SRAM vs Flash based FPGA one more time

Hello, everybody!

I have come with a question concerning FPGA technology. I understand
that this question has been already discussed here, but I still don't
understand one thing..

Everybody knows that from user point of view Flash based FPGA are
better because they are nonvvolatile, tolerable to radiation, do not
consume high current at startup and better secure intellectual
property. On the other hand SRAM FPGA are much easier to manufacture.

That's it? Being easier to manufacture is the point? I don't
understand why large vendors like Xilinx and Altera can not invest
money and design true Flash FPGA, like Actel did.

And the last question: why largest Actel FPGA chip (3 M system gates)
is much smaller than say largest Xilinx's chips? Is it still much
harder to manufacture larger Flash FPGA chip?

Thank you!

urock,

It simply comes down to manufacturing technology. Processes that
combine Flash and CMOS are typically 3 generations behind leading-edge
CMOS processes. As we speak, Actel's flash-based FPGAs are
manufactured at 130nm while the big guys have 40nm SRAM-based devices.
This leads to ~10X density advantage for SRAM over Flash FPGAs,
yielding larger and/or cheaper devices.

 - gael

I got it, gael, thanks!

So you think that when moors law stops working (when manufacturing
process reaches 20 nm), all FPGAs will be Flash-based? I mean will in
your opinion flash technology catch up with cmos?

> Everybody knows that from user point of view Flash based FPGA are
> better because they are nonvvolatile, tolerable to radiation, do not
> consume high current at startup and better secure intellectual
> property. On the other hand SRAM FPGA are much easier to manufacture.
>
> That's it?


No.

All I'll say is that if you're used to the sort of P&R results you get
with Altera/Xilinx tools and devices be wary of committing to a transfer
to a Flash based device without a _lot_ of experimenting first.


Nial. 

On Jun 25, 2:12=A0pm, "Nial Stewart"
<nial*REMOVE_TH...@nialstewartdevelopments.co.uk> wrote:
> > Everybody knows that from user point of view Flash based FPGA are
> > better because they are nonvvolatile, tolerable to radiation, do not
> > consume high current at startup and better secure intellectual
> > property. On the other hand SRAM FPGA are much easier to manufacture.
>
> > That's it?
>
> No.
>
> All I'll say is that if you're used to the sort of P&R results you get
> with Altera/Xilinx tools and devices be wary of committing to a transfer
> to a Flash based device without a _lot_ of experimenting first.
>
> Nial.

well that counts for ACTEL yes!!!!

Lattice is almost like Xilinx, even has distributed RAM (only SRL16
mode is missing)
ic65L is like old Xilinx LUT4FF

but in generic yes, need run real P&R and compare actual designs
before
doing any decisions

Antti

urock wrote:
> Everybody knows that from user point of view Flash based FPGA are
> better because they are nonvvolatile, tolerable to radiation, do not
> consume high current at startup and better secure intellectual
> property. On the other hand SRAM FPGA are much easier to manufacture.

Everybody does not share those positive points. For example I think that
SRAM based FPGA is easier in terms of image management. It can be easily
connected to an external CPU, that has access to big flash etc. You can
easily put many different images to external flash, and load them as
needed, or even use dynamic reconfiguration.

Also the high load current was more of a problem in the past. It's not
that bad anymore compared to the current in normal mode. And IP
can be protected via encrypted configuration streams in most of the new
SRAM based FPGAs.

Radiation tolerance is possibly better for flash based devices, but I
have not compared the FIT rates for configuration data (I think usually
just combined FIT is shown. The SRAM based configuration bits are not
traditional sram cells as far as I know. And for normal functional logic
same problems are shared in terms of SEU.

--Kim

On Jun 25, 9:02=A0am, Kim Enkovaara <kim.enkova...@iki.fi> wrote:
> urock wrote:
> > Everybody knows that from user point of view Flash based FPGA are
> > better because they are nonvvolatile, tolerable to radiation, do not
> > consume high current at startup and better secure intellectual
> > property. On the other hand SRAM FPGA are much easier to manufacture.
>
> Everybody does not share those positive points. For example I think that
> SRAM based FPGA is easier in terms of image management. It can be easily
> connected to an external CPU, that has access to big flash etc. You can
> easily put many different images to external flash, and load them as
> needed, or even use dynamic reconfiguration.

Why can't you do that with Flash based FPGAs?  At least the Lattice
parts are a RAM based FPGA with a flash configuration memory
integrated.  So they have the advantages of both RAM and flash based
parts.  They are low cost too.  That is the main reason that Xilinx
has touted for not making flash based parts, that the flash process is
behind the power curve and parts can be made cost effectively.  BAH!


> Also the high load current was more of a problem in the past. It's not
> that bad anymore compared to the current in normal mode. And IP
> can be protected via encrypted configuration streams in most of the new
> SRAM based FPGAs.
>
> Radiation tolerance is possibly better for flash based devices, but I
> have not compared the FIT rates for configuration data (I think usually
> just combined FIT is shown. The SRAM based configuration bits are not
> traditional sram cells as far as I know. And for normal functional logic
> same problems are shared in terms of SEU.

Not if they are RAM based with flash config mem.  These parts get the
advantages of both, but also the disadvantages of both if that makes
sense...

Rick

On Jun 25, 7:22=A0am, "Antti.Luk...@googlemail.com"
<Antti.Luk...@googlemail.com> wrote:
> On Jun 25, 2:12=A0pm, "Nial Stewart"
>
> <nial*REMOVE_TH...@nialstewartdevelopments.co.uk> wrote:
> > > Everybody knows that from user point of view Flash based FPGA are
> > > better because they are nonvvolatile, tolerable to radiation, do not
> > > consume high current at startup and better secure intellectual
> > > property. On the other hand SRAM FPGA are much easier to manufacture.
>
> > > That's it?
>
> > No.
>
> > All I'll say is that if you're used to the sort of P&R results you get
> > with Altera/Xilinx tools and devices be wary of committing to a transfe=
r
> > to a Flash based device without a _lot_ of experimenting first.
>
> > Nial.
>
> well that counts for ACTEL yes!!!!
>
> Lattice is almost like Xilinx, even has distributed RAM (only SRL16
> mode is missing)
> ic65L is like old Xilinx LUT4FF
>
> but in generic yes, need run real P&R and compare actual designs
> before
> doing any decisions
>
> Antti

Lattice's mixed flash / SRAM parts also don't have the same
instant-on and radiation tolerance features of the Actel parts.
So in essence you lose some of the features to gain better
density and architecture.

On Jun 25, 2:49 pm, gabor <ga...@alacron.com> wrote:
> On Jun 25, 7:22 am, "Antti.Luk...@googlemail.com"
>
>
>
> <Antti.Luk...@googlemail.com> wrote:
> > On Jun 25, 2:12 pm, "Nial Stewart"
>
> > <nial*REMOVE_TH...@nialstewartdevelopments.co.uk> wrote:
> > > > Everybody knows that from user point of view Flash based FPGA are
> > > > better because they are nonvvolatile, tolerable to radiation, do not
> > > > consume high current at startup and better secure intellectual
> > > > property. On the other hand SRAM FPGA are much easier to manufacture.
>
> > > > That's it?
>
> > > No.
>
> > > All I'll say is that if you're used to the sort of P&R results you get
> > > with Altera/Xilinx tools and devices be wary of committing to a transfer
> > > to a Flash based device without a _lot_ of experimenting first.
>
> > > Nial.
>
> > well that counts for ACTEL yes!!!!
>
> > Lattice is almost like Xilinx, even has distributed RAM (only SRL16
> > mode is missing)
> > ic65L is like old Xilinx LUT4FF
>
> > but in generic yes, need run real P&R and compare actual designs
> > before
> > doing any decisions
>
> > Antti
>
> Lattice's mixed flash / SRAM parts also don't have the same
> instant-on and radiation tolerance features of the Actel parts.
> So in essence you lose some of the features to gain better
> density and architecture.

Certainly radiation tolerance is a seldom used feature that has use
only in very specialized applications.  When is the "instant on"
feature needed really?  I don't think I have had an app that needed
"instant on" in the 30 years I have been designing electronics.
Digital stuff always has a reset to hold it off until all power, etc
is ready for operation.  When I use an FPGA, I use one of the FPGA
outputs to hold the rest of the circuit in reset so the FPGA is the
first thing to come alive.  Even if the FPGA is on a PCI bus, I
believe they have provision to give devices time to boot themselves
before they have to respond, no?

Is "instant on" another seldom needed feature?

Rick

On Jun 25, 3:00=A0pm, rickman <gnu...@gmail.com> wrote:
> On Jun 25, 2:49 pm, gabor <ga...@alacron.com> wrote:
>
>
>
> > On Jun 25, 7:22 am, "Antti.Luk...@googlemail.com"
>
> > <Antti.Luk...@googlemail.com> wrote:
> > > On Jun 25, 2:12 pm, "Nial Stewart"
>
> > > <nial*REMOVE_TH...@nialstewartdevelopments.co.uk> wrote:
> > > > > Everybody knows that from user point of view Flash based FPGA are
> > > > > better because they are nonvvolatile, tolerable to radiation, do =
not
> > > > > consume high current at startup and better secure intellectual
> > > > > property. On the other hand SRAM FPGA are much easier to manufact=
ure.
>
> > > > > That's it?
>
> > > > No.
>
> > > > All I'll say is that if you're used to the sort of P&R results you =
get
> > > > with Altera/Xilinx tools and devices be wary of committing to a tra=
nsfer
> > > > to a Flash based device without a _lot_ of experimenting first.
>
> > > > Nial.
>
> > > well that counts for ACTEL yes!!!!
>
> > > Lattice is almost like Xilinx, even has distributed RAM (only SRL16
> > > mode is missing)
> > > ic65L is like old Xilinx LUT4FF
>
> > > but in generic yes, need run real P&R and compare actual designs
> > > before
> > > doing any decisions
>
> > > Antti
>
> > Lattice's mixed flash / SRAM parts also don't have the same
> > instant-on and radiation tolerance features of the Actel parts.
> > So in essence you lose some of the features to gain better
> > density and architecture.
>
> Certainly radiation tolerance is a seldom used feature that has use
> only in very specialized applications. =A0When is the "instant on"
> feature needed really? =A0I don't think I have had an app that needed
> "instant on" in the 30 years I have been designing electronics.
> Digital stuff always has a reset to hold it off until all power, etc
> is ready for operation. =A0When I use an FPGA, I use one of the FPGA
> outputs to hold the rest of the circuit in reset so the FPGA is the
> first thing to come alive. =A0Even if the FPGA is on a PCI bus, I
> believe they have provision to give devices time to boot themselves
> before they have to respond, no?
>
> Is "instant on" another seldom needed feature?
>
> Rick

The point I was trying to make is that all "Flash-based" FPGA's
are not the same.  Lattice's are really SRAM-based using a mixed
SRAM / Flash process with backup storage in the on-chip flash.
Xilinx's 3AN series are multi-die solutions with a standard
SRAM-based part and a SPI flash packaged together.  The OP
seemed to like some of the Actel features that don't exist
in these hybrid devices.

"Instant" on may not be needed often, but there are many
applications where the startup delay time matters.  As the
OP was referring to "the largest parts" in the various
series, these delays are not trivial, and can screw up
a system that needs the FPGA resources before a PCI
starts is enumeration, for example.  Getting a large
Virtex 5 to start up in under 1 second is not a simple
task.  There's no indication of flash-based devices
being anything more than a side line for Xilinx, but
if Virtex 5 or 6 comes out in a Spartan 3AN-like
hybrid version, you can be sure it won't come up
"instantly."

Cheers,
Gabor