Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
13 years ago

SPI Core: Strange problem with sample and shift clocks

Hi, All,

I've an SPI Slave IP in my design, and I'm sending and receiving a single

byte. Every 5-6 time, the bytes sent and received are wrong.

I've looked inside the SPI core with SignalTap II, and I see that when the

problem occurs, the Sample Clock or the Shirt clock (or sometimes both)

don't work. When all is fine, they are in sync with the SPI sclk, but

every so often some clock edges are missing: instead of 8 clock edges per bytes, I get 2-3 edges.

All this time, the sclk itself is received ok.

Some info: system clock is 25MHz, SPI sclk is 50KHz.

Any ideas?

Thanks,

Ran

[UPDATE] I found the problem, I think. It turns out it was a synchronization problem: the SPI Master has a different system clock then the SPI Slave. I thought that the cores had already a synchronization mechanism built-in, but they don't....so I got meta-stability issues.

The solution: add a two flip-flops on the slave SCLK path (And maybe also on the SS_N path, just to be on the safe side), to kill the metastability.

Cheers,

Ran

2 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    13 years ago

    --- Quote Start ---

    I've an SPI Slave IP in my design

    --- Quote End ---

    Which IP is it?

    I expect a serious timing violation causing the problem.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    13 years ago

    Hi, FvM, thanks for the reply :-)

    I'm using the serial peripheral interface - master / slave (http://www.cast-inc.com/ip-cores/interfaces/spi_ms/cast_spi_ms-a.pdf) IP core

    that comes with Qsys. I have one board with a Master instance of

    that core, communicating with another board with a Slave Instance.

    I'll try checking the clock constraints, but I'm not very hopeful: I have

    a NiosII with a TSE core in the same system, and they are working fine. Why should the SPI (which is rather slow) cause timing issues?

    I also tried inserting a 1 us delay between the SS_N and SCLK in the SPI master, but that didn't work either.

    Thanks,

    Ran