Sloppy MicroChips: Can a fair comparison be made between biological and silicon entropy?
June 5, 2012 9:05 AM Subscribe
Was reading about microchips that are designed to allow a few mistakes (known as 'Sloppy Chips'), and pondering equivalent kinds of 'coding' errors and entropy in biological systems. Can a fair comparison be made between the two?
OK, to setup my question I probably need to run through my (basic) understanding of biological vs silicon entropy...
In the transistor, error is a bad thing (in getting the required job done as efficiently and cheaply as possible), metered by parity bits that come as standard in every packet of data transmitted. But, in biological systems error is not necessarily bad. Most copying errors are filtered out, but some propogate and some of those might become beneficial to the organism (in thermodynamics sometimes known as "autonomy producing equivocations").
Relating to the article about 'sloppy chips', how does entropy and energy efficiency factor into this? For the silicon chip efficiency leads to heat (a problem), for the string of DNA efficiency leads to fewer mutations, and thus less change within populations, and thus, inevitably, less capacity for organisms to diversify and react to their environments - leading to no evolution, no change, no good. Slightly less efficiency is good for biology, and, it seems, good for some kinds of calculations and computer processes.
What work has been done on these connections I draw between the biological and the silicon?
I'm worried that my analogy is limited, based as it is on a paradigm for living systems that too closely mirrors the digital systems we have built. Can DNA and binary parity bit transistors be understood on their own terms, without resorting to using the other as a metaphor to understanding?
Where do the boundaries lie in comparing the two?
OK, to setup my question I probably need to run through my (basic) understanding of biological vs silicon entropy...
In the transistor, error is a bad thing (in getting the required job done as efficiently and cheaply as possible), metered by parity bits that come as standard in every packet of data transmitted. But, in biological systems error is not necessarily bad. Most copying errors are filtered out, but some propogate and some of those might become beneficial to the organism (in thermodynamics sometimes known as "autonomy producing equivocations").
Relating to the article about 'sloppy chips', how does entropy and energy efficiency factor into this? For the silicon chip efficiency leads to heat (a problem), for the string of DNA efficiency leads to fewer mutations, and thus less change within populations, and thus, inevitably, less capacity for organisms to diversify and react to their environments - leading to no evolution, no change, no good. Slightly less efficiency is good for biology, and, it seems, good for some kinds of calculations and computer processes.
What work has been done on these connections I draw between the biological and the silicon?
I'm worried that my analogy is limited, based as it is on a paradigm for living systems that too closely mirrors the digital systems we have built. Can DNA and binary parity bit transistors be understood on their own terms, without resorting to using the other as a metaphor to understanding?
Where do the boundaries lie in comparing the two?
Best answer: I don't think the two are analogous at all; they're vaguely similar at best.
Errors in DNA are generally bad for the individual organism; they're only beneficial in the sense that the "bad" ones get filtered out by making those individuals less able to reproduce, with the more or less accidental result of leaving only the "good" ones to survive over the long term.
Sloppy chips appear to be designed to deliberately allow small errors only in areas that won't be significant in whatever the chip is trying to calculate, in exchange for gains in efficiency (in the sense of requiring less work and therefore less heat). A close equivalent to this would be lossy file compression (where we allow unrecoverable but generally unnoticeable imperfections in a compressed image or audio file, in exchange for a greater compression ratio.)
The two just don't map to one another at all, other than that they both involve errors.
An error in a "sloppy chip" might have an equal chance of being accidentally "good" as an error in DNA, but there's no descendant-based filtering process as in evolution to keep the "good" and filter out the "bad". These aren't genetic algorithm style errors, which have the opportunity to improve [future generations of] the algorithm; the best case is only that the sloppy-but-inexpensive algorithm accidentally produces the same result that a perfect-but-expensive algorithm would have done.
(Note also that you're conflating two competing definitions of "efficiency" -- one in the sense of "requiring less work," the other in the sense of "being less error-prone"; the two are basically opposites. The fact that you're using them interchangeably may be contributing to the confusion here.)
TL;DR: the analogy you're looking for is with genetic algorithms (which are deliberately modeled on biology), not "sloppy chips".
posted by ook at 10:46 AM on June 5, 2012
Errors in DNA are generally bad for the individual organism; they're only beneficial in the sense that the "bad" ones get filtered out by making those individuals less able to reproduce, with the more or less accidental result of leaving only the "good" ones to survive over the long term.
Sloppy chips appear to be designed to deliberately allow small errors only in areas that won't be significant in whatever the chip is trying to calculate, in exchange for gains in efficiency (in the sense of requiring less work and therefore less heat). A close equivalent to this would be lossy file compression (where we allow unrecoverable but generally unnoticeable imperfections in a compressed image or audio file, in exchange for a greater compression ratio.)
The two just don't map to one another at all, other than that they both involve errors.
An error in a "sloppy chip" might have an equal chance of being accidentally "good" as an error in DNA, but there's no descendant-based filtering process as in evolution to keep the "good" and filter out the "bad". These aren't genetic algorithm style errors, which have the opportunity to improve [future generations of] the algorithm; the best case is only that the sloppy-but-inexpensive algorithm accidentally produces the same result that a perfect-but-expensive algorithm would have done.
(Note also that you're conflating two competing definitions of "efficiency" -- one in the sense of "requiring less work," the other in the sense of "being less error-prone"; the two are basically opposites. The fact that you're using them interchangeably may be contributing to the confusion here.)
TL;DR: the analogy you're looking for is with genetic algorithms (which are deliberately modeled on biology), not "sloppy chips".
posted by ook at 10:46 AM on June 5, 2012
Genetic variation heads off predators, parasites and unforeseeable environmental changes. Chips are engineered with errors to run with less energy. These subjects are apples and oranges.
posted by Blazecock Pileon at 11:59 AM on June 5, 2012
posted by Blazecock Pileon at 11:59 AM on June 5, 2012
Best answer: Errors in translation (going from RNA-->protein) don't result in mutation, but perhaps looking at the idea of wobble base pairing would be interesting to you and the analogy you're trying to make.
Simplied version of what I remember from mol gen class: watson-crick base-pairing is only needed for the first two nucleotides of the codon between the tRNA (esterified to the amino acid the codon calls for) and the mRNA (the transcript coding for the amino acid sequence of the protein). This is faster than if all three nucleotides had good W-C base pairing. (See table 1 on this page; note the lower Eb (activation energy) for the non CG or AU pairs--lower Eb corresponds to faster kinetics.)
The lower fidelity than if there were 100% W-C bping is "acceptable" because there is degeneracy in the genetic code; this redundancy in the nucleotide-to-amino acid code (while it can be at any position, often it's with just a difference in the last nucleotide). Also, errors in translation (or transcription) don't result in mutation like an error in DNA replication would. But without this "sloppiness" on the last nucleotide during translation, it would take far longer for translation to occur.
posted by neda at 4:13 PM on June 5, 2012
Simplied version of what I remember from mol gen class: watson-crick base-pairing is only needed for the first two nucleotides of the codon between the tRNA (esterified to the amino acid the codon calls for) and the mRNA (the transcript coding for the amino acid sequence of the protein). This is faster than if all three nucleotides had good W-C base pairing. (See table 1 on this page; note the lower Eb (activation energy) for the non CG or AU pairs--lower Eb corresponds to faster kinetics.)
The lower fidelity than if there were 100% W-C bping is "acceptable" because there is degeneracy in the genetic code; this redundancy in the nucleotide-to-amino acid code (while it can be at any position, often it's with just a difference in the last nucleotide). Also, errors in translation (or transcription) don't result in mutation like an error in DNA replication would. But without this "sloppiness" on the last nucleotide during translation, it would take far longer for translation to occur.
posted by neda at 4:13 PM on June 5, 2012
This thread is closed to new comments.
posted by IAmBroom at 10:27 AM on June 5, 2012