Mission critical & low core voltages

Daveb, your question indicates your suspicion that modern ICs ( like FPGAs) offer less reliable system solutions than the older technology. That suspicion is not shared by the (admittedly biased) IC industry, nor by the user community which accepts the newer technologies and higher capabilities gladly (even while they complain about the usual teething problems). You have to put more care (and knowledge) in your pc-board design, you have to pay more attention to signal integrity between ICs, but you get so much more capabilities that you can afford to think about self-checking or redundant designs that increase your reliability. If your suspicion were justified, it would be the kiss of death for progress as we are accustomed to expect... Peter Alfke, Xilinx Applications

Peter, I should have made myself clearer. By older technology I meant 'older' FPGA technology e.g Spartan-II with a 2.5V core voltage. I'm not suspicous of the newer devices' reliability, just keen to see what the consensus is regarding the ever reducing core voltage (i.e. reduced noice margins) & reliability of system implementations.

Dave

Daveb, that's exactly what I was addressing. If you decouple Vccint well, and use the right multi-layer board for good Vcc distribution, there should be no FPGA-internal problem. You can still use 3.3-V I/O, even on the newest devices. We have made sure that the I/O can be 3.3-V capable and tolerant, even though that has increased our chip size and thus the cost. People build extremely reliable complex systems with the modern devices. We used to point to the Mars Rover, which used the then most-modern Virtex devices to control its wheel movement (and other functions). If NASA was willing to use the most modern technology for such a risky and high-visibilty and publicity project, you might also want to do that... Peter Alfke

Thanks Peter. I wonder what devices the Mars Beagle2 Lander used :)

makes no difference if the landing airbag has mechanicaly fail and beagle2 is 6 foot under, embedded into Mars landscape.

Aurash

Well, I think it depends. We are doing electronics for radiation areas I from my experience I can say that old-good devices are doing much bette job than new ones. Specially if we for example take ICs in DIL compare to their counterparts in SOIC package. I don't know what is the internal structure difference, but these DILs can widthstand more of radiation. For FPGAs it's the same thing. Higher integration = more transistors in sq. mm -> less tolerant to single event radiation effects. (of course you have radiation tolerant fpga, but if you compare a cost, you find out that it's better to do that with standard fpgas and change pcb every year :) d.

David,

Our latest Virtex 4 is 600% less likely to be upset by soft errors.

or

As well, its total dose resistance has also increased (improved).

Your intuition is incorrect: the more advanced technologies are improving their ability to withstand the effects of cosmic rays, and radiation.

There may have been intermediate products that were designed poorly (by other manufacturers), but that is no reason to avoid the latest technology when it is properly executed (by everyone else).

Austin

Dave,

If you are concerned about "reliability", then I would assume you are concerned about gate oxide lifetime, hot carrier injection (HCI), negative temperature bias instability (NBTI), electromigration (EM), and so on.

All of these are detailed in our process qualifications which ensure that every device we sell meets the generally accepted commercial (and industrial) lifetimes of better than 20 years.

In fact the latest 90nm triple oxide reliability results are as good, if not better, that the previous technology nodes.

(see page 20)

Austin