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

[Mark Hammer]

Is there any a priori reason to expect tighter tolerances in SMT passive components, compared to their thru-hole counterparts?  Conversely, is there any reason to expect bigger tolerances?

I'm just curious because of recent discussions about some pedals that are SMT, and some vintage pedals that are thru-hole and all over the place.  Should we expect that more recent-issue pedals employing SMT assembly and components will be more consistent?  If they have a digital core, should we expect even MORE consistency within any given product than we are accustomed to in traditional thru-hole analog devices?  Is there any a priori reason to expect less noise or stability issues?

I suppose the qualification is that SMT construction permits cramming more "stuff" in the pedal, and even if tolerances are tighter, more stuff offers more degrees of freedom with respect to pedal-to-pedal variation.

So, like I say, it is a curiosity-driven question.

[R.G.]

Good question, Mark.

There is probably nothing about the SMT parts themselves that are wider tolerance that I know of. I suspect it's a matter of manufacturing getting more automated and precise than it is the nature of the parts themselves.

All parts are tighter tolerance than they used to be, even through-hole parts. Of course this is a statement about the date of manufacture and initial tolerance. At one time, a 5% film cap was a hefty price premium and 2% or ACK! 1% caps were unaffordable. These days, 5% caps are essentially the same price as 10%, and 2% is a modest premium. Resistors can hit 5% with no selection or trimming, so 10% parts (if you can even find something labeled 10%) are probably either very old or if newly made are 5% parts relabeled for the role. 1% resistors used to be 5-10x the price of 10% parts. Now 1% resistors bought in quantity are perhaps 1.5-2x the price of a 5% part, due in part to automated trimming.

SMT parts have brought new materials to capacitors to get the capacitance-per-volume and CV product up. I don't think that has much to do with the scatter, as they all benefit from the tighter manufacturing.

I think newly made parts in general will have a tighter range than "vintage" or NOS parts, and not just SMT.

On the sound quality issues: that's a tough one. We have a 100% chance of a good part of the industry yelling from the rooftops that through hole sounds better than SMT in mysterious and unmeasurable ways. That's already going on. The same thing happened in the transition from tubes to solid state in electronics in general, and persists in the hifi tweako and musical amplifier world. There are even some tidbits of fact underlying that last. But whether a carbon film layer on a ceramic substrate sounds different because the substrate is rectangular instead of cylindrical? I doubt it. Thick film sintered resistors? Maybe... I'd want to see something measured first.

But we are certain to see the waters part into people that verbally abuse SMT and those that think it's OK. The abusers will be composed of those that really hear a difference, real or not; those who have a monetary interest because they can't make SMT stuff, or can't service SMT stuff; and those who believe the advertising from the first two groups. The OK-saying groups will be composed of those who either don't hear a difference or think SMT sounds better; and those who have a monetary interest because SMT is cheaper and more reliable than through-hole and it's accessible to them, and their advertising believers. The tough thing in separating out the blizzard of words is that it's very difficult to tell someone who hears a difference from someone with a financial interest.

[ayayay!]

Man, excellent question.  I wondered this very thing myself working on my DL4 this weekend.  I didn't have any SMT resistors so I was using standard 1/4 watts.  I always metered them before installing to make sure I was tight on the ranges, but they still were quite finicky.  (23.3 being to loud, 23.6 being too soft...)  So I was wondering if the 24K's that the mod calls for would have been dead on, or would it still have a volume drop?  I won't know for sure until I try it will the SMTs. 

Is there inherent drift w/ through-hole components that occurs when mixing w/ SMT?  Is there something to be said for the tighter tolerances of SMT or does that truly take away the romance of it all?  I don't have the time or resources to exhaust just those simple questions. 

...but it sure is fun to experiment...   

[Mark Hammer]

I suppose the other thing to balance off is something alluded to by RG: the relative ease with which SMT-based pedals can be serviced, and I suppose modded too.  We've had a preview of this with the many attempts of members to upgrade a Danelectro Fab Echo; not the easiest of tasks.  So, even where there might be no compelling sonic argument to make for use of thru-hole vs SMT, there is a kind of oblique sonic argument to make in that one can more readily make a thru-hole pedal sound the way you want, whether it sounded better than SMT to start with or not.  Of course, realistically, that particular argument only pertains to a small segment of the market.  For the overwhelming majority of cosumers, it's plug and play all the way.

I concur with RG's point that recency of component manufacture is probably more germane than form factor.  SMT just happens to be more recent.  Incidentally, is it just me or are 1/4W resistors coming with narrower gauge (and shorter) leads these days?

[PerroGrande]

I have a couple of theories related to SMT vs. Through Hole -- trying to stay away from the train-o'-mojo along the way...

1) Depending on how you fabricate the board, etc, you *may* end up with more/better ground planing on an SMT board.   This is doubly true if you use one side of a double-sided board as a large ground plane. 

2) There are some components in the SMT world that do not exist in the through-hole world -- or there are variations that exist within the SMT world that do not occur in through-hole world.  There are a number of examples, but one that comes to mind is the three-leg diode package (SOT-23).  This packaging allows not only single, but dual-diode configurations within a single package.  From a thermal stability and component tolerances, this is an ideal situation.  As a teaser, go look at the MMBD4148SE and think of the various applications of this configuration...

Do these translate to *audible* differences?  Who knows...   I could *potentially* see a situation where the aforementioned tolerances and stability could be a benefit, but I can't say for certain that this is the case.

I *like* SMT stuff.  I got into it about 11 months ago (bought some SMT soldering gear as a Christmas present to myself).  Once I got the hang of working with it, I can produce an SMT board almost as fast as a through-hole.  There is less hole-drilling (in some cases, none whatsoever) and it is MUCH easier to fit an SMT board into an enclosure (there is usually only one "shortable" side -- I've not gotten into double-sided SMT).

[iaresee]

Is there any a priori reason to expect tighter tolerances in SMT passive components, compared to their thru-hole counterparts?

Not in my experience. A 5% tolerance SM resistor can and have swung just as much as 5% through hole resistors. The SMT markings, especially on resistors, might mask this as they often exclude handy pieces of information like tolerance because of space limitations. My experience says much of the part information is lost in the manufacturing process as the details get left behind on the reels the parts were attached to before they were soldered. (Easy) field servicing for SMT-based products is usually not a design consideration. Their either getting returned to the factory where the detailed schematics with tolerances help you find a suitable replacement or they're getting tossed when they break.

Conversely, is there any reason to expect bigger tolerances?

On resistors that don't include the tolerance they're supposed to be "standard tolerance" -- whatever that is. I'd bet it's a moving target, getting better every year as R.G. mentioned, because of improvements in manufacturing.

Should we expect that more recent-issue pedals employing SMT assembly and components will be more consistent?

Consistency is a sum of a few different pieces: part tolerance, manufacturing tolerances and QA testing tolerances come to mind. I think 1 and 2 in bulk productions are easier to control with SMT parts. The volume of things we interact with SMT manufacturing now helps support that. The third part is up to the company: if they let shoddy stuff through previously I see no reason moving to SMT would prevent shoddy products from leaving a line going forward.

If they have a digital core, should we expect even MORE consistency within any given product than we are accustomed to in traditional thru-hole analog devices? Is there any a priori reason to expect less noise or stability issues?

An interesting question and I think it depends on the pedal. If most of the work is being done in the digital core then I think you can expect incredibly good consistency; near perfect since the work is being done in the digital domain with deterministic algorithms. If there's A/D, D/A, analog amplification and analog control involved I'd expect the same kind of swings as in traditional analog design.

[Mark Hammer]

Yeah, I expect that 5% SMT resistors ARE 5% tolerance, but that is separate from the actual spread or distribution of values.  A 5% tolerance could mean that 2/3 of observed production values fall within 1% and only a very few fall outside of +/-3%.  Or it could also mean that 2/3 are found within +/-3.5% and only a few fall outside of 4.5%.  Both of those are a 5% spec.  I was just wondering if the distribution was more "leptokurtic" as the stats folks would say (http://en.wikipedia.org/wiki/Kurtosis).  The presumption is based on what RG noted about more recent production techniques, and the possibility for more consistent production.

[Sir H C]

The dangerous ones are where 5% = those that fail the 1% and 2% test so you don't get a gaussian you get two humps away from the desired value.

The best thing about SMT is that the parasitics are so much smaller for high speed stuff.  For audio, I would say same materials = same sound.

[R.G.]

The dangerous ones are where 5% = those that fail the 1% and 2% test so you don't get a gaussian you get two humps away from the desired value.

I'm guessing that doesn't happen any more with resistors, as the testing is probably too expensive to bother. ICs with long experience in manufacture are sometimes tested only at DC to avoid testing expense. Probably the same with transistors; alternately, all testing would be sampled only, no 100% testing.

[iaresee]

Yeah, I expect that 5% SMT resistors ARE 5% tolerance, but that is separate from the actual spread or distribution of values.  A 5% tolerance could mean that 2/3 of observed production values fall within 1% and only a very few fall outside of +/-3%.  Or it could also mean that 2/3 are found within +/-3.5% and only a few fall outside of 4.5%.  Both of those are a 5% spec.  I was just wondering if the distribution was more "leptokurtic" as the stats folks would say (http://en.wikipedia.org/wiki/Kurtosis).  The presumption is based on what RG noted about more recent production techniques, and the possibility for more consistent production.

I have no insight to offer here. I don't test enough of SMT parts to even have a feel for it I'm afraid. The semi's I work with are binned after manufacture. There's no "manufacturing process for tolerance A, and a different process for B" -- everything is tested off the line and binned after the fact. If the same applies to resistors then a 5% tolerance means it's unlikely you'll have anything that's within +/- 2% of the value because all those chips got harvested for the +/- 2% bin and the +/- %1 bin. 

[Ardric]

Warning, unsubstantiated blather follows!

I've read a few anecdotes about SMT resistors and noise.  Apparently many common lines of metal-film SMT resistors exhibit dramatically more noise than is typical for, say, a 1/4W thru-lead part.  I guess if noise is inversely proportional to bulk it kinda stands to reason.  This doesn't mean that SMT is inherently inferior, but if the application depends on very low noise then you'd be well served to pick your parts carefully and measure to make sure you're getting what you think you are.

[Unbeliever]

This doesn't mean that SMT is inherently inferior, but if the application depends on very low noise then you'd be well served to pick your parts carefully and measure to make sure you're getting what you think you are.

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? Ok. And how do we 'pick our parts carefully'? You mean using rubber-coated tweezers, to avoid making the slightest mark on each resistor? What criteria should be applied to be 'careful'?

Glad you warned us. 

[earthtonesaudio]

The smaller the part is, the smaller the unwanted or parasitic inductance, capacitance, and resistance in the leads becomes.  This makes designing a little easier, as the parts are closer to "ideal" models, and allows for faster speeds.  And potentially it can allow for better energy efficiency.  That much at least is real, and governed by Ma Nature.

Think of the difference in inductance with a wire-wound power resistor compared to a tiny surface mount resistor. 

[gez]

Warning, unsubstantiated blather follows!

I've read a few anecdotes about SMT resistors and noise.  Apparently many common lines of metal-film SMT resistors exhibit dramatically more noise than is typical for, say, a 1/4W thru-lead part.  I guess if noise is inversely proportional to bulk it kinda stands to reason.  This doesn't mean that SMT is inherently inferior, but if the application depends on very low noise then you'd be well served to pick your parts carefully and measure to make sure you're getting what you think you are.

This has come up before.  Absolute rubbish as far as I'm concerned.  You sometimes have to rethink PCB design procedure when dealing with SM stuff.  Very easy to get things wrong as components and traces are closer together, in which case things could become noisy; but with a well-designed board...nah, sorry I don't buy it (and have never had a noisy circuit compared with its thru-hole equivalent).

[R.G.]

The semi's I work with are binned after manufacture. There's no "manufacturing process for tolerance A, and a different process for B" -- everything is tested off the line and binned after the fact.

I believe that's true of all semiconductors. The manufacturing process is so exacting for semis that no one does multiple processes for different grades. In my previous life where we made our own semiconductors, there was A Process for all logic. Your design, by default, used The Process. If you wanted different performance grades, such as perhaps a deep sort for an extra 10-15% of max speed, those were tested after manufacture.

I believe the reverse is true for SMT resistors, possibly caps. The processes are much simpler, and probably result in all parts within basic tolerance. For resistors, the tolerance can be tested in by laser trimming for high precision parts, and I believe that the materials are different (e.g. thick film versus metal film) for different precisions.  Not sure about caps.

"Hole in the middle" parts were common in the 50s and 60s, though.

[Sir H C]

Semiconductor testing is sometimes not done.  If the yield is high enough and the boards they go on cheap enough (no package just bare die glued to a PCB and bonded out and covered in that black goop) then test the final board and toss the failures.  I know that is done, but for some stuff (automotive) exhaustive testing is done for all components driving up the cost.  Over temperature testing of every part, the concept there is that no one wants a recall for their semiconductor.

[Mark Hammer]

Okay, so an adjunct question that those of you with more direct knowledge might be able to answer: 

Just exactly what produces the variation in components that requires the use of tolerances?  In other words, why DOES one capacitor that is intended to be 0.1uf come off the line at 0.1073uf and another comes off at 0.0995uf? 

And related to this, given that chips and transistors are, at one level, essentially constellations of passive components arranged to produce an active result, are the same sorts of factors that produce production variation in caps, resistors, and diodes responsible for production variation in active devices?

Yours truly,
Mark "Walters" Hammer

[earthtonesaudio]

No direct knowledge, but I chat about this kind of stuff with the guys at work all the time.  I think the main cause of variation in passives is impurities in the raw materials, chaos in the manufacturing process (like static electricity or turbulence in pipes), or mechanical inaccuracies in the robots that put the things together. 

With actives, there's not just one technique used to get a particular transistor or IC out of a particular piece of silicon, but any technique you use has some inaccuracies.  ICs tend to have so many things built in for compensation that the end result doesn't care as much about the imperfections in the part.  A discrete transistor on the other hand, varies a lot compared to other in the same batch.  They are prone to the same causes of variance, but you just notice it more with a discrete transistor 'cause there's only one.  Also, the manufacture of the silicon (or germanium, or gallium arsenide, or whatever) is a delicate and tricky process from what I've heard.  The tiniest changes in humidity, temperature, ions in the air, etc can have a big effect on the semiconductor structure.  It's pretty crazy.

[R.G.]

Just exactly what produces the variation in components that requires the use of tolerances?  In other words, why DOES one capacitor that is intended to be 0.1uf come off the line at 0.1073uf and another comes off at 0.0995uf? 

And related to this, given that chips and transistors are, at one level, essentially constellations of passive components arranged to produce an active result, are the same sorts of factors that produce production variation in caps, resistors, and diodes responsible for production variation in active devices?

Most engineering curricula today include an engineering statistics course that contains a section near the end on the analysis of variations. Quoting an old Billy Idol song, there is nothing sure in this world, and there is nothing pure in this world. Well, maybe things are down where quantum effects are supreme. But at our gross level, NOTHING is exact. The basis of all precise human measurement could arguably be said to be the meter. Check this out: http://www.mel.nist.gov/div821/museum/timeline.htm After reading that, what's a meter?  The latest info relates it to the speed of light in a vacuum, but then you have to define the second.

A machinist, when asked to machine up some part will immediately ask you back what the tolerances are on each part. I have been reading up on machining for the last couple of years, and have realized that tight tolerances are very expensive. Imagine an articulated machine arm with one pivot. The pivot has a pin in it, and the arms rotate on this pivot. How precisely can you position the end of the second arm? The diameter of the pin is important, of course; so are the diameters of the holes in the arms. The accuracy of positioning cannot be better than the sum of the tolerances of the parts. A tolerance of 0.025mm is quite involved to hold with human-held measuring devices when machining. Given that the arms and pivot have three of these, the tolerances add to about 0.075mm, and that's the best you can do unless you start adding more zeros onto the cost of the arm assembly. Now imagine designing machines to put photoimages of IC features on a silicon wafer where the dimension of the features is in units of 0.045 micro-meters. Then there's the issue of temperature. Steel expands by about a factor of 13uM per meter for every degree of change in temperature. Aluminum expands by 23uM per meter. Are we going to use steel and aluminum in the same machine? Then there's the machine's flexibility. Machinists develop a mind set that all materials are best modeled as a block of rubber of different hardnesses. NOTHING is rigid, unfortunately not even the foot-thick cast iron and steel frames of the precision machines which make precision machines.

Measurement of anything should include a tolerance on the measurement, because there is no perfect way simply to measure what you have. A voltmeter measuring a 9V battery? Um, when was that voltmeter calibrated? And was it calibrated to a standard voltage with the standard traceable to the National Institute of Standars and Technology? What was the thickness of that plastic film we used to make those capacitors? Did we measure by micrometer? Or by stacking 1000 layers and then using a micrometer? And at what temperature? Was the micrometer measuring surface free of silicon and carbon particles (i.e. dust) and an oil film? AGGGGGHHH!!

About the only thing which is easy to get as true as you like it is a flat surface. By using three same-material surfaces and abrasives, one can (eventually) grind one to match the other and then both to match a third, and get a set of three surfaces as flat as you have the patience and attention to detail in cleaning abrasives from surfaces to get. This is how optical flats are made, and they are very flat indeed. This process is the "mother process" of all mechanical accuracy in actual practice.

I'm ranting. There is variation and tolerance on everything. Period. Measurement of that variation, let alone control of it, is a huge and continuing issue for all technical disciplines.

[iaresee]

And related to this, given that chips and transistors are, at one level, essentially constellations of passive components arranged to produce an active result, are the same sorts of factors that produce production variation in caps, resistors, and diodes responsible for production variation in active devices?

Yes. For integrated circuits there are myriad factors that are responsible for part variations. Wire sizes cannot be image with 100% consistency (especially as the process size gets smaller than the wavelengths used to etch the non-photoresist pieces away...like 45nm) -- we're talking atom-level variations in width here. You have atom-level variations in the thicknesses of the substrates: the insulator, the dielectric material you're using for your wells in your transistors, etc.

The smaller the process for the part, the more complicated the part, the more variations in the part. I'm not in IC design at my company, but we get regular presentations on the bleeding-edge-of-the-art of new fabrication technologies and I have to say the stuff we're doing with TSMC for our 45nm parts is MIND BLOWINGLY COMPLICATED. Just the fact that they can accurately and repeatedly use interference patterns to image things smaller than the width of the light you're using hurts my head to think about. It gets harder every time we go smaller to make ICs work right. And that's why FPGAs are getting more popular -- fewer and fewer companies can really afford to handle the complexities of complete custom ASIC solutions at modern process sizes. The tools are hella expensive (think $1M/seat), the masks are hella expensive (we're budget tens of millions for masks for our handful of devices being released next year), mistakes cost you time and major money so time-to-market can suck. Yay for me.