Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!rex!ukma!seismo!dimacs.rutgers.edu!aramis.rutgers.edu!athos.rutgers.edu!nanotech From: forbis@milton.u.washington.edu (Gary Forbis) Newsgroups: sci.nanotech Subject: Re: But are they safe? Keywords: mutations, control Message-ID: Date: 26 Mar 91 23:10:52 GMT Sender: nanotech@athos.rutgers.edu Organization: University of Washington, Seattle Lines: 45 Approved: nanotech@aramis.rutgers.edu I'm wondering what error rate is being hypothesised here. How many units are people thinking about. Might the codes necessary to keep the probability of undetected error low enough slow down the machines so much as to make them inviable solutions to the problems for which they are being considered? I see that CD are now being advertised with eight times oversampling and this is just for audio applications. The space shuttle uses five flight computers, four run the same program with semaphores to indicate proper functioning and the fifth runs an independently created program and serves as an arbitrator. Even with this at least one flight was delayed in the final seconds because of a program malfunction. I think that in some cases some of the computers were shut off during reentry. The SpaceLab had (I believe) seven synchronized gyros yet it had hardly been placed in orbit before they started to fail. One would think that if there was a way to prevent these errors it would have been taken in these cases. My first home computer was an Altair 8800, my second, a TRS80 mod I. The first had a problem with the cpu the second a memory problem. The cpu failure was pretty straight forward, it had a single bit error in one of the registers. The memory problem wasn't so easy, it had two cells welded together. All of the standard memory tests showed that the memory was good yet some programs failed in strange ways. I found the problem by setting memory each location then testing all of the other locations in the bank. Please note that a any arbitrary number of bits could have been welded together by having a single short or open in the address lines. I don't get the feeling everybody is considering this with open eyes. Assuming that every one alive today had a million nanocomputers working for them this would be about 5*10^15 machines. Is this number too high? Even if most failures are hard failures this is still about 10^19 machine hours per year. Even if each person only has a hundred computers this is still 10^15 machine hours per year. I'm having a hard time believing this many machine hours per year could be achieved without some soft failures. --gary forbis@u.washington.edu [I had an Altair too; the memory didn't even have parity. I imagine that nanocomputers will have oodles and gobs of soft errors; this comes with the territory of sizes and speeds we're talking about, and from working in a domain where quantum effects are sure to raise their ugly heads. :^) That's why it's almost certain that nanocomputers are virtually certain to have a major error-correcting component and that coping with the probabilistic nature of the data will be a central design issue. --JoSH]