RF Active probe input impedance

[Nitrousoxide]

I'm currently designing an active RF probe (1MHz - 1GHz). I have a prototype designed, however, I have run into an issue.

Im able to impedance match all the components (doesn't really matter since the distance between nodes is very minimal) fine. It's the input impedance that Im having issues with.

The input impedance is meant to be in the order of megaohms (1-10 Mohm), but due to the nature of the internal capacitances (most likely body or cgs), the input forms an (almost) direct short to ground. Its not possible to place a high value resistor in series with the gate as that just forms a low pass filter with the capacitance which is causing the impedance to drop.

It seems like there's a massive tradeoff between bandwidth and input impedance. I know that there will always be a compromise between specifications, but is there a way to alleviate the issues? Is this something thats impossible to combat since you'd always have some form of junction capacitance? Could this just be an over-conservative model of the bf998 that Im using?

I've tried experimenting with both common source and common drain topologies, both present the same issue.

Schematics:

[awallin]

the owls are not what they seem...

2.5kOhm @ 70 MHz is par for the course if you believe e.g. this:
https://www.electronicdesign.com/test-amp-measurement/choose-best-passive-and-active-oscilloscope-probes-your-tasks

can you simulate an input impedance curve vs. frequency, like in the article above?

[David Hess]

At the high impedance input, impedance matching transmission line sections loses meaning and reducing capacitance is what is required.  It becomes more of a mechanical problem than an electrical problem.  Materials with a low dielectric constant can be used and holes cut to remove as much material as possible.  A ground plane in this area is contraindicated and if present should be removed.

Bob Pease showed an example in his book shown below.

[gcewing]

Instead of a resistor in series with the input, use a capacitor. This will form a capacitive voltage divider with the input capacitance, so you don't end up with a low pass filter. This is how the "x10" setting on a scope probe works -- it switches in a series capacitor with 1/9 the capacitance of the cable and scope input, giving 10 times the impedance at the expense of 1/10 the sensitivity.

[Nitrousoxide]

Thanks for the tips.

I added a nominal 1pF capacitance in series with the gate which improved the input impedance. I'm getting around 500 ohms at 500 MHz and 3.8 Kohms at 70 MHz, however, as expected the signal is attenuated (can be fixed). This seems to agree with the literature that was linked in the thread.

I understand that the model presented in SPICE is only a model, and that the final implementation will probably vary by a significant margin.

Screenshot of performance for completion sake:


Browsing literature for buffer circuitry, I see the dual gate NMOS being used as a source follower (common drain). However, I would have expected a cascode configuration to preserve the bandwidth the best.

Also, Should I add an inductor (to reject RF noise) between the bias and the gate (G1) of the nmos device?

[bson]

The BF998 has different input capacitance on the two gates; looking at the NXP datasheet it states Cg1 = 2.1-2.5pF and Cg2 = 1.2pF.  Assuming 2.5pF, if you use both gates like in your schematic you get 3.7pF.  That's very high for a wideband probe, typically they sit at 1pF give or take a little.  You could use only G2; look for example at Fig 17 on p. 8, where G1 is used for AGC and G2 for signal input.  (https://www.nxp.com/docs/en/data-sheet/BF998.pdf)  1.2pF is more reasonable.

[Wolfgang]

Did you see this one ?

https://electronicprojectsforfun.wordpress.com/rf-measurement-techniques/high-frequency-probes/

You can cur back on input capacitance using a 10:1 divider.

[Nitrousoxide]

The BF998 has different input capacitance on the two gates; looking at the NXP datasheet it states Cg1 = 2.1-2.5pF and Cg2 = 1.2pF.  Assuming 2.5pF, if you use both gates like in your schematic you get 3.7pF.  That's very high for a wideband probe, typically they sit at 1pF give or take a little.  You could use only G2; look for example at Fig 17 on p. 8, where G1 is used for AGC and G2 for signal input.  (https://www.nxp.com/docs/en/data-sheet/BF998.pdf)  1.2pF is more reasonable.

I dont get it? I have the circuit configured similarly as shown in figure 17. The only difference is that im using dual supply (hence no need for bias circuitry on the input) and I have Vagc implemented as a voltage divider. Isnt that just using G2 for the signal input? Since everything else would be RF ground (or close to) upon small signal analysis?

It then follows that the only relevant capacitance affecting the bottom mos is the Cgs of the top mos, which would always be there, no?

Did you see this one ?

https://electronicprojectsforfun.wordpress.com/rf-measurement-techniques/high-frequency-probes/

You can cur back on input capacitance using a 10:1 divider.

Thanks for the article, will give it a thorough read. Looks interesting.

[bson]

I dont get it? I have the circuit configured similarly as shown in figure 17.

My bad, for some reason I got the impression you had your gates connected together, with the two resistors for bias between the supplies! 

What does your board look like?

[Nitrousoxide]

Schematic:


Its a four layer board. I originally decided to attempt with a two layer. However, I just couldn't justify the elimination of the power planes. Thus the stackup from top to bottom:
- Signal. Controlled impedance traces to 50 ohms. Flood fill polygon is grounded with via stitch.
- +5V
- -5V
- GND with via stitch.

I have removed all planes over/under most of the high impedance section. The rest is flood filled to act as coplanar waveguide.

Top:

Bottom:


Component values may be tweaked, some points will be 0 ohm jumpers too.