Curiosity-driven query: SMT vs thru-hole differences in components

[Sir H C]

We once calculated that the gate oxide thickness for one of the 90nm (maybe 180nm) processes was about 50 atoms thick and the variance was +- 5 atoms. 

[Mark Hammer]

These are lovely and delicate responses folks.  I'm very glad I asked the question.

I'm sensing that pedal-to-pedal variation can be expected to be somewhat smaller in the "future", as a result of more consistent fabrication techniques, and possibly even better access to reliable supplies of components (unlike the circa-1972 "Matthews Method" of who in the greater NYC commutable distance has caps of that value cheap this week ).  At the same time, people have eloquently explained why there are, and probably will continue to be, finite limits to the consistency with which a given pedal is produced.  Not the wider range of possibilities we have come to expect from many discrete-transistor vintage pedals, but still nothing you'd call faithful clonings where pedal 10,000 is absolutely identical to pedal 500.

The other sense I'm getting is that, even with component to component variation, the very compexity of circuits suggests a kind of increasing homogeneity (accidental consistency, really) arising out of what you might call "regression to the mean" in statistical terms.  That is, with so many parts, all the variation within tolerances this way and that even out in the wash.   As RG has pointed out in past, with respect to the Fuzz Face, each part in the Fuzz Face matters a lot because it is such a simple circuit.  When a circuit depends on the joint interaction of literally thousands of transistors, caps, and resistors embedded in silicon wafers, no single part is going to have quite as much impact, or produce quite as much sonic variation.

Is that a fair distillation of the thread so far?

[iaresee]

 As RG has pointed out in past, with respect to the Fuzz Face, each part in the Fuzz Face matters a lot because it is such a simple circuit.  When a circuit depends on the joint interaction of literally thousands of transistors, caps, and resistors embedded in silicon wafers, no single part is going to have quite as much impact, or produce quite as much sonic variation.

Is that a fair distillation of the thread so far?

Absolutely.

[Sir H C]

And you can make designs that are more "on the edge" with more consistent parts, you don't have to design for that worst case.

For instance in ICs often the current used is higher than it "needs" to be to handle worst case processing.

[puretube]

Watchit:
The smaller the components, the closer the connections, the narrower the traces,
the LARGER the capacitances between the nodes...

(audibly heard observations of a high-impedance nerd...)

[Paul Perry (Frostwave)]

When you look at a microphotograph of an IC, it looks like a formal garden, complete with rough edges. VERY rough edges.
And even when stuff is laser trimmed for accuracy, it looks as though someone took a machine gun & blasted away chunks - because that is pretty well exactly what happened, at a very tiny scale.
Making semis has everything in common with printing, and with baking pies. When you magnify the details up, it's always lumpy & uneven, to some degree. Hence the variations.
As for form factor of resistors and 'noise' - there are two kinds of noise, the 'white' noise, and the 1/F or 'flicker' noise which appears at low frequencies, and gets progressively larger as you move down in the far infrasonic range. The latter is associated in mysterious ways with the uniformity of the resistive material and the way in which the electrons move through the material, also with surface impurities.

[reverbie]

Drop an old rotary phone...then drop your new Apple I-Phone.

Which one still works?

[Sir H C]

Drop an old rotary phone...then drop your new Apple I-Phone.

Which one still works?

The iPhone, because I would dive to catch that.

[Ardric]

This has come up before.  Absolute rubbish as far as I'm concerned.

I think it's pretty well established that resistor noise is dependent on at least some factors like chemical composition, physical structure, process details, etc.  One resistor can certainly be more or less noisy than another.  If that's rubbish, I'm pretty confused.

Lots of literature on the topic:
http://www.google.ca/search?&q=film+resistor+excess+noise

Narrowing that down to SMD vs thru-hole isn't very fair or accurate, granted.  That was the ancedote.  And I'll also agree that for the vast majority of applications the difference is negligible.

You are suggesting to measure nosie levels of each and every surface mount resistor to be used in a circuit, based on anecdotes you've read somewhere? [...]

I wasn't, but you gave me a chuckle, thx.  I was suggesting that if low noise is really important, check the mfr specs.  Don't just assume that film=good.  Select on specs, like any other part.  But since it isn't really that important for stompboxes, nevermind... I'm just a nerd who thought it was an interesting topic.

At the same time, people have eloquently explained why there are, and probably will continue to be, finite limits to the consistency with which a given pedal is produced.  Not the wider range of possibilities we have come to expect from many discrete-transistor vintage pedals, but still nothing you'd call faithful clonings where pedal 10,000 is absolutely identical to pedal 500.

Mark, there's a common class of part that can do everything any other part can do, but do it with perfect 100% absolute repeatability.  It's called software!  We'll never reach a point where we can make two fuzz faces perfectly identical, but we WILL reach a point where it's cheaper to build a fuzz-face out of software than hardware.  That's because we'll reach a point where the simulations can be made more faithful to the original than an analog copy could possibly be.  It's inevitable.

The other sense I'm getting is that, even with component to component variation, the very compexity of circuits suggests a kind of increasing homogeneity (accidental consistency, really) arising out of what you might call "regression to the mean" in statistical terms.

I'll argue with this too, wee!  I think when you're laying down thousands of transistors in a chip, you're sensitive to yield issues, and one bad part in the works cannot be repaired.  Therefore you change your habits to favor designs that work ok within whatever tolerances your mfg process gives you, using extra parts in the process.  The extra parts are worth the increase in yield.

You're saying if part A is +ve wrt spec and part B is -ve, then when the circuit snippet is A+B you're golden, the tolerances have canceled out.  When it's A-B you're in the scrap bin.  I'm not sure why the A+B case is more common than A-B, but if it really is, then I'd agree.

[R.G.]

Mark, there's a common class of part that can do everything any other part can do, but do it with perfect 100% absolute repeatability.  It's called software!  We'll never reach a point where we can make two fuzz faces perfectly identical, but we WILL reach a point where it's cheaper to build a fuzz-face out of software than hardware.  That's because we'll reach a point where the simulations can be made more faithful to the original than an analog copy could possibly be.  It's inevitable.

It is inevitable. Just look at the progress we've made.   

An analogy is in order. Solid state reproduction of tubes, whether analog or digital (i.e., by software) is just around the corner - just like it has been for the last 30 years. 

I suspect that software modelling of the Fuzz Face and others is also just around the corner.

We are already at the place where it's cheaper to build a Fuzz Face out of software than hardware. Making it faithful to the original will be tough because (a) it's hard to model all of the interactions on a real-time basis, just as for tubes and (b) you first have to define which original you're going to match with your software version; there is no "the original", only a continuum of similar pedals.

But it is inevitable. Someday.