What kind of SMD resistors to use in RF circuits?

[sundance]

I wonder what SMD resistors I should use in an input attenuator stage which has to dissipate about 1 watt.
Mouser has dedicated RF resistors (https://www2.mouser.com/Passive-Components/Resistors/SMD-Resistors-Chip-Resistors/_/N-7h7yu?P=1y95kbl) but you only get a handful of values which makes the attenuator designing a little difficult. And they are expensive...
Since MELFs are metal film types with a helical cut in the film they will have some significant inductance.
Thin film resistors are quite pricey at $1.64  (https://www2.mouser.com/ProductDetail/Vishay-Thin-Film/PHP01206E40R2BST5?qs=sGAEpiMZZMvdGkrng054txRufvdcoZTXWzQJLohsX8s%3D)
Thick film resistors are reasonably priced at 0.46$, but are they usable for RF? (https://www2.mouser.com/ProductDetail/Yageo/RC2512FK-0740R2L?qs=sGAEpiMZZMvdGkrng054tx7%2F6%252BNA3LAJ%252BNJncdlxtPs%3D)

Or might combining 4 "run-of-the-mill" 1206 SMDs do the job?
(We're talking about frequencies below 1.5 GHz)

Addition info I found: http://www.resistorguide.com/thin-and-thick-film/
High frequency behavior: Thin film features lower parasitic inductance and capacitance. However, inductance may be high if thin film is executed with a cylindrical shape that is spiral cut.

[thinkfat]

You would not want to dissipate that much energy in a single stage, I guess? 1 Watt is a lot. What's the total attenuation you're looking for?

[T3sl4co1l]

1.5GHz is a wavelength of 200mm, so a body or path length of several mm won't make much difference.

2512 resistors should be perfectly suitable.

Mind, the transmission line approximation is modified by impedance ratio as well.  It's not enough to have a path length of λ/10 or so, but also as the element impedance differs from system impedance, the length needs to be as many times shorter.  So, if you have a 50 ohm attenuator with a 300 ohm series element, or 8 ohm parallel element, those elements need to be six times shorter still, or λ/60 say.

But a 6.4mm resistor still isn't far off so I don't think that will be much of a big deal.

You can also consider how to constrain those limits further.  What if you use more stages cascaded, to get the same total attenuation?  The extreme values required, are less extreme, allowing better performance with larger components.

You can also consider cascading stages of telescopic attenuation, to distribute the power dissipation and therefore enable smaller resistors, say 0805s (1/8W) instead.  Downside is, you'll need many more values (BOM items); series or parallel combinations (would allow some optimization) are probably discouraged for bandwidth reasons.

Here's an example with equal dissipation and 50 ohm ports (and, I forget what the output was, -40dB maybe?). https://www.seventransistorlabs.com/Images/DistAttenuator.png  Mind, the extreme-value problem is back, because the first attenuator is like 0.5dB (large shunt, small series values), while the last is like 20dB (large series values).  There's probably an intermediate attenuation that gives the least-extreme resistor values, so attenuation stages around that value would give the best compromise between dissipation and part value.

Tim

[sundance]

My total projected attenuation is 20 dB. For mechanical reasons I'd like to use a T-attenuator configuration.
Using two stages with 10 dB each would mean that my 1st stage's resistors have to dissipate 519 and 338 mW. Too much for a 1206.
The most I can do with 250 mW resistors would be 3 dB, reducing the power to 500 mW. And if my 2nd stage had 7 dB (max. power dissipation here is 190 mW), my final 10 dB stage can easily handle the rest (assuming that my math is correct...).
Thanks for your valuable input - greatly appreciated.

@Tim:
Frankly I was more concerned about the parasitic inductance of the SMD resistors. Like you said, the mechanical dimensions can be pretty small with 1206 components.
So both thin and thick film types should be OK with 1.5 GHz?
What's the bandwidth problem when using parallel/series combinations?

[T3sl4co1l]

Not ESL, necessarily -- as I noted the absolute ratio away from Zo is what your problem is.  (More precisely, (Z/Zo + Zo/Z) / 2 or something like that.)

R < Zo has ESL as the dominant limitation.
R > Zo has Cp as the dominant limitation.
BW is maximum at R = Zo, where sqrt(ESL/Cp) = Zo as well.

Exact values depend on resistor geometry, but somewhere in the 50-100 ohm range is typical.

Parallel and series resistors add extra pads and width or length, and therefore exacerbate the mismatch problem.  To some extent, impedance can be controlled by dropping or removing ground plane underneath the resistors.  Probably not enough to keep a 2512 resistor (3.2mm wide) flat further into the GHz, but maybe a 2010, and likely 1206 or a parallel pair of 0805s.

Note that 0603 and 0.8mm traces over 1.6mm substrate is about 50 ohms flat.  On a thinner substrate, or with closer inner layers, narrower traces and components are needed.  So, you'll probably want something on the thicker side to keep trace and component widths reasonable.

Tim

[sundance]

Just checked some datasheets: https://www.bourns.com/docs/Product-Datasheets/CHPREZTR.pdf
Unfortunately there's no mention of parasitic L/C.
Have you got any ballpark figures for ESL/C_P for this kind of resistors?

[T3sl4co1l]

Basically just physical dimensions.  So, 1nH/mm or thereabouts, and, Idunno, 30fF/mm or whatever?

ESL will be slightly more for spiral cut parts but it's really not much.  What underlies the figure is the resistivity of the materials, relative to the impedance of free space.  (Would seem to suggest that thick film or metal film is worse, but -- who knows what formulations they use; I wouldn't worry about it.)  Point is, the spiral parts are for higher resistances where Cp dominates instead, so you aren't worried about ESL.

Most importantly -- if you're concerned about this, then I'm guessing you have the means to measure if it's behaving or not, and therefore you can at least try different types or sizes and see which ones behave best.  Don't take my word for it, MEASURE.

Tim

[pigrew]

You may want to look into dedicated surface-mount attenuators. I've never used them, but the specs look reasonable and the price acceptable (a few US$). For example, Skyworks ATN3590-20 which are rated at 2 W and 40 GHz. Anaren also has D10AA20Z4, which are 10 W, but only 3 GHz.

[sundance]

MEASURE, I will! 

@pigrew:
I had no idea such smd attenuators existed! Anaren's 10 watts version is < 2$ @digikey! The specs look promising and I will give it a try, too.

Thank you very much for your posts; I definitly (or is it definetly?) learned a lot today!