Bad batch of CS8900's?

[Robert Reimiller]

A customer has a problem with the latest batch of CS8900 chips. It
seems that TCP packets with a segment size of greater than or equal to
1030 bytes are not being transmitted correclty by the chip. The chip
is changing the data to FFFF words, every other word, for about 8
times in a row, and then resumes sending the correct data.

We also tried doing Internal Loopback and External Loopback tests
using large (1460 byte) UDP packets and these chips fail this also.
Previous batches (with different date codes) work fine. The chips are
on a custom PC-104 board in conjunction with a commercial CPU board.
We've checked the bus timing and everything looks fine.

The chips are part number CS8900A-IQ and the batch code is KTAALD0239.
Anyone else see similar problems?

Bob

 

[Jamie]

sounds like bad run of internal memory.

 

[Robert Reimiller]

The chips are part number CS8900A-IQ and the batch code is KTAALD0239.

Just an update. The client had four boards that failed loopback
modified with a different datecode of CS8900 and the problem went
away. At this point it suggests bad chips.

 

[Hal Murray]

Just an update. The client had four boards that failed loopback

modified with a different datecode of CS8900 and the problem went
away. At this point it suggests bad chips.

It's could be, but I'm not sure that batch of chips is bad.

It could be something that is marginal with the design and that
particular batch of chips is within spec but enough slower/faster
than the other batches so that it doesn't work on that board.