Hi,
My customer is struggling with difficulties to get stable performance on their design, which seems to be related to the DDR configuration. Can someone please help us to verify the configuration below:
Settings of DDR3 SDRAM CK=666.67MHz; tCK=1/CK=1.500ns; tRCD=tRP=CL=13.75ns; Speed -125 (1.25ns or 800MHz)
DDR memory is: TN-47-02 DDR2 from Micron (www.micron.com/.../tn4702.pdf)
- OLD SETTINGS: Two Versions in step (iv).
(i) CL=floor(13.75n/(1/666.67M))=9=0x9
(ii) CWL=7 (0x7) since 1.875ns>tCK>1.500ns
(iii) AL arbitrarily set via step (iv) and not equal to 0, CL-1 or CL-2.
(iv) DDR_PHY_CTRL_1[4:0]=[10=0x0A|14=0x0E] with 0x0A older and 0x0E being newer settings thus RL=CL+AL=[10+1=11=0xC|14+1=15=0xF]
- Latest NEW SETTINGS
(i) CL=ceiling(13.75n/(1/666.67M))=10=0xA
(ii) CWL=7 (0x7) since 1.875ns>tCK>1.500ns
(iii) Selected AL=CL-2=10-2=8=0x8 to meet DDR_PHY_CTRL_1[4:0] value constraint.
(iv) DDR_PHY_CTRL_1[4:0]=17=0x11 (Register value required is 18-1=17=0x11).This meets CL+1=10+1=11<=Register Value<=CL+7=10+7=17
We would like you to review the latest settings in (2) and also possibly comment on the settings in older versions in (1). We are experiencing AC power drops from large fixed FFT size computation from captured samples with instability of >=0.3/0.5dB, over multiple sample intervals in time (few hours to a couple of days), and moving from older settings to new settings we have improved the instability – but still needs further improvement. What is the implication of improper setting of CL and specifically AL (Posted CAS Additive Latency) with respect to data gap (excerpt from pp15-16 TN4702.pdf from Micron Tech. Inc.), and its impact on data read out from DDR3.
Best regards,
Joakim