anywhere, you need to order the part number which stays the same, see what product you happen to get, then design the product around this. So shit like this sometimes happens.

errorprone:
It also depends on the manufacturer.  For the i.mxrt series the silicon revision is also tacked on to the part number.  Unfortunately nothing is consistent and you just have to read the datasheets and errata.

westfw:

--- Quote ---I think you are mixing up part number suffixes and silicon revision codes.

This thread discusses revision codes which are not visible in the part number
--- End quote ---
My point is that they ARE mixed up; some vendors will put them in the part number, and some won't.

JPortici:

--- Quote from: westfw on June 20, 2021, 10:54:01 pm ---
--- Quote ---I think you are mixing up part number suffixes and silicon revision codes.

This thread discusses revision codes which are not visible in the part number
--- End quote ---
My point is that they ARE mixed up; some vendors will put them in the part number, and some won't.


--- End quote ---

like TI and their LDOs that apparently have logic so complex they need different silicon revisions because safe modes are nonfunctional, and they will happily use an "A" for both a different version and a different pinout.. TPS709A is first revision, pinout A. TPS709AB is second revision, pinout B. TPS709BA is third revision, pinout A. With also a multitude of other letters after to complete the shitshow, the wrong revision/pinout can slip

(I don't remember if it was actually with the TPS709 that i had these issues, but you get the point)

Siwastaja:

--- Quote from: JPortici on June 22, 2021, 03:45:41 pm ---like TI and their LDOs that apparently have logic so complex they need different silicon revisions because safe modes are nonfunctional, and they will happily use an "A" for both a different version and a different pinout.. TPS709A is first revision, pinout A. TPS709AB is second revision, pinout B. TPS709BA is third revision, pinout A. With also a multitude of other letters after to complete the shitshow, the wrong revision/pinout can slip

--- End quote ---

Oh that sucks but it's totally manageable, all you need to do is to stupidly treat the part number as an identifier for a part, each unique part number is an unique part, bypassing all assumptions and all generic wisdom about similar numbers. The longer the part number is the easier it gets since it overwhelms your brain and makes you not look at it, only copy-paste. When ordering though be very careful you are actually getting what you copypasted in the search box.


But ST doesn't include the silicon revision in the part number, and silicon revisions are not necessarily binary or even source level compatible, and no, this isn't due to just bugfixes but other changes too, like arbitrarily adding a fixed, unnecessary and un-bypassable /2 clock divider into the ADC. There simply is no process to overcome this as far as I know, unless the process is:
* Every week or so, download the full st.com and all documentation, store in your archive
* Very early in design, order parts for the whole expected batch of the product
* Open the physical package
* Look at the device markings
* Now from your archive, find the documentation for that revision, since it's not publicly available anymore even though distributors sell you those parts
* Now design!
* Manufacture all units at once

... although I'm sure if you were a large enough player like a big automotive customer ST would help you in this process, I'm pretty sure.

Navigation

[0] Message Index

[#] Next page

[*] Previous page

There was an error while thanking
Thanking...
Go to full version