SDRAM Controller

Hi!
I am a bit new to FPGAs, so far I have only worked with CPLDs ( Xilinx
9500 family ). Now I would like to use a Spartan 2E ( with WebPack 5.2
and VHDL ) to make a SDRAM controller.

I have searched this archive but I haven't found any topic related to
my question. Here's the deal. When data needs to be transfered to the
SDRAM the controller sends out data prior to generating the rising
edge of the SDRAM's clock. So the way I see it, the fpga must generate
two clocks, one for the controller and one for the SDRAM. The SDRAM
clock must be by 90� out of phase of the controller clock ( the SDRAM
clock must be delayed for 1/4 of the cycle of the controller clock ),
in order for data to appear on DQ lines before the rising edge of
SDRAM's clk. I am right so far, or am I way off? And If I am right,
how do you generate the SDRAM clock ( do you use a DLL to phase-shift
the input clock ? ).

Best regards
George Mercury

If you are using Virtex-II or Spartan-III, the Xilinx DCM block
provides 4 clock outputs, at 90 degree intervals.
See XAPP134 & 200 for some ideas.

george_mercury@hotmail.com (George) wrote:

:Hi!
:I am a bit new to FPGAs, so far I have only worked with CPLDs ( Xilinx
:9500 family ). Now I would like to use a Spartan 2E ( with WebPack 5.2
:and VHDL ) to make a SDRAM controller.
:
:I have searched this archive but I haven't found any topic related to
:my question. Here's the deal. When data needs to be transfered to the
:SDRAM the controller sends out data prior to generating the rising
:edge of the SDRAM's clock. So the way I see it, the fpga must generate
:two clocks, one for the controller and one for the SDRAM. The SDRAM
:clock must be by 90� out of phase of the controller clock ( the SDRAM
:clock must be delayed for 1/4 of the cycle of the controller clock ),
:in order for data to appear on DQ lines before the rising edge of
:SDRAM's clk. I am right so far, or am I way off? And If I am right,
:how do you generate the SDRAM clock ( do you use a DLL to phase-shift
:the input clock ? ).
:
:Best regards
:George Mercury

You run the state machine of you controller on negedge of clk and
SDRAMs sample data of posedge of clock.

Sumit

george_mercury@hotmail.com (George) wrote in message news:<6d167a0a.0310260419.50c5076d@posting.google.com>...
> Hi!
> I am a bit new to FPGAs, so far I have only worked with CPLDs ( Xilinx
> 9500 family ). Now I would like to use a Spartan 2E ( with WebPack 5.2
> and VHDL ) to make a SDRAM controller.
> 
> I have searched this archive but I haven't found any topic related to
> my question. Here's the deal. When data needs to be transfered to the
> SDRAM the controller sends out data prior to generating the rising
> edge of the SDRAM's clock. So the way I see it, the fpga must generate
> two clocks, one for the controller and one for the SDRAM. The SDRAM
> clock must be by 90&#4294967295; out of phase of the controller clock ( the SDRAM
> clock must be delayed for 1/4 of the cycle of the controller clock ),
> in order for data to appear on DQ lines before the rising edge of
> SDRAM's clk. I am right so far, or am I way off? And If I am right,
> how do you generate the SDRAM clock ( do you use a DLL to phase-shift
> the input clock ? ).
> 
> Best regards
> George Mercury

george_mercury@hotmail.com (George) writes:

> Hi!
> I am a bit new to FPGAs, so far I have only worked with CPLDs ( Xilinx
> 9500 family ). Now I would like to use a Spartan 2E ( with WebPack 5.2
> and VHDL ) to make a SDRAM controller.
> 
> I have searched this archive but I haven't found any topic related to
> my question. Here's the deal. When data needs to be transfered to the
> SDRAM the controller sends out data prior to generating the rising
> edge of the SDRAM's clock. So the way I see it, the fpga must generate
> two clocks, one for the controller and one for the SDRAM. The SDRAM
> clock must be by 90� out of phase of the controller clock ( the SDRAM
> clock must be delayed for 1/4 of the cycle of the controller clock ),
> in order for data to appear on DQ lines before the rising edge of
> SDRAM's clk. I am right so far, or am I way off? And If I am right,
> how do you generate the SDRAM clock ( do you use a DLL to phase-shift
> the input clock ? ).
> 

Xilinx have an appnote on this.  And Altera also.  As for me, assuming
you're talking about single data-rate:

I just ran my state machine off the same clock as the SDRAMs.  If you
drive the data on the rising edge, it won't appear on the outputs
until some time later (about 6/7 ns IIRC for my chip), so the SDRAM
will then latch it in on the *next* clock cycle.

No need to faff about generating mutplile clock domains, using
negative edges or any of the other nasties.  Just make sure the
latency in whatever is provding the data is right.

Micron provided good VHDL models to get the simulation right - much
easier than wiring up a logic analyser!


If you're doing DDR, ignore me :-)

HTH,
Martin

-- 
martin.j.thompson@trw.com
TRW Conekt, Solihull, UK
http://www.trw.com/conekt

Thanks, I will consider using both propositions! One more thing, what
exactly is a state-machine? I have read this expression about a milion
times in datasheets and application notes, but I stil don't know what
it is.

George

george_mercury@hotmail.com (George) wrote in message news:<6d167a0a.0310260419.50c5076d@posting.google.com>...
> Hi!
> I am a bit new to FPGAs, so far I have only worked with CPLDs ( Xilinx
> 9500 family ). Now I would like to use a Spartan 2E ( with WebPack 5.2
> and VHDL ) to make a SDRAM controller.
> 
> I have searched this archive but I haven't found any topic related to
> my question. Here's the deal. When data needs to be transfered to the
> SDRAM the controller sends out data prior to generating the rising
> edge of the SDRAM's clock. So the way I see it, the fpga must generate
> two clocks, one for the controller and one for the SDRAM. The SDRAM
> clock must be by 90&#4294967295; out of phase of the controller clock ( the SDRAM
> clock must be delayed for 1/4 of the cycle of the controller clock ),
> in order for data to appear on DQ lines before the rising edge of
> SDRAM's clk. I am right so far, or am I way off? And If I am right,
> how do you generate the SDRAM clock ( do you use a DLL to phase-shift
> the input clock ? ).

Assumption: Single-data-rate SDRAM.  

You're making the whole thing WAAAAY too complicated.

Your SDRAM controller should obviously be a synchronous state machine.
 Clock the SDRAM and the controller with the same clock -- it's common
to use special clock-buffer chips to do this.

You have a synchronous controller, and you should register the data
outputs, too: this way, the SDRAM command and data lines all have the
same clock to out.

Say you're running your SDRAM and controller at 100 MHz (to keep the
math simple).  Say your FPGA (or whatever) has clock-to-out on the
pads spec'd at 5 ns.  Say the SDRAM has 2 ns setup and 1 ns hold
times.  Say you've laid out your board such that the flight time of
the traces between the controller and the SDRAM is 1 ns.

This all means that command and data asserted on clock edge 1 will be
stable by edge 1 + 5 ns.  Therefore, that command and data are valid
on the SDRAM pins on clock edge 1 + 6 ns.  They meet SDRAM setup with
a 2 ns margin, and will be registered in the SDRAM on clock edge 2.

In other words, it's Synchronous Design 101, and nothing special is
required in order for you to meet the SDRAM's timing specs.

-a

George wrote:

> Thanks, I will consider using both propositions! One more thing, what
> exactly is a state-machine? I have read this expression about a milion
> times in datasheets and application notes, but I stil don't know what
> it is.
> 
> George

A state machine is like a little hardwired program inside of your design. 

It's called a state machine because it works on a so-called state register
and a number of inputs to generate outputs, and a new value of the state
register.

This state register can be implemented in umpteen ways, and how the state
machine decides to jump from one state to the next can be implemented in
even more ways. I'd say just look up "Moore Mealy FSM" on Google to get the
hang of the idea.

Best regards,


-- 
Ben

Hello George,

> One more thing, what exactly is a state-machine?

I have a presentation that discusses finite state
machines -- with a simple example in Verilog, at:

http://www.engr.sjsu.edu/crabill/vlogfsm.pdf

If you are looking for a ready-made SDRAM controller,
you can find one on the Xilinx website, as XAPP134.
I've used it in a project and it works properly.

Good luck,
Eric