Cyclone III and 2xSDRAM

[Altera_Forum]

I need to open topic because I couldn't find answers anywhere so maybe someone can help. 

Card is designed around EP3C40F484C6N with 2 SDRAM chips IS42S16320B-6TL but they are connected on opposite sides of FPGA. Goal was to create 32bit interface but without sharing signals(ADDR and such like normal 32bit interface should be). Idea was to be able to control each chip better by direct interface to the FPGA and now I see that this was bad idea. Running LFSR tests on both chips is possible up to 80MHz but after that all fails. First problem is that each SDRAM_CLK uses his own PLL so there might be problems with that. SDRAM1_CLK is connected to PLL_4 and SDRAM2_CLK to PLL3. To avoid Quartus using random PLL I did this in *.qsf:    --- Quote Start ---  set_location_assignment PLL_4 -to "pll:pll|altpll:altpll_component|pll_altpll:auto_generated|pll1" set_location_assignment PLL_3 -to "pll2:pll2|altpll:altpll_component|pll_altpll:auto_generated|pll1"  --- Quote End ---    This says that 2 PLLs are generated who needs to be in sync. As I know they should be in sync by default but maybe there is something I don't know. After that I have edited SDC file and added there what's needed to separate both chips.  INPUT SIGNALS   --- Quote Start ---   create_generated_clock -name sdram_clk -source [get_pins {pll|altpll_component|auto_generated|pll1|clk[2]}] [get_ports sdram_clk_o]; create_generated_clock -name sdram2_clk -source [get_pins {pll2|altpll_component|auto_generated|pll1|clk[0]}] [get_ports sdram2_clk_o]; set input_clock [get_clocks sdram_clk];# Name of input clock set input_clock2 [get_clocks sdram2_clk];# Name of input clock set tco_max 5.800; # Maximum clock to out delay (external device) set tco_min 3.200; # Minimum clock to out delay (external device) set trce_dly_max 0.240; # Maximum board trace delay set trce_dly_min 0.200; # Minimum board trace delay set input_ports [get_ports {sdram_dq_io[*]}];# List of input ports set input_ports2 [get_ports {sdram2_dq_io[*]}]; # Input Delay Constraint set_input_delay -clock $input_clock -max [expr $tco_max + $trce_dly_max - $trce_dly_min] $input_ports; set_input_delay -clock $input_clock -min [expr $tco_min + $trce_dly_min - $trce_dly_max] $input_ports;  set_input_delay -clock $input_clock -max [expr $tco_max + $trce_dly_max - $trce_dly_min] $input_ports2; set_input_delay -clock $input_clock -min [expr $tco_min + $trce_dly_min - $trce_dly_max] $input_ports2;  # set_multicycle_path -setup -from $input_clock -to [get_clocks {pll|altpll_component|auto_generated|pll1|clk[0]}] 2 set_multicycle_path -from [get_clocks {sdram_clk}] -to [get_clocks {pll|altpll_component|auto_generated|pll1|clk[0]}] -hold -end 2   --- Quote End ---   OUTPUT SIGNALS   --- Quote Start ---   set fwclk [get_clocks sdram_clk];# forwarded clock name (generated using create_generated_clock at output clock port) set fwclk2 [get_clocks sdram2_clk]; set tsu 1.500; # destination device setup time requirement set thd 0.800; # destination device hold time requirement set trce_dly_max 0.240; # maximum board trace delay set trce_dly_min 0.200; # minimum board trace delay set output_ports [get_ports {  sdram_cke_o  sdram_csn_o  sdram_rasn_o  sdram_casn_o  sdram_wen_o  sdram_dq_io[*]  sdram_ba_o[*]  sdram_a_o[*]  sdram_dqm_o[*]  sdram_dq_io[*] }]; # list of output ports  set output_ports2 [get_ports {  sdram2_cke_o  sdram2_csn_o  sdram2_rasn_o  sdram2_casn_o  sdram2_wen_o  sdram2_dq_io[*]  sdram2_ba_o[*]  sdram2_a_o[*]  sdram2_dqm_o[*]  sdram2_dq_io[*] }]; # list of output ports  # Output Delay Constraints set_output_delay -clock $fwclk -max [expr $trce_dly_max + $tsu - $trce_dly_min] $output_ports; set_output_delay -clock $fwclk -min [expr $trce_dly_min - $thd - $trce_dly_max] $output_ports;  set_output_delay -clock $fwclk -max [expr $trce_dly_max + $tsu - $trce_dly_min] $output_ports2; set_output_delay -clock $fwclk -min [expr $trce_dly_min - $thd - $trce_dly_max] $output_ports2;   --- Quote End ---    Next problem is this and could be related to phase shift coz for both chips needs to be different. On SDRAM1 all traces are 28mm, all but CLK who is 23mm. However on SDRAM2 all traces are also 28mm, only CLK is 14mm. This complicates things even more and right now I don't see option how to solve this. If someone is willing to discuss this serious complication I would really appreciate. If there is need for more data i can provide.

[Altera_Forum]

I see that lot of people opened this thread but no one replied. Is the problem so complicated or no one was interested to try to resolve it. I would at least want that someone who knows tells me that this is impossible to do so I can see my other options.

[Altera_Forum]

The 14 mm difference in trace lengths only amounts to about 100 ps skew, this would only be of impact at the higher frequencies. 

There are a few other unknowns: Do you use 1 or 2 controllers? Do you use Altera IP, Third Party IP, or did you write your own? What is your target operating frequency? The full 166 MHz?

[Altera_Forum]

Hi, thank you a LOT for your reply! I m using controller my friend did but we stuck.  

What is interesting here is when I run tests on one SDRAM chip it goes even more than 166MHz and behaves stable but problem is to run them together at that freq. Like I mentioned I think that problem is that each SDRAM uses different PLL and that clock traces are also different. Can you point me somewhere to read or what to do, what to investigate first. PLLs are using 7MHz source so since 2 PLLs are generated question is are they in sync, if they are not how to make them? I m talking about clk from FPGA to SDRAM. Next question is about phase shift because I think that each SDRAM for each PLL used needs different shift.

[Altera_Forum]

If you are using a single controller, you should be able to run the two sides with a single PLL.

[Altera_Forum]

Yes that's something I was doing before but since SDRAM chips are placed on opposite sides of FPGA like you can see on the picture on first post so I was thinking that there is my problem. Then I decided to generate two PLL and tell quartus what PLL should use. Before situation was that quartus used only PLL4 and not PLL3 where 2nd SDRAM is connected.