MPC8313 PCI Bridge

[denisfournier]

I interfaced a DSP (TMS320C6424) with a Freescale CPU (MPC8313E) using the 32 bit PCI Interface. Assuming that both devices are PCI compliant.

In the PCI V2.2 specifications, a PCI bridge usually has relatively weak output driver and assume a trace impedance of 65 ohms with no termination. In this case, series resistors are not required. But apparently, neither of the devices are fully PCI compliant.

I did not use any serial resistors on the PCI signals as the PCI specification don't require any termination due to the week driver. The nets are about 4 inches long with impedance controlled to 55 ohms +/- 10%. The MPC8313 is configured as PCI host and the TMS320C6424 as PCI device.

There is actually no problem. Everything work perfectly fine. But I probed some signals on the PCI interface to check the waveforms and they are mostly very badly shaped: Down to -1.05V undershoot and up to +4.24V overshoot. The PCI driver is apparently too strong and the waveform certainly does not meet the absolute maximum rating of the receiver chips.

I think that +4.24V overshoot may be harmful for the internal PCI clamping diode and some precautions need to be considered. A 33 ohms series resistors may be required on both end of each net. To make the interface PCI compliant, I will then need to add 54 series resistors on each of the devices.

In the event that I don't have enough space on the PCB to add 108 series resistors to reshape the waveform to match the PCI specifications, can I assume that the clamping diode on the receiver will be sufficient to absorb the energy of the overshoot and undershoot without the risk of damaging the chip on a long term?

In other words, the serial damping resistors, are they mandatory on the MPC8313E PCI interface?

Thanks,

[alexander_yakov]

For MPC8313 device output impedance of PCI signals is 25 Ohms, this is specified in “MPC8313 Hardware Specifications”, Table 3.

http://cache.freescale.com/files/32bit/doc/data_sheet/MPC8313EEC.pdf

Looking to the PCI specification you referring, I do not see a strict requirement for output driver to be 65 ohms, so I do not think PCI specification is violated here.

For overvoltage and undervoltage specifications please refer to PCI specification, Figure 4-5 “Maximum AC Waveforms for 3.3V Signaling”. Our device is done in accordance with PCI specification, so should meet these values.

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

[denisfournier]

Thanks Alexander,

Knowing the output impedance of the buffers will be very helpful.

I am referring to the following document:

- PCI System Architecture (4th Edition): Page 26, "PCI Method: Reflected-Wave Switching"

The PCI bus is unterminated and uses wavefront reflection to advantage. A carefully selected, relatively

weak output driver is used to drive the signal line partially towards the desired logic state. The driver

only has to drive the signal line partially towards its final state, rather than completely

(as a strong incident-wave driver would).

- PCI System Architecture (4th Edition): Page 728, "Card Characteristic Impedance"

The characteristic impedance range of on-card signal traces must be 65 ohms ±10%.

This impedance only includes the impedance of the on-card trace.

- PCI Bus Demystified Second Edition: Page 148,

The PCI specification has no requirement for the impedance of an unloaded motherboard.

However, the tighter electrical requirements of Compact PCI require that an

unloaded backplane have an impedance of 65 ohms ±10%.

Regards,

[alexander_yakov · Accepted solution]

As far as I can understand, the first section explains the fact that PCI is based on reflected wave, other two sections specify 65 ohm impedance for on-card traces and Compact PCI backplane.

[denisfournier]

This is correct.

The reflected wave switching of the PCI requires a week driver and no termination. The impedance of the bus is not specified.

The Compact PCI uses the same reflected wave switching technique on a 65 ohms backplane. (This suppose a 65 ohms driver but the driver impedance is still not specified.)

Thanks,

[alexander_yakov]

"Weak" does not mean 65 ohms. As you already mentioned, PCI bus is unterminated, so assuming any termination techniques like impedance matching is incorrect.

In other words, it is not correct to assume output impedance of the output driver should be equal to line impedance. As I can understand, this is a root cause of the topic.

[denisfournier]

You are correct,

This is not a requirement.

This is indeed a mathematical conclusion based on the information in the PCI specifications.