OT 0805 resistor noise

residual stress.

surface tension holds them in place at all times, and

more theoretical than a practical "problem".

for

We have a very nice semi-automatic p&p, made in Switzerland! I think it cost $25,000 or so. I'll look up the model when I get to work. My guys just did my new VME prototype board on it; it took about 3 hours to place 1050 parts on a very dense board.

I've seen a Panasonic turret-feeder place 6 parts per second, but that's a $300,000 or so machine, and setup is a major deal.

John

How does setup work for p&p machines? Do you feed it Gerbers or something, along with telling it what part numbers are on what reel? Or is the p&p setup completely separate from the PCB layout process?

The physical origin of shot noise is the flow of electrons whose mutual correlations are small--deriving the usual formula [ = 2eI] requires assuming that the electrons arrive in a Poisson process, i.e. totally uncorrelated.

Conduction electrons in metals at ordinary temperatures are very highly correlated, on account of the Pauli exclusion principle--the Fermi level is ~100x larger than the thermal energy (a few volts vs 25 mV at 300K). Where the conduction electrons are less dense, e.g. in vacuum or in the depletion region of PN junction, the electron correlations are much smaller, so effectively you have full shot noise in tubes and semiconductors without applied feedback.

The shot noise PSD of a metal resistor of physical length L and electron mean free path lambda is suppressed by a factor of roughly lambda/L compared with the 2eI formula. You can make a zero-order model of this by taking a resistor to be a series string of L/lambda resistors, each with a full shot-noise voltage source =sqrt(2eI)*(R*lambda/L) in series with it, to model the effect of the scattering destroying the correlations. When you add this up in power, it comes out to be

= sqrt(L/lambda*2eI)*R*lambda/L = R*2eI*sqrt(lambda/L).

Since lambda is on the order of a hundred angstroms, whereas L is on the order of a millimeter, this is a pretty big number, 50 dB or thereabouts.

Tbe basic assumption involved in this model is that scattering is an elastic dephasing event, leaving the electron energies unchanged, so that the noise currents are forced to be correlated the rest of the time. This is true in a metal, but not true in a poor conductor or free space, where the occupation numbers of the conduction states are much less than 1, so that a short mean free path doesn't help there. (Finite temperature inelastic effects, e.g. phonon scattering, are less important in resistors because they're specifically designed to have near-zero temperature coefficients--i.e. the scattering is dominated by crystal defects and impurities.) Carbon-loaded thick-film resistors are a more complicated case, because a lot of the resistance occurs at the contact points between carbon grains. I also don't know enough about the physics of conductive polymers to be sure of how they'd behave.

Cheers,

Phil "noise 'R' us" Hobbs

Whoops, the third term should be R*sqrt(2eI*lambda/L).

Cheers,

Phil Hobbs

(more at

the plug at the beginning, the papers are further down the page)

residual stress.

surface tension holds them in place at all times, and

more theoretical than a practical "problem".

Interesting posts, thanks!

I am considering starting a SMT placement business here in the EU. How much does a smaller maybe used pick and place cost in the US? Just considering my options for a conservative start.

--
Siol
------------------------------------------------
Rather than a heartless beep
Or a rude error message,
See these simple words: "File not found."

Whoops, the third term should be R*sqrt(2eI*lambda/L).

Cheers,

Phil Hobbs

(More at

-- ignore the plug at the beginning, the papers are further down the page.)

Our semi-automatic (made by Essemtec) has a PC that spins the parts carousel and guides the operator to pick up and place the parts. We get a file from PADS that has the part locations, and we run a Basic program to reformat and tweak the data for the p&p. PADS coordinates are usually pin 1, and generally have to be offset to part-center for placement.

Most of the parts are loose, in little triangular bins around a replacable lazy-susan sort of thing. After we set up for a board, we usually keep that carousel loaded for future use. We keep a few common parts, like bypass caps, on a few reel feeders on the side. I could post a pic to abse, maybe.

Works great. Out first proto of this new board is working, three working days after we got the first stuffed board. It was perfect:

1050 parts, two FPGAs, 3000 lines of code so far, runs right, and still untouched by a soldering iron.

John

If you get a chance, a pic would be great.

Even things like resistors are loose in bins?? Seems like a hassle to get them properly oriented and right-side-up. (That's one of the time consuming steps for me in manual assembly.)

See pics in abse.

Cecily is assembling a board now, so I asked her about the flip thing. So she just hits the side of a resistor, in the feed bin, with the pickup tool, and it flips over; "just like playing tiddly-winks" she says.

John

Thanks.

So, is there any measurable shot noise from, say, a metal-oxide or carbon-film resistor?

It would seem to be easy to measure if shot noise were anywhere significant in comparison to Johnson noise. Get a decent power supply, RC filter it a couple of times, and apply to a 2-resistor DUT divider to ground. The halfway point is AC coupled to a low-noise amp, probably a jfet thing. More gain, highpass some to avoid 1/f stuff, TRMS voltmeter. With resistors in, say, the 20K range, the Johnson noise would be plain, a microvolt or two. Frob power supply voltage and note results.

Add a good wirewound or Vishay foil reference-resistor pair and a spdt switch or better yet a software-controlled relay, interface to the DVM, and do the noise-figure-meter/Dicke radiometer trick to resolve small noise levels.

That wouldn't be hard.

John

Google -> Groups -> Advanced Groups Search, etc...

--
 Thanks,
    - Win

I tried Google, and like i said, it was useless, and i do not know how you restrict it like you mention. Pleas either give an example or references.

residual stress.

surface tension holds them in place at all times, and

more theoretical than a practical "problem".

does

options for

Seems the minimum charge is around $100; there is the NRE for the "silk screen" for the solder paste, and the quote-per-part for assembly is definitely higher for low quantity than for higher quantities. Some assembly houses may break that down for "set-up charge" which (i think) would be a constant, and then quote-per-part for assembly, which would go down a little at (say) 1,000 then 5,000 etc. There also is an extra charge if you want parts sooner than standard thruput.

Thanks, John. Great pictures. I'll show those to my current p&p machine (aka my wife).

We do the same thing Cecily does, only with tweezers and a magnifying headset. It is just like tiddlywinks. But I worry about the possibility of damaging the part. And, as with tiddlywinks, if you're not careful the part goes flying somewhere... that's a big problem if it's a capacitor, because if it lands amidst a bin of other-valued capacitors there's no way to tell it apart. We're very careful to only work with one value at a time, for that reason, but it slows us down.

We do have an older, completely non-computerized, pantographic sort of manual p&p that cost a few K. It just helps pick up and place the parts... it has a vacuum pickup/rotation thing that rides on x-y rails. They still use it sometimes for doing small boards. It's still a lot better than placing parts with tweezers.

Of course, any p&p needs to be used with a board that has stenciled solder paste. If you're hand-soldering, you need to place and solder one part at a time.

We recently did our first 456-pin BGAs... and it worked!

John

That's one of the annoyances for me, too. Interesting.

--
Siol
------------------------------------------------
Rather than a heartless beep
Or a rude error message,
See these simple words: "File not found."

I was hoping to pay that much for a used automatic SMT p&p. There was one on sale that I just missed :(

And than there's owen that'll probably have to be purchased new due to rohc.

--
Siol
------------------------------------------------
Rather than a heartless beep
Or a rude error message,
See these simple words: "File not found."

Preheat with hot air from below, heat it with hot air from above, and after a while it just "sits down" all of a sudden?

--
 Thanks,
    - Win

It went through the conveyer on the reflow oven, deep dark tunnel, so we couldn't see what happened! They do ride a thermocouple through the oven, attached to a data-logging Fluke, to get the time-temp profile right. That's apparently the most important thing. They've done maybe

20 boards so far, with 100% success. They prefer BGAs to TSOPs now, although they may change their mind after they have to rework one.

It's sure nice to toss all your logic into one huge chip. No interconnect problems, and massively inefficient logic works just fine. Multipliers, adders, fifos, PID controllers, serial bangers, phase detectors, DPMs, whatever, just heap them on. It's almost obscene.

John