Radiation-safe orbits

Henry Spencer wrote:
In article ,
th wrote:

LEO satellites don't really need anything beyond error-correcting memory...
and in an equatorial orbit, probably not even that. (Almost all of the
bit-flips in memory occur during passage through either the South Atlantic
Anomaly or one of the auroral ovals, and an equatorial orbit encounters
neither.)

This is a very optimistic statement! Bit flips may occur in memories
already when you are flying in a passenger aircraft across the poles.
IIRC the shuttle computers get a couple of hundred bit flips per mission
at 28 degrees inclination, even more at ISS orbit inclination.


Please read what I wrote: the auroral ovals (around the magnetic poles)
and the South Atlantic Anomaly (which both a 28deg orbit and the ISS orbit
pass through) are the big hot spots for radiation effects. If you plot
memory errors vs. location on a map, they're very obvious. An equatorial
orbit *doesn't pass through those hot spots*.

True, but have a look at http://www.ngdc.noaa.gov/stp/GOES/sts.pdf where
you can see the location of the shuttle computer upsets. The SAA is
obviously the most contributing area but an equatorial orbit is not free
from bit flips, which means that you need error correction mechanisms
unless you can tolerate frequent computer failures or restarts (provided
that you detect the error before something drastic occurs)

Now, if I were building a satellite for an equatorial LEO, I probably
*would* put error-correcting memory in it, just on general principles.
But one might well be able to get away without it.

You could get away with it but how to treat the tens or hundreds of
upsets per year (depends on memory technology) that will occur? Software
mechanisms are expensive to design and validate and if you are using for
instance a COTS operating system with no source code availability the
behaviour in case of error is totally unpredictable.

The severity of the radiation problem in space is much exaggerated. The
MOST astronomy satellite, in about the worst possible LEO -- relatively
high and polar -- has error-correcting memory, and some care was taken in
the design of its electronics, but it has no rad-hard parts. (The project
couldn't afford them.) It's coming up on two years in orbit, and the only
radiation effect yet visible is some drift in the calibration of some
sensors.

The severity of the total dose radiation problem is a heritage from old
days and is quite exaggerated today since most commercial technology has
inherently good total dose radiation tolerance due shrinking geometries
and processes. 10 - 20 years ago with chips having geometries of several
microns you were lucky to find a chip that tolerated more than 5 krad.
Today many modern complex chips like processors easily tolerate 50 krad
or more but instead issues like latch-up, SEU and SET are dominating.
Thus the market for dedicated rad hard parts has decreased as you can
find commercial parts that tolerate the rather comfortable environment
in LEO (1000 km altitude and 1 krad per year total accumulated dose
with typical shielding).

--
th

Rick Jones wrote:
In sci.space.science th
wrote:

This is a very optimistic statement! Bit flips
may occur in memories already when you are flying
in a passenger aircraft across the poles.

Or anywhere else; part of the bit flipping is from
alpha particle radiation derived from the material
used to construct the computer integrated circuits,
so is internal to the computer and not location
dependent.

Heck, terrestrial computers were/are getting bit
flips and whatnot in their caches and such even in
places like Denver, necessitating ECC on the
caches and the like.

Funny you should mention Denver...

With enough bits sitting there waiting to be
flipped...?

Dr. Seymore Cray famously got caught by this with
his Cray I computer serial number 1, which he
decided to build without error correcting memory and
circuits. The result was a machine that failed in a
statistical manner that prevented the length of
calculations desired for the first supercomputer,
[two hours mean time to failure sticks in my mind,
but this was 30 years or so ago, I may not remember
correctly]. The end result was that every purchaser
of [$20 million] Cray I supercomputers first
received serial number 1, on which to develop code,
while the eventual one they got to keep for running
that code was constructed and delivered later, with
error correcting circuitry now very much part of the
construction.

I happened to get to _sit on_ Cray 1 #1 when it
was doing its initial delivery stint at the
Boulder Colorado National Center for Atmospheric
Research (NCAR), just down the road from Denver.
It was built as two nested truncated cones, with
seat cushions atop the short outer cone to make
it double as a piece of furniture, perhaps to
seem less threatening.

The problem was exactly your "enough bits waiting to
be flipped". The smaller numbers in smaller, earlier
computers had made the issue seem ignorable to Dr.
Cray, but at the size of Cray I #1, that was no
longer true.

Today's massively larger memories just make the
problem even worse, the need for error correcting
code and circuitry more pronounced. After Cray I #1,
probably no one would build a computer without such
technology today. The difference for use in orbit is
just that you'd want more of it for orbits where
exterior radiation now becomes a greater issue.

FWIW, and pardon my anecdote.

xanthian.

What's interesting is that SEUs aren't strictly a space-based problem and more
and more manufacturers are measuring the effects of radiation on their ground-
based devices and doing things about it. Modern central processing units use
internal parity and EDAC, for example, in many places, not just on the
external memory bus.

Here are the slides of a presentation that I found quite interesting. I
haven't been able to locate the url for the white paper on it. I have a copy
but it's copyrighted.

"The NSEU Sensitivity of Static Latch Based FPGAs and Flash
Storage Devices"
Joe Fabula, Xilinx
2004 MAPLD International Conference
September 8-10, 2004, Washington, D.C.

http://klabs.org/mapld04/presentatio...9_fabula_s.ppt

I agree with most of the message below, total dose is not as much of an issue,
at least for digital circuitry, with effects such as SET becoming more
prominent. But it seems as though single event latchup is on the decline with
the shrinking voltages, seeing fewer problems there.


th wrote:

Henry Spencer wrote:
In article , th
wrote:

LEO satellites don't really need anything beyond error-correcting
memory... and in an equatorial orbit, probably not even that. (Almost
all of the bit-flips in memory occur during passage through either the
South Atlantic Anomaly or one of the auroral ovals, and an equatorial
orbit encounters neither.)

This is a very optimistic statement! Bit flips may occur in memories
already when you are flying in a passenger aircraft across the poles.
IIRC the shuttle computers get a couple of hundred bit flips per mission
at 28 degrees inclination, even more at ISS orbit inclination.


Please read what I wrote: the auroral ovals (around the magnetic poles)
and the South Atlantic Anomaly (which both a 28deg orbit and the ISS
orbit pass through) are the big hot spots for radiation effects. If you
plot memory errors vs. location on a map, they're very obvious. An
equatorial orbit *doesn't pass through those hot spots*.

True, but have a look at http://www.ngdc.noaa.gov/stp/GOES/sts.pdf where
you can see the location of the shuttle computer upsets. The SAA is
obviously the most contributing area but an equatorial orbit is not free
from bit flips, which means that you need error correction mechanisms
unless you can tolerate frequent computer failures or restarts (provided
that you detect the error before something drastic occurs)

Now, if I were building a satellite for an equatorial LEO, I probably
*would* put error-correcting memory in it, just on general principles.
But one might well be able to get away without it.

You could get away with it but how to treat the tens or hundreds of
upsets per year (depends on memory technology) that will occur? Software
mechanisms are expensive to design and validate and if you are using for
instance a COTS operating system with no source code availability the
behaviour in case of error is totally unpredictable.

The severity of the radiation problem in space is much exaggerated. The
MOST astronomy satellite, in about the worst possible LEO -- relatively
high and polar -- has error-correcting memory, and some care was taken
in the design of its electronics, but it has no rad-hard parts. (The
project couldn't afford them.) It's coming up on two years in orbit,
and the only radiation effect yet visible is some drift in the
calibration of some sensors.

The severity of the total dose radiation problem is a heritage from old
days and is quite exaggerated today since most commercial technology has
inherently good total dose radiation tolerance due shrinking geometries
and processes. 10 - 20 years ago with chips having geometries of several
microns you were lucky to find a chip that tolerated more than 5 krad.
Today many modern complex chips like processors easily tolerate 50 krad
or more but instead issues like latch-up, SEU and SET are dominating.
Thus the market for dedicated rad hard parts has decreased as you can
find commercial parts that tolerate the rather comfortable environment
in LEO (1000 km altitude and 1 krad per year total accumulated dose
with typical shielding).




--
rk, Just an OldEngineer
"Engineers abhor extrapolation"
-- Ken Iliff, from _Runway to Orbit_, 2004