5V PCI interface

Hi!

I'm making a PCI card which currently has the Xilinx Spartan-II
designed in beacuse of its ability to interface directly to the 5V bus.

I notice that the Spartan-II has partly disappeared from the Xilinx site,
as in hard to find (not longer listed amongst their FPGA products)
and not available in their online shop anymore.

I strikes me that "the world" really needs a 5V capable device,
and that it's just a bit strange that a major manufacturer decides
to no longer make devices that has 5V support. It is likely that
this capability will be useful for another few tens of years.

Of course there are level converters and that is probably the obvious
solution. However, that seems inelegant, and they bring in a new set
of problems, such as increased cost and complexity and increased
set up and hold times etc.. Also there seems to not be a particularily
good selection of level converters.

Thus I have two questions:

1.
How do you go about building FPGA based PCI cards, or indeed
anything that needs to interface to 5V; What's your favourite
device (family) for this? Do you use level converters, if so which ones?

2.
I'd prefer that the parts we end up with will be easily available
in small quantities for the next 4 or 5 years.
Does that make the Spartan-II a bad choice?

Any advice will be appreciated - thanks!

DJ
--

Dr,

Spartan 2 will be around a long time.  That we have demoted it from the 
limelight is a marketing issue (just so much shelf space for the new 
products to showcase).

As you may be aware, we still provide the 3100A series of FPGAs, which 
are still supporting designs done 15 years ago!

We discontinue devices once they are not able to be manufactured and 
sold economically.  This means that there is little business, and the 
process used to make the chips has become obsolete at the fabrication 
facilities.  We also may discontinue a particular part/package 
combination when that package is running at extremely low volumes or 
becomes difficult to procure.

Since we are still making almost all of our FPGA products, I don't think 
you have anything to worry about with Spartan II.

The original Virtex, and Spartan II are a lot like classic Coca-Cola -- 
they may never go away.

However, the cost/function of newer devices is so much better than the 
older devices, that you may want to consider designing with the latest 
devices (at some point).

The app notes we have published for 5V PCI details all of the tricks to 
make the latest 90nm devices work on the 5V PCI bus. (Xapp 646, 311)

I hope this helps,

Austin

Austin,

thanks a lot for your reply - that was most reassuring,
and I got both questions answered!

> The original Virtex, and Spartan II are a lot like classic Coca-Cola -- 
> they may never go away.

That's kind of what I was hoping for :-)

> The app notes we have published for 5V PCI details all of the tricks to
> make the latest 90nm devices work on the 5V PCI bus. (Xapp 646, 311)

I have overlooked said application note - shall download it at once
and study it carefully. Anyway it's good to know that the Spartan-II
will remain in production.

Regards,

DJ
--

Austin,
Maybe you can give me more insight to a problem I have with xapp646. The 
note states that "Since the device is a set of series-connected NMOS 
transistors, any voltage larger than a few hundred millivolts below the VCC 
pin voltage will be cut off."
From reading the IDT appnotes and what I'm seeing on a circuit board, the 
output will always  be limited to less than VCC-1. With VCC at 3.3v as shown 
in xapp646, under light loading, the output voltage is about 2.3v, and with 
a 10k load, it's closer to 2v which means essentially no noise margin for 
TTL.  Look at figure 4 of http://www1.idt.com/pcms/tempDocs/AN_11.pdf  or 
figure 5 of  http://www1.idt.com/pcms/tempDocs/quickswitch_basics.pdf
Do you think that I should be seeing around 2 to 2.3v output with the ckt 
shown in xapp646?

Dr, take a look at TI's sn74cb3t3384 or sn74cbtd3384c as well as some 
appnotes on their site.


gja

"Austin Lesea" <austin@xilinx.com> wrote in message 
news:d3gogs$lr91@cliff.xsj.xilinx.com...
> Dr,
>
> Spartan 2 will be around a long time.  That we have demoted it from the 
> limelight is a marketing issue (just so much shelf space for the new 
> products to showcase).
>
> As you may be aware, we still provide the 3100A series of FPGAs, which are 
> still supporting designs done 15 years ago!
>
> We discontinue devices once they are not able to be manufactured and sold 
> economically.  This means that there is little business, and the process 
> used to make the chips has become obsolete at the fabrication facilities. 
> We also may discontinue a particular part/package combination when that 
> package is running at extremely low volumes or becomes difficult to 
> procure.
>
> Since we are still making almost all of our FPGA products, I don't think 
> you have anything to worry about with Spartan II.
>
> The original Virtex, and Spartan II are a lot like classic Coca-Cola -- 
> they may never go away.
>
> However, the cost/function of newer devices is so much better than the 
> older devices, that you may want to consider designing with the latest 
> devices (at some point).
>
> The app notes we have published for 5V PCI details all of the tricks to 
> make the latest 90nm devices work on the 5V PCI bus. (Xapp 646, 311)
>
> I hope this helps,
>
> Austin 

"gja" <geeja@hotmail.com> wrote in message
news:luY6e.1145$FQ.507@fe10.lga...
[snip]
> Dr, take a look at TI's sn74cb3t3384 or sn74cbtd3384c as well as some
> appnotes on their site.

Will do. Thanks for the tip!

DJ
--

gja,

Basically, I am using the fact that the IDT device is just a simple NMOS 
transistor, and since I know how that works (physically) I am ignoring 
the data sheet (as it is misleading in this case).

I know that IDT does not support this from their data sheet 
specifications, and they actually called me to tell me that they would 
not support this.

Odd.  It works fine.  They are sand-bagging their specifications like 
crazy here, and their parts work far better than the data sheet implies 
(in this circuit).

Could be the loading (none), could be the voltages (less variation than 
what they spec), could be they don't want to support the application. 
Fine, call Xilinx.  I'd much rather you call us than IDT.  OK by me.  We 
have built it, used it, tested it, and are still doing so.  I know a lot 
of folks out there who have done likewise.  Haven't heard a single 
complaint.

Officially, the PCI specification does not allow any devices to be 
placed in series with a PCI compatible part.  That is fine as well.

Austin

gja wrote:
> Austin,
> Maybe you can give me more insight to a problem I have with xapp646. The 
> note states that "Since the device is a set of series-connected NMOS 
> transistors, any voltage larger than a few hundred millivolts below the VCC 
> pin voltage will be cut off."
> From reading the IDT appnotes and what I'm seeing on a circuit board, the 
> output will always  be limited to less than VCC-1. With VCC at 3.3v as shown 
> in xapp646, under light loading, the output voltage is about 2.3v, and with 
> a 10k load, it's closer to 2v which means essentially no noise margin for 
> TTL.  Look at figure 4 of http://www1.idt.com/pcms/tempDocs/AN_11.pdf  or 
> figure 5 of  http://www1.idt.com/pcms/tempDocs/quickswitch_basics.pdf
> Do you think that I should be seeing around 2 to 2.3v output with the ckt 
> shown in xapp646?
> 
> Dr, take a look at TI's sn74cb3t3384 or sn74cbtd3384c as well as some 
> appnotes on their site.
> 
> 
> gja
> 
> "Austin Lesea" <austin@xilinx.com> wrote in message 
> news:d3gogs$lr91@cliff.xsj.xilinx.com...
> 
>>Dr,
>>
>>Spartan 2 will be around a long time.  That we have demoted it from the 
>>limelight is a marketing issue (just so much shelf space for the new 
>>products to showcase).
>>
>>As you may be aware, we still provide the 3100A series of FPGAs, which are 
>>still supporting designs done 15 years ago!
>>
>>We discontinue devices once they are not able to be manufactured and sold 
>>economically.  This means that there is little business, and the process 
>>used to make the chips has become obsolete at the fabrication facilities. 
>>We also may discontinue a particular part/package combination when that 
>>package is running at extremely low volumes or becomes difficult to 
>>procure.
>>
>>Since we are still making almost all of our FPGA products, I don't think 
>>you have anything to worry about with Spartan II.
>>
>>The original Virtex, and Spartan II are a lot like classic Coca-Cola -- 
>>they may never go away.
>>
>>However, the cost/function of newer devices is so much better than the 
>>older devices, that you may want to consider designing with the latest 
>>devices (at some point).
>>
>>The app notes we have published for 5V PCI details all of the tricks to 
>>make the latest 90nm devices work on the 5V PCI bus. (Xapp 646, 311)
>>
>>I hope this helps,
>>
>>Austin 
> 
> 
> 

Hi Austin,

> Officially, the PCI specification does not allow any devices to be
> placed in series with a PCI compatible part.  That is fine as well.

Hmmm... I thought that PCI-Compliant was "nothing in series", and
PCI-Compatible was "whatever you need to do to make it work".

Apart from that, I second all your statements there. Plus, I would like to
repeat here, TI seems to be more open to use for this type of application
of their sn74cbtd3384c parts.

Best regards,


Ben

Austin, are you saying that you've actually seen the circuit actually pass 
3v across the switch ? Or are you saying that the output SHOULD be clamped 
around 2.3v ?

Perhaps I should've been more clear, but I am currently testing a new ckt 
board built with the circuit in xapp646 and a virtex2 part.
With the IDT  vcc at 3.3v, the maximum voltage I am seeing with a scope is 
about 2.3v on the output side of the switch. IE., when the virtex2 part is 
driving (lvttl) one side of the switch at 3.3v, the other side of the switch 
is about 2.3v. When the 5v ttl part 74FCT645 is writing to the virtex2 part, 
I see 4v on that side of the switch and 2.3v on the virtex2 side. When I 
originally looked at xapp646, it stated that it would clamp to less than ~3v 
outputs, I didn't think it would be closer to 2.3v.  If I had used the 
implementation in IDT's  or TI's appnote where they drive the vcc pin at 
4.3v using a diode from 5v, I wonder if that would have worked better.

gja


"Austin Lesea" <austin@xilinx.com> wrote in message 
news:d3jd3k$mb11@cliff.xsj.xilinx.com...
> gja,
>
> Basically, I am using the fact that the IDT device is just a simple NMOS 
> transistor, and since I know how that works (physically) I am ignoring the 
> data sheet (as it is misleading in this case).
>
> I know that IDT does not support this from their data sheet 
> specifications, and they actually called me to tell me that they would not 
> support this.
>
> Odd.  It works fine.  They are sand-bagging their specifications like 
> crazy here, and their parts work far better than the data sheet implies 
> (in this circuit).
>
> Could be the loading (none), could be the voltages (less variation than 
> what they spec), could be they don't want to support the application. 
> Fine, call Xilinx.  I'd much rather you call us than IDT.  OK by me.  We 
> have built it, used it, tested it, and are still doing so.  I know a lot 
> of folks out there who have done likewise.  Haven't heard a single 
> complaint.
>
> Officially, the PCI specification does not allow any devices to be placed 
> in series with a PCI compatible part.  That is fine as well.
>
> Austin
>
> gja wrote:
>> Austin,
>> Maybe you can give me more insight to a problem I have with xapp646. The 
>> note states that "Since the device is a set of series-connected NMOS 
>> transistors, any voltage larger than a few hundred millivolts below the 
>> VCC pin voltage will be cut off."
>> From reading the IDT appnotes and what I'm seeing on a circuit board, the 
>> output will always  be limited to less than VCC-1. With VCC at 3.3v as 
>> shown in xapp646, under light loading, the output voltage is about 2.3v, 
>> and with a 10k load, it's closer to 2v which means essentially no noise 
>> margin for TTL.  Look at figure 4 of 
>> http://www1.idt.com/pcms/tempDocs/AN_11.pdf  or figure 5 of 
>> http://www1.idt.com/pcms/tempDocs/quickswitch_basics.pdf
>> Do you think that I should be seeing around 2 to 2.3v output with the ckt 
>> shown in xapp646?
>>
>> Dr, take a look at TI's sn74cb3t3384 or sn74cbtd3384c as well as some 
>> appnotes on their site.
>>
>>
>> gja
>>
>> "Austin Lesea" <austin@xilinx.com> wrote in message 
>> news:d3gogs$lr91@cliff.xsj.xilinx.com...
>>
>>>Dr,
>>>
>>>Spartan 2 will be around a long time.  That we have demoted it from the 
>>>limelight is a marketing issue (just so much shelf space for the new 
>>>products to showcase).
>>>
>>>As you may be aware, we still provide the 3100A series of FPGAs, which 
>>>are still supporting designs done 15 years ago!
>>>
>>>We discontinue devices once they are not able to be manufactured and sold 
>>>economically.  This means that there is little business, and the process 
>>>used to make the chips has become obsolete at the fabrication facilities. 
>>>We also may discontinue a particular part/package combination when that 
>>>package is running at extremely low volumes or becomes difficult to 
>>>procure.
>>>
>>>Since we are still making almost all of our FPGA products, I don't think 
>>>you have anything to worry about with Spartan II.
>>>
>>>The original Virtex, and Spartan II are a lot like classic Coca-Cola -- 
>>>they may never go away.
>>>
>>>However, the cost/function of newer devices is so much better than the 
>>>older devices, that you may want to consider designing with the latest 
>>>devices (at some point).
>>>
>>>The app notes we have published for 5V PCI details all of the tricks to 
>>>make the latest 90nm devices work on the 5V PCI bus. (Xapp 646, 311)
>>>
>>>I hope this helps,
>>>
>>>Austin
>>
>> 

gja,

See below,

Austin

gja wrote:
> Austin, are you saying that you've actually seen the circuit actually pass 
> 3v across the switch ?
Yes, I have.
  Or are you saying that the output SHOULD be clamped
> around 2.3v ?
No, I am not.
> 
> Perhaps I should've been more clear, but I am currently testing a new ckt 
> board built with the circuit in xapp646 and a virtex2 part.
> With the IDT  vcc at 3.3v, the maximum voltage I am seeing with a scope is 
> about 2.3v on the output side of the switch. IE., when the virtex2 part is 
> driving (lvttl) one side of the switch at 3.3v, the other side of the switch 
> is about 2.3v.
Into what load?  On a "real" PCI bus is where I made the measurements. 
Perhaps if you use the resistor "standard termionation" (which is 
anything but a standard) you will get some other result?
  When the 5v ttl part 74FCT645 is writing to the virtex2 part,
> I see 4v on that side of the switch and 2.3v on the virtex2 side.
OK.  I will admit that the TI part has some advantages, but it is also 
an active device, and adds delay, doesn't it?
  When I
> originally looked at xapp646, it stated that it would clamp to less than ~3v 
> outputs, I didn't think it would be closer to 2.3v.  If I had used the 
> implementation in IDT's  or TI's appnote where they drive the vcc pin at 
> 4.3v using a diode from 5v, I wonder if that would have worked better.
Except that then you do not limit the voltages to less than what we require.

Hi Austin,

> OK.  I will admit that the TI part has some advantages, but it is also
> an active device, and adds delay, doesn't it?

Yep. About 500ps worst-case. I tell my users to edit the constraint file so
that because of this, all PCI signals should have 1ns subtracted from their
timing constraints. This takes care of the combinatorial delays as well. 

In Quartus, the difference is unnoticeable with a 33MHz bus and usually adds
a little compilation time with a 66MHz bus. I'd hope the same situation
would apply to Xilinx users.

Best regards,



Ben