Re: SDRAM failures on MPC5200B

[lokowich]

Roman Fietze wrote: Hallo Mark, On Tuesday 13 November 2007 14:59:56 lokowich wrote: A follow-up: The first version of our termination network was based on the Lite5200 reference design, but was confirmed by Freescale to be inadequate. We revised the hardware per their errata, terminating both ends of the data bus. This fixed the problem. Thanks for the helpful feedback. If I can remember it correctly, the is no errata on the Freescale documentation web pages specifying to terminate the SDRAM data lines on both ends. The only document I can remember that specifies something similar is AN3045, "A comparison of the MPC5200B (Mask Set M62C) with prior MPC5200 Versions" on page 2, in the chapter "1. Overview". Could you please give me some hint about what errata or other document we probably missed? This is message from Freescale to our HW engineer and was in reference to Figure 1 in AN3045 A design could avoid termination resistors if the trace delay is less than 1/6 of the rise fall/time, if the output impedance matches the trace impedance or if the switching waveform is not so important (like PCI synchronous signals). The first two conditions are not met for MPC5200B DDR signals with 5 cm traces. The MPC5200B memory controller's nominal output impedance is 22 Ohms (16..35 worst case range). The nominal rise/fall time is ~400ps. However, worst case r/f time is about 200ps. Thus, MPC5200B memory design must include signal terminations. The most critical signals are memory clocks and DQS. The ringing on these signals (at the threshold level) is not allowed. The recommended series resistor value is 20..30 Ohms (may depend on the load). Bidirectional signals (DQM) should have series termination at both ends of data bus. It is common recommendation for MPC5200B memory design as well as for MPC5200 Rev.A. Lite5200B board doesn't follow these recommendations. This was done before the importance of the correct termination resistors was known. A large number of the issues people are having in moving from the MPC5200 to the MPC5200B are caused by not using the correct termination resistors. The Lite5200B board should not be used as an example. Memory system design must ensure AC timing specs for MPC5200 and SDRAM. Memory signals must have valid levels within setup/hold time window. The ringing caused by the PCB trace reflections must be damped before the input setup window begins. The ringing reduces time margins provided for signals. Proper termination of the signal removes the ringing. Using proper series termination solves most issues caused by poor signal integrity. In order to find optimal termination, use IBIS models of MPC5200 and SDRAM to simulate signal waveforms Maybe I have to tell our HW people that it will be necessary to fix our design as well. Thank you Roman _______________________________________________ Linuxppc-embedded mailing list Linuxppc-embedded@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-embedded

_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded