Large/Fast static RAM

  • Thread starter Martin Euredjian
  • Start date
M

Martin Euredjian

Guest
I'm trying to determine if anyone makes a large/fast static RAM part. 16M
(or more) x 16 bits (or more), 10ns.

I can't afford the address-to-data-out latency of dynamic RAM. There are
ways around this, of course, but SRAM would be so much simpler.

Any ideas?

Thanks,


--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian

To send private email:
0_0_0_0_@pacbell.net
where
"0_0_0_0_" = "martineu"
 
I don't believe that you'll find a monolithic 16Mx16 SRAM. You'll need
multiple devices. You can start your search on QuestLink. There are plenty
of 10 ns, x16 asynchronous RAMs out there. There are also some x32, 200
MHz+ synchronous RAMs.

http://www.questlink.com/categorySearch_sub.jhtml?type=cat&param=130700

-- Steve Knapp

"Martin Euredjian" <0_0_0_0_@pacbell.net> wrote in message
news:lEgLb.1476$rt.397@newssvr29.news.prodigy.com...
I'm trying to determine if anyone makes a large/fast static RAM part. 16M
(or more) x 16 bits (or more), 10ns.

I can't afford the address-to-data-out latency of dynamic RAM. There are
ways around this, of course, but SRAM would be so much simpler.

Any ideas?

Thanks,


--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian

To send private email:
0_0_0_0_@pacbell.net
where
"0_0_0_0_" = "martineu"
Click to expand...
 
Check ISSI, they have a 18 Mb (x18,x36,x72), synchronous static ram.
Seems they have a very low latency, and support bursts.


"Martin Euredjian" <0_0_0_0_@pacbell.net> ha scritto nel messaggio
news:lEgLb.1476$rt.397@newssvr29.news.prodigy.com...
I'm trying to determine if anyone makes a large/fast static RAM part. 16M
(or more) x 16 bits (or more), 10ns.

I can't afford the address-to-data-out latency of dynamic RAM. There are
ways around this, of course, but SRAM would be so much simpler.

Any ideas?

Thanks,


--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian

To send private email:
0_0_0_0_@pacbell.net
where
"0_0_0_0_" = "martineu"
Click to expand...
 
I see IDT has 1M x 18 ZBT. You are right, 256M is probably far off. I
posted on the outside chance that there was an obscure part out there that I
didn't come across. These days searching on the net can produce so many
links that it is humanly impossible to sort through all of them.


--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian

To send private email:
0_0_0_0_@pacbell.net
where
"0_0_0_0_" = "martineu"


"Uwe Bonnes" <bon@elektron.ikp.physik.tu-darmstadt.de> wrote in message
news:btk8ac$3ur$1@news.tu-darmstadt.de...
Martin Euredjian <0_0_0_0_@pacbell.net> wrote:
: I'm trying to determine if anyone makes a large/fast static RAM part.
16M
: (or more) x 16 bits (or more), 10ns.

: I can't afford the address-to-data-out latency of dynamic RAM. There
are
: ways around this, of course, but SRAM would be so much simpler.

4Mibit Devices are in full production (256kix16). I guess it will take
some
time until 256MiBit are available.

Bye

--
Uwe Bonnes bon@elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------
Click to expand...
 
Martin Euredjian <0_0_0_0_@pacbell.net> wrote:
: I'm trying to determine if anyone makes a large/fast static RAM part. 16M
: (or more) x 16 bits (or more), 10ns.

: I can't afford the address-to-data-out latency of dynamic RAM. There are
: ways around this, of course, but SRAM would be so much simpler.

4Mibit Devices are in full production (256kix16). I guess it will take some
time until 256MiBit are available.

Bye

--
Uwe Bonnes bon@elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------
 
at Thu, 08 Jan 2004 17:42:41 GMT in <lEgLb.1476$rt.397
@newssvr29.news.prodigy.com>, 0_0_0_0_@pacbell.net (Martin Euredjian) wrote
:

I'm trying to determine if anyone makes a large/fast static RAM part. 16M
(or more) x 16 bits (or more), 10ns.

I can't afford the address-to-data-out latency of dynamic RAM. There are
ways around this, of course, but SRAM would be so much simpler.

Any ideas?

Thanks,
Click to expand...
The biggest async parts (the classic SRAM, super-simple interfacing) are
the 16 Mbit parts from Cypress (CY7C1061AV33) and Toshiba (TC55V16100FT).
Sync parts (much more complex interfacing, a bit of a PITA) are the 72 Mbit
Cypress CY7C1482V33. Don't expect anything much larger, especially on the
async parts, anytime soon. These aren't cheap, either, although you gain an
incredible amount, especially with the asynchronous parts, without all the
overhead associated with DRAM. Manufacturer interest will depend on your
order size. How many are you looking to get? Millions? Or onesy-twosey? If
the former, you may get Cypress' attention. If the latter, you'll no doubt
have to be satisfied with what's available.
--
Alex Rast
ad.rast.7@nwnotlink.NOSPAM.com
(remove d., .7, not, and .NOSPAM to reply)
 
Let's remember that the original question was for 256 Mbits = 32 MBytes.
That's several generations of Moore's Law away.
If you figure six transistors per memory bit (plus decoding), that
device would be getting close to 2 Billion transistors. Even the largest
FPGAs have a factor 4 fewer tightly-packed transistors, and they command
a price of >$ 1000.

If you really need RANDOM access in 10 ns, this may be impossible today.
If there is some structure, predictability etc, then you might be able
to do it with DRAMs plus caching (???)
Peter Alfke
==========
Uwe Bonnes wrote:
Alex Rast <ad.rast.7@nwnotlink.nospam.com> wrote:
...
: The biggest async parts (the classic SRAM, super-simple interfacing) are
: the 16 Mbit parts from Cypress (CY7C1061AV33) and Toshiba (TC55V16100FT).
: Sync parts (much more complex interfacing, a bit of a PITA) are the 72 Mbit

At least the Cypress page doesn't sound like instant availability for the
16M asynchronous parts
--
Uwe Bonnes bon@elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------
Click to expand...
 
Alex Rast <ad.rast.7@nwnotlink.nospam.com> wrote:
....
: The biggest async parts (the classic SRAM, super-simple interfacing) are
: the 16 Mbit parts from Cypress (CY7C1061AV33) and Toshiba (TC55V16100FT).
: Sync parts (much more complex interfacing, a bit of a PITA) are the 72 Mbit

At least the Cypress page doesn't sound like instant availability for the
16M asynchronous parts
--
Uwe Bonnes bon@elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------
 
What's you quantity per year.

You could have 16 of the cypress sevices places on a PBC or even die
bonded into a module. And that module may even have a market on it's
own. THis is what the IC mfgs do when they are stepping up to a
larger size device.

gm
 
Lurking around I found 72Mbit parts. These are MCM's, of course.
http://www.gsitechnology.com/72MbBurst.htm


--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian

To send private email:
0_0_0_0_@pacbell.net
where
"0_0_0_0_" = "martineu"




"Peter Alfke" <peter@xilinx.com> wrote in message
news:3FFDF8A2.A649E2FA@xilinx.com...
Let's remember that the original question was for 256 Mbits = 32 MBytes.
That's several generations of Moore's Law away.
If you figure six transistors per memory bit (plus decoding), that
device would be getting close to 2 Billion transistors. Even the largest
FPGAs have a factor 4 fewer tightly-packed transistors, and they command
a price of >$ 1000.

If you really need RANDOM access in 10 ns, this may be impossible today.
If there is some structure, predictability etc, then you might be able
to do it with DRAMs plus caching (???)
Peter Alfke
==========
Uwe Bonnes wrote:

Alex Rast <ad.rast.7@nwnotlink.nospam.com> wrote:
...
: The biggest async parts (the classic SRAM, super-simple interfacing)
are
: the 16 Mbit parts from Cypress (CY7C1061AV33) and Toshiba
(TC55V16100FT).
: Sync parts (much more complex interfacing, a bit of a PITA) are the 72
Mbit

At least the Cypress page doesn't sound like instant availability for
the
16M asynchronous parts
--
Uwe Bonnes bon@elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik Schlossgartenstrasse 9 64289 Darmstadt
--------- Tel. 06151 162516 -------- Fax. 06151 164321 ----------
Click to expand...
 
"Martin Euredjian" <0_0_0_0_@pacbell.net> wrote in message news:<lEgLb.1476$rt.397@newssvr29.news.prodigy.com>...
I'm trying to determine if anyone makes a large/fast static RAM part. 16M
(or more) x 16 bits (or more), 10ns.

I can't afford the address-to-data-out latency of dynamic RAM. There are
ways around this, of course, but SRAM would be so much simpler.

Any ideas?

Thanks,
Click to expand...
As everyone has said, SRAM is not an option at 16.16M for many many
years.

Ordinarily DRAM isn't an option either at 10ns atleast not SDRAM &
DDRAM.

But if you can live with 20ns RAS cycle RLDRAM can do it, at 256M
today. Its 8 way banked and has separate IO for brutal bandwidth at
400MHz by 8,16,32. look at Infineon and Micron sites.

Xilinx has docs on it as well in their memory section. Avnet IIRC has
an eval board with a slightly slower version.

There are also other faster DRAMs from the other guys, but not quite
as fast as I can tell.

IBM makes DRAM that cycles in 5ns or so for internal cache (EET
article), but I don't think there will be any external product at that
speed till the DRAM guys get demand from...

Hope that helps

johnjaksonATusaDOTcom
 
Peter Alfke wrote:

Let's remember that the original question was for 256 Mbits = 32 MBytes.
That's several generations of Moore's Law away.
If you figure six transistors per memory bit (plus decoding), that
device would be getting close to 2 Billion transistors. Even the largest
FPGAs have a factor 4 fewer tightly-packed transistors, and they command
a price of >$ 1000.

If you really need RANDOM access in 10 ns, this may be impossible today.
If there is some structure, predictability etc, then you might be able
to do it with DRAMs plus caching (???)
Click to expand...
You may also be able to throw more pins at the problem, and use multiple
RAM banks to get some more bandwidth (and thus lower nett access times).
Not nice to add pins, but sometimes the design dictates it ...

-jg
 
John,

RLDRAM looks very interesting. A quick price and availability in the Avnet
site does not show these parts as being readily available. Nothing like the
bleeding edge! Maybe a phonecall or two is required. I also saw the Memec
P160 demo module. I have the Memec V2 Microblaze board that takes these
P160's, so it would be very easy to evaluate the solution.

Thanks,


--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Martin Euredjian

To send private email:
0_0_0_0_@pacbell.net
where
"0_0_0_0_" = "martineu"





"john jakson" <johnjakson@yahoo.com> wrote in message
news:adb3971c.0401091904.8a294ef@posting.google.com...
"Martin Euredjian" <0_0_0_0_@pacbell.net> wrote in message
news:<lEgLb.1476$rt.397@newssvr29.news.prodigy.com>...
I'm trying to determine if anyone makes a large/fast static RAM part.
16M
(or more) x 16 bits (or more), 10ns.

I can't afford the address-to-data-out latency of dynamic RAM. There
are
ways around this, of course, but SRAM would be so much simpler.

Any ideas?

Thanks,

As everyone has said, SRAM is not an option at 16.16M for many many
years.

Ordinarily DRAM isn't an option either at 10ns atleast not SDRAM &
DDRAM.

But if you can live with 20ns RAS cycle RLDRAM can do it, at 256M
today. Its 8 way banked and has separate IO for brutal bandwidth at
400MHz by 8,16,32. look at Infineon and Micron sites.

Xilinx has docs on it as well in their memory section. Avnet IIRC has
an eval board with a slightly slower version.

There are also other faster DRAMs from the other guys, but not quite
as fast as I can tell.

IBM makes DRAM that cycles in 5ns or so for internal cache (EET
article), but I don't think there will be any external product at that
speed till the DRAM guys get demand from...

Hope that helps

johnjaksonATusaDOTcom
Click to expand...
 
You must log in or register to reply here.

Welcome to EDABoard.com

Sponsor

Back
Top