PDA

View Full Version : multi-fault circuits



safemode
02-04-2004, 07:06 PM
I was on google for quite a while today trying to find information on multi-fault circuits and such but kept coming up with basically the same few pdf's of papers that simply mentioned multi-fault. There has to be some specific info on multi-fault circuits, it's kind of hard to believe that nobody's been able to create a multi-fault circuit. Our circuit uses an oscillation to do the job, has anyone else even theorized of this method or are there other deadends people have been trying to use? These circuits seem like they would have real value with companies like intel and ibm and amd and such who have huge numbers of transisters and require very good error detection to not corrupt data, has there been any sort of notice by these companies to this project? Is the multi-fault circuit we have discovered the first of even it's size to be made? If so I think a little post to sites like slashdot and newsforge or others could help bring in more computers to the project since that would be a major accomplishment.

michaelgarvie
02-05-2004, 05:58 AM
You will sometimes find these circuits refered to as fail-safe. They are usually very bulky and require analog components (big silicon) as well as full duplication of mission logic. All these do however is make sure that if something fails the mission system goes to a safe state such as a Red Light or a Train Stopping. They also normally don't achieve full diagnosis under multiple faults.

The circuits evolved here are purely at the digital gate level which will make them small. Also they might be able to avoid full duplication making them even smaller. And also they may be able to achieve full diagnosis. The difference is for a medical equipment these would not only put themselves in a safe state but have a flashing light informing you that its doing it.

ABS systems for cars use full duplication sometimes and complex self-checking microprocessors in others. The former is susceptible to stuck-at faults in the comparison logic and other multiple faults in both copies. The latter has a fault latency which the evolved circuits here would not have.

The first circuit which was evolved here under multiple faults, the adder had:
0 fault latency
0 fault time overhead
100% fault coverage
100% redundancy.
This kind of quality is unheard of in conventional design. The last figure rarely goes below 80% even when the first three are very relaxed. So hopefully this scales up!

Hope this helps.

safemode
02-06-2004, 02:52 AM
i think it will scale up, it would be great if we had more people to expedite things but hopefully this goal is allowed to continue to completion no matter how long it ends up taking. :)

michaelgarvie
02-06-2004, 06:41 AM
The good thing is this field is becoming bigger and bigger. As more and more mission critical tasks are automated: medical equipment like pacemakers or heart monitoring, transport systems like high speed trains, aeroplanes, traffic lights, ABS brakes; apart from the traditional needs of space missions and nuclear power stations. Wherever there is human life or big economic loss at risk these circuits will be needed.

The circuits produced by this projects are truly better than those of conventional design so would lead to safer controllers in all these applications and more saving lives and money. I think if people were aware of this they might join the project.

safemode
02-06-2004, 10:22 AM
what happened ? the circuit changed?

safemode
02-06-2004, 10:43 AM
that was like 2+ days of circuit evolution gone to waste. Assumption A sounds like it puts too big of a handicap on the circuit to be very useful as a "multi-fault" circuit. Kinda sounds like a single fault circuit having one fault after the other.

michaelgarvie
02-07-2004, 05:58 AM
Nothing was wasted. The previous run (no Ass. A) has been saved and will be continued in the future. I thought it wise to start with the easier problem.

Both with and without A are useful. With A they're easier. Without A they're alot harder but alot stricter and potentially more useful in cases where faults could come in pairs or bursts, or where the circuit does not go through all its input vectors regularly.

By the way, the last circuit in the Hall Of Fame does not assume A, so it is possible; at least for circuits of size 2. Hopefully it will be possible for larger circuits, and that would be a real breakthrough.

For now, though I think its wise to concentrate on assuming A (which everyone does in the self-test community) and improve on their results in terms of fault coverage and gate count. I'm confident we'll achieve this for a few benchmark circuits using oscillation error signalling and dual rail signalling.

Once these benchmarks are done, we can move on to dropping assumption A and there we'll need as much processing power as we can get!

safemode
02-09-2004, 10:02 AM
ok, i sorta got the assumption A thing, but now we're steering evolution toward specific answers and that no longer makes it evolution. Steering toward specific solutions seems to me to be going a bit too far. We killed the 6 gate answer, on to 10 and the faster we're out of assumption A and back to the true multi-fault circuits the better. I think it's best to leave how the circuit accomplishes what it needs to do to evolution, that is the purpose of the project and what makes it so interesting.

michaelgarvie
02-10-2004, 04:46 AM
Think of it as an easier version of the problem, which is the one that is solved in industry. It would be good to make sure that evolution is better at this simpler version before moving to the hard one! Trust me, baby steps are good.