FPGA Central

All,

http://tinyurl.com/yd7p3ah

Above is a discussion of how Xilinx will now list the contribution of
alpha particles emitted by the packaging materials in their latest
generation of FPGA devices as a contributor to soft errors.

Obviously, when something becomes significant, it has to be taken into
account!

All other manufacturers with-hold this information, as it is VERY BAD,
and the alpha contribution to upsets is from 1 to 10 times as bad as
the atmospheric neutron contribution in the latest technology nodes
(45/40).

In Virtex 6, and Spartan 6, not only is the intrinsic soft failure
rate low, it is now lower than that of ASIC and ASSP devices. We
crossed over (were equal) at 65nm.

This rate of increase in soft error rate was predicted by Robert
Bauman of TI in 2005, and it now has come to pass ...

Austin

On Nov 13, 12:25*pm, austin <aus...@xilinx.com> wrote:
> All,
>
> forums.xilinx.com/t5/PL...
>
> Above is a discussion of how Xilinx will now list the contribution of
> alpha particles emitted by the packaging materials in their latest
> generation of FPGA devices as a contributor to soft errors.
>
> Obviously, when something becomes significant, it has to be taken into
> account!
>
> All other manufacturers with-hold this information, as it is VERY BAD,
> and the alpha contribution to upsets is from 1 to 10 times as bad as
> the atmospheric neutron contribution in the latest technology nodes
> (45/40).
>
> In Virtex 6, and Spartan 6, not only is the intrinsic soft failure
> rate low, it is now lower than that of ASIC and ASSP devices. *We
> crossed over (were equal) at 65nm.
>
> This rate of increase in soft error rate was predicted by Robert
> Bauman of TI in 2005, and it now has come to pass ...
>
> Austin

I seem to recall that in the very early days of dynamic RAM,
it was discovered that ceramic packaged parts had more soft
errors than plastic parts. This led to the discovery of
package-generated alpha particles. Are ceramics still used in
the new packages, or are some other materials emitting
alpha particles?

Regards,
Gabor

Gabor,

Those ceramic packages had been contaminated in a bath of nitric acid
that had been used to dissolve some lead, so it bathed them in the
actinides, and plutonium decay isotopes...famous story.

If manufactured from low, or ultra low alpha materials, ceramics can
be as good as anything else.

Basically, if it ever got close to lead, you have 4.5 billion years of
decaying uranium and plutonium to deal with.

If people smoke cigarettes, you have Polonium 210 in the fertilizer
used, so the smoke is radioactive (one reason why people get cancer
from smoking). Get that smoke into the chips, and you are in
trouble...upset rates go up by a factor of ten (I have measured it).

So, today, we buy laser separated lead for bumping (ultra low alpha
lead), and epoxies, mold compounds, etc all of the highest purity.

So does (most) everyone else who cares about alpha particles. Tested
before the material is shipped, tested again before we use it.
Assembled in a line that has no non-pure materials, only on their own
dedicated machines, never used for a "regular" material assembly
lot....and no smoking.

There are lots of chips out there where the manufacturer doesn't care
how they get manufactured/assembled, as the application has such buggy
software, that soft errors are the least of their problems.

The recent paper by google is pretty scary, and typical of this market
(PC's).

http://www.cs.toronto.edu/~bianca/pa...gmetrics09.pdf

So, bad signal integrity, poor design, is thousands of times more
errors than alpha particles....but, maybe some of these are alphas?

Regardless, there are those that do care, and we do spend a lot of
effort to keep the bad isotopes at bay, and out of the process, but it
is impossible to remove all of them altogether.

As is typical of alpha contamination, NO ONE will confess to having
the problem, and they bury their dead quickly, and quietly.

But, the stories leak out: an ASSP this year (recall), an ASIC
packaged wrongly (another recall), a 'bad' DRAM chip (another very
quiet wholesale recall).

Like I say, we "been there, done that" and have no intention of having
that happen ever again.

Honesty (still) is the best policy.

Oh, and verify, test, qualify.

Austin

austin wrote:

> Those ceramic packages had been contaminated in a bath of nitric acid
> that had been used to dissolve some lead, so it bathed them in the
> actinides, and plutonium decay isotopes...famous story.

And there were white ceramic packages that let in light.
I once heard a story of product that worked on the
bench, but not in the box.
For rev 1.0 the workaround was a light bulb inside the box
near the custom chip in the white ceramic package.

-- Mike Treseler

In article <[email protected]>,
austin <[email protected]> wrote:
>So, today, we buy laser separated lead for bumping (ultra low alpha
>lead), and epoxies, mold compounds, etc all of the highest purity.

Cool, I hadn't realised that was a commercial product.

http://www.osti.gov/bridge/servlets/...ble/231307.pdf
is a reprint of a (rather content-free) thirteen-year-old LLNL paper
about setting up to make the stuff.

>The recent paper by google is pretty scary, and typical of this market
>(PC's).
>
>http://www.cs.toronto.edu/~bianca/pa...gmetrics09.pdf

That very explicitly says that hard errors are more common than soft
errors; I've certainly had, of the couple of dozen RAM modules that
I've bought, one on which a sense amp died and every 4096th bit
started reading back as zero. Wish I'd figured out that this was what
was happening before defragmenting a disc on that system!

The big recent packaging embarrassment is nVidia's inadequate bumping
in very hot graphics processors, which seems to have cost them a few
hundred million dollars and a lot of goodwill with Apple.

Tom

Mike Treseler <[email protected]> wrote:
(snip)

> And there were white ceramic packages that let in light.
> I once heard a story of product that worked on the
> bench, but not in the box.
> For rev 1.0 the workaround was a light bulb inside the box
> near the custom chip in the white ceramic package.

One I heard some years ago (2708 days) was an EPROM that worked
in the light, not in the dark. It was part of a movie camera
controller that had to work in the dark. Debugging in the dark
can be hard, though.

-- glen

On Nov 14, 1:38*pm, glen herrmannsfeldt <g...@ugcs.caltech.edu> wrote:
> Mike Treseler <mtrese...@gmail.com> wrote:
>
> (snip)
>
> > And there were white ceramic packages that let in light.
> > I once heard a story of product that worked on the
> > bench, but not in the box.
> > For rev 1.0 the workaround was a light bulb inside the box
> > near the custom chip in the white ceramic package.
>
> One I heard some years ago (2708 days) was an EPROM that worked
> in the light, not in the dark. *It was part of a movie camera
> controller that had to work in the dark. *Debugging in the dark
> can be hard, though.
>
> -- glen

Getting a bit further off-topic, at my first job we made a LORAN
receiver with planar-gas-discharge display (those bright orange
seven- or fourteen-segment displays you probably saw on
pinball machines). The first winter after product release we
started getting calls from the field that the display would not
come on on dark, cold nights. We finally reproduced the
problem in the screen-room with all lights off. The instant the
fluorescent lamps were turned on, the displays started up.

Apparently the 900V or so was not enough to get the lamp
started without some help from external radiation. The
solution came from the PGD manufacturer: add a bit of
radioactive Kr-85 to the gas in the tubes.

Regards,
Gabor

On Nov 15, 1:16*pm, Gabor <ga...@alacron.com> wrote:
> On Nov 14, 1:38*pm, glen herrmannsfeldt <g...@ugcs.caltech.edu> wrote:
>
>
>
> > Mike Treseler <mtrese...@gmail.com> wrote:
>
> > (snip)
>
> > > And there were white ceramic packages that let in light.
> > > I once heard a story of product that worked on the
> > > bench, but not in the box.
> > > For rev 1.0 the workaround was a light bulb inside the box
> > > near the custom chip in the white ceramic package.
>
> > One I heard some years ago (2708 days) was an EPROM that worked
> > in the light, not in the dark. *It was part of a movie camera
> > controller that had to work in the dark. *Debugging in the dark
> > can be hard, though.
>
> > -- glen
>
> Getting a bit further off-topic, at my first job we made a LORAN
> receiver with planar-gas-discharge display (those bright orange
> seven- or fourteen-segment displays you probably saw on
> pinball machines). *The first winter after product release we
> started getting calls from the field that the display would not
> come on on dark, cold nights. *We finally reproduced the
> problem in the screen-room with all lights off. *The instant the
> fluorescent lamps were turned on, the displays started up.
>
> Apparently the 900V or so was not enough to get the lamp
> started without some help from external radiation. *The
> solution came from the PGD manufacturer: *add a bit of
> radioactive Kr-85 to the gas in the tubes.
>
> Regards,
> Gabor

Oops... that should have been 90V, not 900V...