SO14 package at 154C/W? Really?

[Joerg]

Hal Murray wrote:

No, but I have seen too many BGA failures. Not in my designs because
(with one exception) I never used BGA.

Were those recent or back in the startup days?

I thought reliability was one of the reasons for shifting to BGAs.
Consider the alternatives if you have a lot of pins.

Last case was in 2007.

Would you please ask your assembly people what they think about BGAs.

I am self-employed and most of my clients farm out assembly.

I'd expect they work fine after they get the process debugged. There
are a lot of them in use these days.

Sometimes you've got no choice, like where there are 400+ contacts. But
I try to stay away from such monster chips in my designs. I find that
gull-wing contacts take flexing and G-force in the most graceful manner.

I don't know if you've worked in hi-rel design: System goes onto a
launch pad, lifted x feet, dropped, crashes onto concret surface, lifted
x feet again, and so on. Once in a while someone pours on another layer
of concrete because the previous surface has hammered itself farther
into the ground. After a while they protrude into the soil like stalactites.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

 

[th]

Joerg wrote:

th wrote:
Joerg wrote:
th wrote:
Joerg wrote:
Eeyore wrote:

Joerg wrote:

Eeyore wrote:
Joerg wrote:

So here I am, needing more than the usual 32V opamp. Chose ye olde
MC33174 because it can take 44V but was surprised when I read
that the
usual SO14 package is only rated at 154C/W. Since I am burning
500mW
that would become too toasty, ouch. Or I'll have to idle two
amps each
in there and double the number of chips.

So, is 500mW in a SO14 really too much? What do thee say?
NE5532 ? That's rated at +/- 22V IIRC. I think I've seen them in
SMT in
recent years.

Large input bias current IIRC though.
Max supply current is 16mA. At 40V VCC it'll almost unsolder
itself

I've NEVER known one take 16mA. I use 8mA as a guideline when
estimating PSU
requirements (TI typical 352mW). Typical from Fairchild is 6mA
i.e. 264 mW.


Yes, sure, but this is a medical design. There you must always do a
hazard analysis (FMEA and all that) and assume datasheet limits. If
no limits are given you cannot use the part.


Don't you mix up hazard analysis with worst case analysis here? If
the data sheet says 16 mA and a part consumes only 8 mA over its
operating temperature range, it will not suddenly start consuming 16
mA just because the data sheet says so. Depending on how many units
you are going to manufacture you can always buy a lot of parts and
screen out the ones exceeding your requirements, let's say 10 mA.
Now use the measured values and add some margin due to ageing
effects in your WCA.


No cherry picking in this business.


It is not cherry picking if you write a proper specification with your
specific values and buy the parts via an independent test house. They
typically charge you a few dollars per device in test costs plus a
fixed sum for the test setup. Could be worth looking into that
alternative depending on your volumes.


"a few Dollars per device" would be the killer here. Medical might be a
lucrative market but not that lucrative ;-)


This is the same procedure as the microprocessor manufacturers use,
except that they do the screening in-house. There is no physical
difference between a 2 GHz and a 2.4 GHz processor, it just happens
that some wafer lots end up faster than others during manufacturing.


But then they are shipped with different part numbers. In medical you'd
have to also laser-mark the devices. Anyhow, I really never do that but
always look for a solution that works with off-the-shelf part. A
solution which I now have, by using MC33171 amps. Same as the MC33174
but singles instead of quads. Cheap, works.

Nice to hear that you found a solution. But I'm astonished that the

medical authorities are satisfied with with you just using the data
sheet values without any evidence that the manufacturer has really
tested them for the parts delivered to you. In other areas considered
critical for human life, the authorities would never accept using just
data sheet values from the manufacturer but rather require detailed
procurement specifications clearly showing which parameters are tested.

It would be interesting to hear about the rules used for designing
medical equipment (to see if you can rely on these devices if you end up
in a hospital )

--
th

 

[Joerg]

th wrote:

Joerg wrote:
th wrote:
Joerg wrote:
th wrote:
Joerg wrote:
Eeyore wrote:

Joerg wrote:

Eeyore wrote:
Joerg wrote:

So here I am, needing more than the usual 32V opamp. Chose ye
olde
MC33174 because it can take 44V but was surprised when I read
that the
usual SO14 package is only rated at 154C/W. Since I am burning
500mW
that would become too toasty, ouch. Or I'll have to idle two
amps each
in there and double the number of chips.

So, is 500mW in a SO14 really too much? What do thee say?
NE5532 ? That's rated at +/- 22V IIRC. I think I've seen them
in SMT in
recent years.

Large input bias current IIRC though.
Max supply current is 16mA. At 40V VCC it'll almost unsolder
itself

I've NEVER known one take 16mA. I use 8mA as a guideline when
estimating PSU
requirements (TI typical 352mW). Typical from Fairchild is 6mA
i.e. 264 mW.


Yes, sure, but this is a medical design. There you must always do
a hazard analysis (FMEA and all that) and assume datasheet limits.
If no limits are given you cannot use the part.


Don't you mix up hazard analysis with worst case analysis here? If
the data sheet says 16 mA and a part consumes only 8 mA over its
operating temperature range, it will not suddenly start consuming
16 mA just because the data sheet says so. Depending on how many
units you are going to manufacture you can always buy a lot of
parts and screen out the ones exceeding your requirements, let's
say 10 mA. Now use the measured values and add some margin due to
ageing effects in your WCA.


No cherry picking in this business.


It is not cherry picking if you write a proper specification with
your specific values and buy the parts via an independent test house.
They typically charge you a few dollars per device in test costs plus
a fixed sum for the test setup. Could be worth looking into that
alternative depending on your volumes.


"a few Dollars per device" would be the killer here. Medical might be
a lucrative market but not that lucrative ;-)


This is the same procedure as the microprocessor manufacturers use,
except that they do the screening in-house. There is no physical
difference between a 2 GHz and a 2.4 GHz processor, it just happens
that some wafer lots end up faster than others during manufacturing.


But then they are shipped with different part numbers. In medical
you'd have to also laser-mark the devices. Anyhow, I really never do
that but always look for a solution that works with off-the-shelf
part. A solution which I now have, by using MC33171 amps. Same as the
MC33174 but singles instead of quads. Cheap, works.

Nice to hear that you found a solution. But I'm astonished that the
medical authorities are satisfied with with you just using the data
sheet values without any evidence that the manufacturer has really
tested them for the parts delivered to you. In other areas considered
critical for human life, the authorities would never accept using just
data sheet values from the manufacturer but rather require detailed
procurement specifications clearly showing which parameters are tested.

It would be interesting to hear about the rules used for designing
medical equipment (to see if you can rely on these devices if you end up
in a hospital )

This is not a life support application. In life support apps you
basically have to make your own parts. One company even ran their own
foundry (no, I am not kidding).

In (most) non life support apps you can use datasheet values but you
cannot use the typical values. Not the graphs, just the hard min-max
data in the tables, that's it. One reason why I never use LDOs because
they only give a typical stability range graph. That doesn't fly.

At the end of the day the hazard analysis will tell the story. Along the
lines of "If opamp U31 fails what effects will there be, what are the
likely consequences and what will mitigate these effects?"

Whether a mfg tests or not that says nothing about whether that part
will fail on December 23 or not.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

 

[Joerg]

Phil Hobbs wrote:

John Larkin wrote:
On Sat, 6 Sep 2008 23:24:06 +0000 (UTC), Andre Majorel
cheney@halliburton.com> wrote:

On 2008-09-06, John Larkin
jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:

One nice thing about BGA solder joints is that you can't
inspect them, which saves a lot of production time.
I thought it was done, with X rays or something ?

It can be, and was more often when BGAs were new. There are also some
optical things that can peek under the chips and let you see 4 or 5
balls deep, and also look for crud between the rows. We actually have
one of those, and my production people sometimes use it, mostly to see
that their temperature profiles look right, that the solder has flowed
nicely. But once a process is up and boards are flowing down the line,
individual board inspection isn't commonly done, by us or by our
outside contractors. Like I said, we've scored 100% so far.

I think I could program a BGA to inspect its own solder joints, by
measuring pin capacitance. Maybe I'll do that some day.

John


We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way, so
iiuc it's used for sample testing and failure analysis--hopefully the
failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

 

[Joerg]

Phil Hobbs wrote:

John Larkin wrote:
On Sat, 06 Sep 2008 11:08:16 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Eric Smith wrote:
Joerg wrote:
A heatsink glued onto a BGA? Wow, that takes guts.
John Larkin wrote:
Guts? Why? It's soldered to the board in 456 places, which should be
pretty stiff.
Joerg wrote:
Yeah but if that number is reduced to 455 places some grief could
set in
You'd have to be doing something extremely wrong in order for epoxying
a small heat sink to a properly soldered BGA to result in any of the
bonds failing. Are you applying the heat sink with a hammer or
crowbar
or something?

No, but I have seen too many BGA failures. Not in my designs because
(with one exception) I never used BGA.

Dropping the board just once is often enough--the acceleration during
impact is

a = g*(height of fall)/(crumple distance)

which can easily be thousands of times g, especially on hard floors.

Our server blades have to pass a 3000g shock test, which is really
tough if there's no compliant layer (such as paste or liquid metal)
between the HS and the processor. C4 balls are under a lot of static
stress to begin with (although little ones with under 500 connections
are easier than large ones).

Cheers,

Phil Hobbs

I'd suspect that board flex could be a killer in high-shock
situations, or where a lot of force is applied by a heatsink clamp or
some such. Gluing a fairly light pin-fin heat sink to the top of a bga
FPGA doesn't seem very hazardous to me.

No problems so far.


Yours are probably fine (and your customers seem to be technical folk
who probably don't expect badly dinged hardware to still work). Some of
our heatsinks weigh pounds apiece, and that 3000g test is no joke.

Depends on the customers. Remember that ad picture in EE Design or
something where a missile test went wrong, big fire ball, little arrow
points to small black box up in the air, all dinged up? "This is one of
our power supply modules and afterwards it still worked!"

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

 

[Phil Hobbs]

Joerg wrote:

Phil Hobbs wrote:

We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!


What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

 

[John Larkin]

On Sat, 06 Sep 2008 19:28:19 -0500,
hal-usenet@ip-64-139-1-69.sjc.megapath.net (Hal Murray) wrote:

I think I could program a BGA to inspect its own solder joints, by
measuring pin capacitance. Maybe I'll do that some day.

I think that would work for signal pins on an FPGA, maybe even
an ASIC if you had good JTAG.

Not jtag. I was thinking of programming the fpga to put a weak pullup
on all pins. Then pull a pin low and turn it loose and see how long it
takes to go high. An unsoldered pin will have very little capacitance,
so will go high pretty fast. Done right, this could catch all opens
and shorts on all the i/o pins, without the need for jtag gadgets on
the other ends.

Essentially none of the things we connect to fpga's have jtag of their
own.

I don't think it will work with power/ground pins. If, say, one
of many power pins doesn't connect the chip will probably work fine
in normal usage but might go flaky when adjacent pins are doing a lot
of I/O.

Maybe so.

John

 

[Spehro Pefhany]

On Fri, 05 Sep 2008 16:58:23 -0700, John Larkin
<jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:

On Fri, 05 Sep 2008 17:26:36 -0400, Spehro Pefhany
speffSNIP@interlogDOTyou.knowwhat> wrote:

On Fri, 05 Sep 2008 12:48:03 -0700, John Larkin
jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:

On Fri, 05 Sep 2008 10:18:59 -0700, Joerg
notthisjoergsch@removethispacbell.net> wrote:

So here I am, needing more than the usual 32V opamp. Chose ye olde
MC33174 because it can take 44V but was surprised when I read that the
usual SO14 package is only rated at 154C/W. Since I am burning 500mW
that would become too toasty, ouch. Or I'll have to idle two amps each
in there and double the number of chips.

So, is 500mW in a SO14 really too much? What do thee say?

Epoxy a heat sink on top. I have a board right here on my desk that
has a heatsink on an FPGA, and two more on some dual dacs.

Are those commercial heatsinks? They don't look as nice as the rest of
the board. I wonder why they didn't machine square or rectangular
posts rather than round.

Die cast, I think. They are from Cool Innovations, who are up near
you. What I'd really like is anodized colors.

I bought an anodizing kit from Caswell.. hopefully sometime in the
next year I'll get around to trying it out.

Is that LPI 'silk screen' ?

ftp://66.117.156.8/DSC01786.JPG

Nice.

Dunno, but the ref desigs are 60 mils high, and are pretty readable.
That doesn't always happen.


Is 'MS' your layout guy/gal?

Mike. He does our pcb stuff and Autocad/Solidworks stuff. He's good,
so I won't give his last name.

Justifiable!

Incidentally, this is rev A of the board, and it works: no kluges.

Lower-left is where the Xport ethernet brick goes. We press-released
it and got a lot of hits, so you may see it in a mag or something.

Hey, google T346

This is amazing. Google has changed the whole game.

John

Cool.
Best regards,
Spehro Pefhany
--
"it's the network..." "The Journey is the reward"
speff@interlog.com Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog Info for designers: http://www.speff.com

 

[John Larkin]

On Tue, 09 Sep 2008 11:15:04 -0400, Spehro Pefhany
<speffSNIP@interlogDOTyou.knowwhat> wrote:

On Fri, 05 Sep 2008 16:58:23 -0700, John Larkin
jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:

On Fri, 05 Sep 2008 17:26:36 -0400, Spehro Pefhany
speffSNIP@interlogDOTyou.knowwhat> wrote:

On Fri, 05 Sep 2008 12:48:03 -0700, John Larkin
jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:

On Fri, 05 Sep 2008 10:18:59 -0700, Joerg
notthisjoergsch@removethispacbell.net> wrote:

So here I am, needing more than the usual 32V opamp. Chose ye olde
MC33174 because it can take 44V but was surprised when I read that the
usual SO14 package is only rated at 154C/W. Since I am burning 500mW
that would become too toasty, ouch. Or I'll have to idle two amps each
in there and double the number of chips.

So, is 500mW in a SO14 really too much? What do thee say?

Epoxy a heat sink on top. I have a board right here on my desk that
has a heatsink on an FPGA, and two more on some dual dacs.

Are those commercial heatsinks? They don't look as nice as the rest of
the board. I wonder why they didn't machine square or rectangular
posts rather than round.

Die cast, I think. They are from Cool Innovations, who are up near
you. What I'd really like is anodized colors.

I bought an anodizing kit from Caswell.. hopefully sometime in the
next year I'll get around to trying it out.

Is that LPI 'silk screen' ?

ftp://66.117.156.8/DSC01786.JPG

Nice.

Dunno, but the ref desigs are 60 mils high, and are pretty readable.
That doesn't always happen.


Is 'MS' your layout guy/gal?

Mike. He does our pcb stuff and Autocad/Solidworks stuff. He's good,
so I won't give his last name.

Justifiable!


Incidentally, this is rev A of the board, and it works: no kluges.

Lower-left is where the Xport ethernet brick goes. We press-released
it and got a lot of hits, so you may see it in a mag or something.

Hey, google T346

This is amazing. Google has changed the whole game.

John

Cool.
Best regards,
Spehro Pefhany

Yeah, it used to be that "you can't base a company on a product",
which meant that you had to have a grand scheme and a product line
with critical mass to reach a market. Now, one person, with one good
home-brew product, an assembly line in a garage or a big closet, and a
decent web site, can get started and sell and grow. And a small
company, using mostly free press releases and a keyword-loaded web
site, can be as visible in some specialties as Tek or Agilent. The
levelling power of the search engine is astounding.

John

 

[JosephKK]

On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:

We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!


What's your impression so far with respect to BGA versus PQFP and others?


I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

What the heck do you use for a mating connector?

 

[Phil Hobbs]

JosephKK wrote:

On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

What the heck do you use for a mating connector?

A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It's been quite educational working in the packaging research area, even
though I don't personally do any significant amount of electrical
packaging work. You pick up stuff by osmosis and from people's talks.

Cheers,

Phil Hobbs

 

[legg]

On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

What the heck do you use for a mating connector?


A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It sounds like the wrong place, in the system, to interconnect.

RL

 

[John Larkin]

On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

What the heck do you use for a mating connector?


A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It's been quite educational working in the packaging research area, even
though I don't personally do any significant amount of electrical
packaging work. You pick up stuff by osmosis and from people's talks.

Cheers,

Phil Hobbs

Electronics is 50% packaging.

John

 

[Phil Hobbs]

legg wrote:

On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs
What the heck do you use for a mating connector?

A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)


It sounds like the wrong place, in the system, to interconnect.

RL

Nope, just a really big system. The on-board bandwidth is 5 or 10 times
higher--we could use twice that much off-board bandwidth.

Cheers,

Phil Hobbs

 

[legg]

On Wed, 10 Sep 2008 06:51:53 -0700, John Larkin
<jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:

On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
snip
It's been quite educational working in the packaging research area, even
though I don't personally do any significant amount of electrical
packaging work. You pick up stuff by osmosis and from people's talks.

Cheers,

Phil Hobbs

Electronics is 50% packaging.

13.6%, until the sales team get involved.

RL

 

[JosephKK]

On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

What the heck do you use for a mating connector?


A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It's been quite educational working in the packaging research area, even
though I don't personally do any significant amount of electrical
packaging work. You pick up stuff by osmosis and from people's talks.

Cheers,

Phil Hobbs

Either way, it sounds like a good reason to create a ZIF connector
with wiping action. 4736 pins at even 0.05 oz insertion force is no
joke. And move the power to another connector. 2368 signals is a lot
of signals, even if half of them are power (leaving 1184 signals).

 

[JosephKK]

On Wed, 10 Sep 2008 10:43:03 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

legg wrote:
On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs
What the heck do you use for a mating connector?

A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)


It sounds like the wrong place, in the system, to interconnect.

RL

Nope, just a really big system. The on-board bandwidth is 5 or 10 times
higher--we could use twice that much off-board bandwidth.

Cheers,

Phil Hobbs

About the top clock frequency that you can go through a backplane of
useful size is about 500 MHz. The propagation speeds eat you alive,
not to mention skew, especially at wide transaction widths (say 256
bits). I figure they use a multiplicity of busses, at least 1184 but
not more than 2368 signals was it? And going differential gets you
higher clocks but with worse skew problems. Does re-aggregating
several very high speed serial signals (10,000 MHz each) back to a
parallel form do the job often enough to help?

 

[Phil Hobbs]

JosephKK wrote:

On Wed, 10 Sep 2008 10:43:03 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

legg wrote:
On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs
What the heck do you use for a mating connector?

A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It sounds like the wrong place, in the system, to interconnect.

RL
Nope, just a really big system. The on-board bandwidth is 5 or 10 times
higher--we could use twice that much off-board bandwidth.

Cheers,

Phil Hobbs

About the top clock frequency that you can go through a backplane of
useful size is about 500 MHz. The propagation speeds eat you alive,
not to mention skew, especially at wide transaction widths (say 256
bits). I figure they use a multiplicity of busses, at least 1184 but
not more than 2368 signals was it? And going differential gets you
higher clocks but with worse skew problems. Does re-aggregating
several very high speed serial signals (10,000 MHz each) back to a
parallel form do the job often enough to help?

These are (iirc) 5 Gb/s lines, with forward error correction, adaptive
thresholds, preemphasis, decision feedback equalization, yada yada yada.
The whole thing is tweaked right to the eyeballs, but it meets a
5-nines availability spec (5 minutes down time per year).

Cute gizmos, but all that stuff costs power and latency, which is
another reason for going optical. We need to get below 100 uW/(Gb/s)
(100 fJ/bit), which is going to be a challenge.

Cheers,

Phil Hobbs

 

[krw]

In article <8i6hc4ts7a2dhvflhukm1h4dh000fk8rdt@4ax.com>,
quiettechblue@yahoo.com says...

On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs

What the heck do you use for a mating connector?


A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It's been quite educational working in the packaging research area, even
though I don't personally do any significant amount of electrical
packaging work. You pick up stuff by osmosis and from people's talks.

Cheers,

Phil Hobbs

Either way, it sounds like a good reason to create a ZIF connector
with wiping action. 4736 pins at even 0.05 oz insertion force is no
joke. And move the power to another connector. 2368 signals is a lot
of signals, even if half of them are power (leaving 1184 signals).

Moving the power to another connector doesn't help your signal
integrity.

--
Keith

 

[JosephKK]

On Thu, 11 Sep 2008 01:07:32 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Wed, 10 Sep 2008 10:43:03 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

legg wrote:
On Tue, 09 Sep 2008 23:51:27 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

JosephKK wrote:
On Mon, 08 Sep 2008 08:28:30 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

Joerg wrote:
Phil Hobbs wrote:
We actually have a CT scanner for this up in Poughkeepsie. Takes a
pretty long time when there are high atomic weight metals in the way,
so iiuc it's used for sample testing and failure analysis--hopefully
the failures happen during qualification and not in the field!

What's your impression so far with respect to BGA versus PQFP and others?

I don't know--we basically don't use any packages with leads!

I can tell you that 32x148 pin surface mount connectors are hard to use.

Cheers,

Phil Hobbs
What the heck do you use for a mating connector?

A male version surface-mounted on the backplane. That's the standard
backplane connector for the current P- and Z-series machines. Half of
the pins are grounds, so for signals it's only 16x148. There's a _big_
preload put on it before reflow. (Yes, I realize that this is a
ridiculous connector. That's one reason we're trying to move towards
optical interconnection.)

It sounds like the wrong place, in the system, to interconnect.

RL
Nope, just a really big system. The on-board bandwidth is 5 or 10 times
higher--we could use twice that much off-board bandwidth.

Cheers,

Phil Hobbs

About the top clock frequency that you can go through a backplane of
useful size is about 500 MHz. The propagation speeds eat you alive,
not to mention skew, especially at wide transaction widths (say 256
bits). I figure they use a multiplicity of busses, at least 1184 but
not more than 2368 signals was it? And going differential gets you
higher clocks but with worse skew problems. Does re-aggregating
several very high speed serial signals (10,000 MHz each) back to a
parallel form do the job often enough to help?


These are (iirc) 5 Gb/s lines, with forward error correction, adaptive
thresholds, preemphasis, decision feedback equalization, yada yada yada.
The whole thing is tweaked right to the eyeballs, but it meets a
5-nines availability spec (5 minutes down time per year).

Cute gizmos, but all that stuff costs power and latency, which is
another reason for going optical. We need to get below 100 uW/(Gb/s)
(100 fJ/bit), which is going to be a challenge.

Cheers,

Phil Hobbs

Oh yeah, doing anything large scale really fast is a serious alls to
the walls type endeavor. Gets pricey as well.