UPGRADE YOUR BROWSER

We have detected your current browser version is not the latest one. Xilinx.com uses the latest web technologies to bring you the best online experience possible. Please upgrade to a Xilinx.com supported browser:Chrome, Firefox, Internet Explorer 11, Safari. Thank you!

cancel
Showing results for 
Search instead for 
Did you mean: 
Visitor ane093
Visitor
675 Views
Registered: ‎09-20-2017

Question regarding UG116, Device Reliability Report

The rightmost column in table 1-19 and 1-20 says FIT/Mb (Real-Time Soft Error Rate Per Event).
What does an event mean in this context?

Also, I see that FIT/Mb caused by thermal neutrons and alpha particles are listed in table 1-19 and 1-20, but cannot find similar data for high energy neutrons. Is this available?

Thanks

0 Kudos
3 Replies
Explorer
Explorer
657 Views
Registered: ‎05-08-2018

Re: Question regarding UG116, Device Reliability Report

Well,

 

Although Xilinx asked me to leave March 10 of this (RIF'd after 19 years, 9 months), I created the SEU program, created the remote arrays, and managed the program for 7 years, handing it of to reliability team, who runs it now.

 

The atmospheric neutron flux is from thermal to greater than 1 Gev (highest energy neutrons are infrequent but there is no known upper limit.  By having real parts, taking data, one then knows absolutely what will happen in a real system, anywhere on earth (flux is well known by longitude, latitude, altitude).

 

The highest energy neutron will not create 'more' events.  Physics doesn't work like that.  The neutron slits the silicon atom (spallation),  which produces  heavy ions ranging from protons to silicon itself (no split only recoil).  These secondaries range in energy from 0.5 Mev LET (a measure of total energy deposited:  multiply by 10 to get approximate charge deposited in femto coulombs) to 14.2 Mev LET.  Testing at CERN with 25 GeV neutrons (generally everything known as Hadrons with energy) reveal nothing new, or interesting, confirming my statements above.

 

The neutron that created the initial ions may go on to create more ions if its initial energy is not fully absorbed by the original interaction.  Eventually, one ends up with a neutron at rest (probably outside the device) which then decays to a proton in a few minutes (neutrons by themselves are unstable).

 

Event in real time testing is defined as a bit flip found during readback.  The initial value was known, the readvack sees one difference.  Some neutrons will cause multiple upsets (MCU, MBU), with some being up to 16 or more upsets (of the angle of the secondary is just right).  Atmospheric testing cannot detect SEU from MBU.  Beam testing is used to characterize MBU at design thge interleaving so that  MBU are mostly correctable (MBU does not break the frame ECC).

 

So in the LANSCE testing (neutron beam) data identifies SEU and MBU, so the flux is known (get a neutron count).

 

This is mostly an academic matter (event vs. neutron count) and is unimportant, as in real use, it only matters the rate of failures (FIT), and if that will affect your system, and how you mitigate.

 

 

Tags (1)
Visitor ane093
Visitor
631 Views
Registered: ‎09-20-2017

Re: Question regarding UG116, Device Reliability Report

Thanks for the thorough answer.

But how can FIT be related to both events and time?


I would suppose that:
We would express the probability in of failure, in a time span, given certain intensity of radiation.
or
The probability that a bit flip cause a failure (independent of time?)

 

0 Kudos
Explorer
Explorer
622 Views
Registered: ‎05-08-2018

Re: Question regarding UG116, Device Reliability Report

It is a probability (statistics) process:

 

A FIT/Mb of 6 (latest US+ CRAM) for 200 Mb of CRAM is then 1200 FIT.  So, on average, one gets 1200 upsets in 1 billion hours, or roughly, an upset every 800,000 hours.

 

You need at least 100 upsets to determine the rate to a 95% confidence level (still have +/- 5% your answer is too low, or too high), and your result is within +/- 20 % ofthe actual value.

 

As upsets are perfectly random, you coukd get one, and get another one the next day.  No way to know for any event, all you can say is that over time, and over events, the answer will converge eventually to the predicted value.

 

0 Kudos