Measuring ~1 nH parasitic inductance in SMT resistors

[jleom7]

We have discovered that the ESL of a 2010 package SMT resistor is important in a switch mode power supply.  The resistor is a current sense resistor where the parasitic inductance actually is an issue.  See the attached article which describes the impact of ESL (equivalent series inductance) on some power supply circuits.

ESL is not a typical parameter spec'd for current sense resistors yet can vary depending on base resistor construction.  I need to find a test house to measure sub 5nH inductance of two different manufacturers so I can confirm my suspicion that ESL is measurably different.  We have two different manufacturers, one causes our power supply to go unstable, the other works fine.   I suspect one manufacturer has an ESL of sub 1nH and the other is somewhere around 10nH.

Here's the article on ESL impact on some switcher designs:  https://www.analog.com/en/technical-articles/switch-mode-power-supply-current-sensing-part-3-current-sensing-methods.html

So, does anyone know of a test house which can measure ~0.2 to 10nH resolution of 2010 package SMT resistors?

We believe the internal construction difference between two manufactures is the reason for different ESL.  Attached are two xrays.  We suspect the serpentine resistor has higher ESL.

[DaJMasta]

Can you just select parts based on X-Rays then?  If you know one works and one doesn't, you should have a pretty good success rate comparing new parts with the same methodology, I'd think.

If 10nH or less is effecting the design, is it possible to reroute the PCB to reduce the loop area involved instead?  A cm of trace length could easily be as much difference as the two parts, so if you can shave that amount off of your layout, it could be more tolerant of higher inductance resistors.

Why do you need a test house to do it, business certification reasons?  You should be able to measure the difference on a consistent jig with a good LCR meter or a VNA, and setting it all up should be pretty short work.

[graybeard]

What is the resistor value?  The smaller the resistor the easier it will be to measure the ESL.

The trick to measuring such tiny packages in primarily in the design of the test fixture.  I recently was at a convention with a Hioki was showing some of their impedance meters.  They has some very impressive fixtures that could measure components as small 0201 up into the GHz.

[rfeecs]

So, does anyone know of a test house which can measure ~0.2 to 10nH resolution of 2010 package SMT resistors?

https://www.modelithics.com/

[T3sl4co1l]

Simple, set up a fixture with a toy version of your circuit and measure what's upsetting your circuit!

A switched current for example, will generate a measurable voltage blip, V = L * dI/dt.  For testing purposes, this might simply be a current from a voltage source, through a resistor, at low duty cycle so as not to waste too much power overall.

More resolution can be obtained by measuring the out-of-phase component of the voltage, i.e., if a square wave current is applied, and the measured voltage passed to a mixer (multiplier) that's also passed the quadrature (90° phase shift) of the current, then the average result is the inductive voltage drop -- read out as a DC voltage.  (Preferably this is done with a sine wave; square waves work, given suitable assumptions about the circuit.)

Or the opposite, applying a square voltage and measuring the out-of-phase current (or its rise time).

Tim

[thm_w]

http://www.vishay.com/docs/30100/wsl.pdf vishay datasheet states "Very low inductance 0.5 nH to 5 nH" but doesn't give the actual value. I'm assuming it varies with package size.
https://www.rohm.com/datasheet/MCR18EZHFL/mcr-low-e Rohm doesn't mention it at all.

Have you tried asking the vendors of the resistors yet?

[snoopy]

If your circuit is so sensitive to chip inductance wouldn't pcb trace be adding to this inductance ? Sounds like this circuit is very marginal and would be flaky out in the field with a lot of returns from customers.

[floobydust]

Welcome to the forum.
You are doomed if nH on a sense resistor matters in your SMPS. The power devices alone will have much greater inductance on their leads.

This is the kind of thing a LED streetlight SMPS would encounter 

[floobydust]

Component leads and PCB trace inductance are many times greater than the squiggle OP is chasing after in a 2010 resistor.
Just get an RF resistor usable to GHz instead of designing and measuring parasitics to save pennies.
Would you think this is realistic, based on your power experience?

[Berni]

I'm guessing the PCB design is more important than your resistor.

But if you really wanted to measure its inductance you can certainly do it. The typical LCR meter could have issues with such a sensitive mesurement, but there are big boatanchors of LCR meters that are typically called Impedance analyzers. For example the HP 4294A can measure 1nH to 3% accuracy when tested at 10MHz or above, but at this point the performance of the test jig that holds the resistor also becomes important.

Don't know of any labs that would offer such a service but you can likely lease one of these impedance analyzers from the various companies that rent out equipment. The things are quite pricey bits of kit so it doesn't make sense to buy one for a one off thing.

[mzzj]

If your circuit is so sensitive to chip inductance wouldn't pcb trace be adding to this inductance ? Sounds like this circuit is very marginal and would be flaky out in the field with a lot of returns from customers.

OP said that it's a current sense resistor so you don't need to (necessarily) worry about pcb traces.

1mOhm 1nH shunt resistor works well to about 100khz (1.5dB) and if you are interested about phase information its only good for something like 10khz. 

--
Sorry, no idea of test houses.
If the frequency stays within sub 10Mhz its reasonably easy to rig up your own test setup from function generator, preamplifier and scope. Add power amplifier to the mix if you wish.

[T3sl4co1l]

Component leads and PCB trace inductance are many times greater than the squiggle OP is chasing after in a 2010 resistor.
Just get an RF resistor usable to GHz instead of designing and measuring parasitics to save pennies.
Would you think this is realistic, based on your power experience?

Is there such a thing as an "RF" resistor that's not in the 50-100 ohm range?

The underlying pattern is this: the resistor is simply some conductor over the ground plane.  The higher above it is, and the narrower it is, the higher the transmission line impedance is.

There isn't such a thing as an "RF" 1mΩ resistor, because no one has a transmission line that matches 1mΩ.  (There can be a resistor with Kelvin leads, with matched mutual inductance, such that the current-sense bandwidth is flat into the GHz; but the inductance of the main current path, will necessarily be limited by physical dimensions, period.)

We're talking short transmission line segments here (<< 1ns), but the conclusion is the same: make it wider and lower, and you'll be better off than you were.  In short: use "wide" resistors (say, 1020 vs. 2010), use more in parallel, and use more vias flanking the common pad (or even via-in-pad if your soldering process or fab cost can support it).

Incidentally, a lot of "wide" resistors are trimmed in the same way as usual (i.e., with slots that neck down the current path, increasing inductance).  Perhaps try the Susumu(?) resistors, that claim to be many segments in parallel, in a wide body?

Beyond that -- if this is, like, GaN switching shit, there's not much you can do as long as the components have to remain on the top surface of the PCB.  Even if you integrate resistance into the PCB itself, you still have the stray inductance of pads, traces and vias, carrying that current from the device, into the depth of the PCB.  There is only one possible solution at this point: choose a lower switching current, with more channels acting in parallel.  This relaxes the impedance demands per channel, making the layout tractable.

Tim

[RFDUK]

Connect a known high tolerance chip capacitor in parallel (on top of) with the resistor under test. For example 5pf +/-0.05pf.

Use a return loss bridge or VNA with small coupling loop on the end of a coax. 2 turn loop diameter say 1cm. This allows you to couple into the resonance in the above with only a small loading impact. Set the bridge to display a return loss of sub 1dB and perhaps you will see the resonance frequency of the components under test. From this you can calculate inductance and compare one resistor to another.

If the loop isn't successful in resolving a resonance due to insufficient coupling a direct connection may be necessary. Connect two high tolerance capacitors in series to form a matching network, connect these across the resistor. Say one 5pF, one 20pF. Connect the return loss bridge coax across the 20pF, inner to the capacitor junction. Again calculate inductance from the observed resonance.

Absolute accuracy of the calculated inductance may not be great, but will facilitate A to B comparisons.

[rf-messkopf]

The underlying pattern is this: the resistor is simply some conductor over the ground plane.  The higher above it is, and the narrower it is, the higher the transmission line impedance is.

Correct. For this reason it is best to measure the resistor in the environment in which it will be used, i.e., the same PCB material, dimensions, ground planes, etc. Therefore a test fixture should be realized on a suitable PCB. The resistor can then be accurately measured with a VNA (provided its resistance is not too large) if the reference planes of the measurement are located right at the SMD pads. One way to achieve this is a TRL calibration. You can realize the necessary calibration standards on the same board as the test fixture. This will give you the complex impedance between the reference planes, from which you can work out the series inductance of an equivalent circuit, just like an LCR meter does.

I recently had a play with TRL calibration and test fixtures in my hobby lab, see my homepage at https://www.mariohellmich.de/projects/trl-cal/trl-cal.html for some information.

[JohnG]

RF resistors that can handle some current and are labeled as such are expensive, and tend to have values like 25, 50, 100, or 200 ohms, although if you look really hard, you can sometimes find 10 ohm. Often 10-100X the price of other resistors.

It was brought up that wide resistors often have a single cut from the edge, so they might don't buy you a lot - check carefully. However, in principle, this is the right approach, i.e. make the resistor current path wide and short. With a good ground plane underneath and an ideal (homogeneous, zero thickness) resistor, a 1020 footprint will have about 25% of the inductance of a 2010.

Of course, things are never ideal. If you don't have a ground plane and guarantee that the return current flows under the resistor in the opposite direction, you need to fix your layout first. If you do, I have found that you can use multiple smaller resistors instead of a wide resistor, which reduces the effect of resistive element etching pattern.

Finally, if everything else is optimized, the thickness of the resistor can matter. You want the resistor current to flow as close to the ground plane as possible to minimize inductance. but most resistors have the element on the top side of the chip substrate. You can mount these upside down, which manufacturers absolutely hate, or they may just laugh at you (and I mean this literally). Or, both Susumu (look at KRL and PRL series) and Rohm (UCR) make resistors with the element on the bottom of the chip.

Hope this helps,
John

[JohnG]

I forgot to add:

If you are making power supplies (or probably anything) in any quantity, using a particular resistor is a particular pain in the ***. Inevitably, some genius in sourcing will find another, lower cost resistor and pester you repeatedly (if you are lucky), or change the part anyway and then blame the engineer when problems start (if you are not lucky). This is less likely if you use several more standard resistors in parallel.

John

[Doctorandus_P]

Have you considered to solder a string of 10 to 100 of them in series for measuring?

[JohnG]

I forgot to add anything that actually answers your question . You might want to give the app note below a look. The estimation method may be good enough for your needs, and does not require an impedance analyzer or VNA. As always, YMMV...

http://www.ti.com/lit/an/sboa094/sboa094.pdf

John