The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> RF Design >> 50 ohm buffer 10MHz to 40 MHz (wafer probe)
https://designers-guide.org/forum/YaBB.pl?num=1400556531

Message started by mixed_signal on May 19^th, 2014, 8:28pm

Title: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by mixed_signal on May 19^th, 2014, 8:28pm

Hi,
I want to design a 50 ohm buffer in frequency range 10MHz-40MHz which will be followed by wafer probes and then instruments. The baseband output of a receiver will be buffered.

I am planning for PMOS source follower with
a) current source load
b) resistive load

PMOS for low flicker noise. Which one is better a) or b).
Any other suggestion? I have no power limit. I need low noise.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by RFICDUDE on May 20^th, 2014, 8:16pm

I would think you would want a current source load just to avoid any biasing issues. If you use a resistor load then you have to make sure the pmos is sized to deliver enough current to keep the output voltage where you want it to nominally be.

Perhaps you can use a long channel device for the current source to reduce its flicker noise contribution.

Even with a current source load, you may want some resistance between the pmos source and the output pad if linearity is a concern.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by tm123 on May 22^nd, 2014, 10:59am

I agree that biasing the PMOS source follower with a current source is a better way to go. I assume you need to impedance match to the load, so you can either make the 1/gm of the PMOS source follower equal to 50 ohms or make the 1/gm smaller than 50 ohms and put a resistor in series with the output pad so that the total output resistance is equal to 50 ohms.

Tim

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by mixed_signal on May 22^nd, 2014, 11:22am

Hi RFICDUDE,

Thank you very much! I have attached the circuit. i am using 3.3V I/O PMOS devices in 65nm process. The bias current is 5mA.

1. Is the current too high?

2. gm of M1 =1/Rload=1/50 (for S11=-15dB)

Gain=gm*Rload/(1+gm*Rload)=0.5
Is there a way to increase gain without matching circuit?

3. The sizes are
(w/l)1=600u/400n
(w/l)2,3,4,5=400u/400n
Is it ok to have different length devices in current source & M1? I am worried since thresholds may not track each other.

Is the design OK?

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by tm123 on May 23^rd, 2014, 7:56am

Hi mixed_signal,

What is the output power you are trying to deliver to the 50 ohm load? To determine the necessary current, calculate the peak signal voltage divided by the load impedance (50 ohms in this case). For example if you need to deliver a sine wave of 0dBm output power, that is 316mV peak and 316mV/50=6.3mA so you need AT LEAST 6.3mA to drive the load. You should put something more than this to be sure the source follower device never turns off. Any linearity requirement will determine how much more than 6.3mA you need.

If you are impedance matching with this type of circuit your voltage gain will be 0.5. You can get more voltage gain by mismatching the impedances, it's just a voltage division. The tradeoff is S22 will be worse. In most cases, return loss better than 10dB should be good enough.

I do not see a reason to match the lengths of the current source and the source follower transistor. You want to maximize the output impedance of your cascode current source so that implies large length for M4 and high gm for the cascode M2 (large W/L). The transistor M1 should be sized to optimize its performance as a source follower.

Hope this helps.

Tim
(edited to correct transistor names in last paragraph)

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by tm123 on May 23^rd, 2014, 8:01am

One other comment, the decoupling capacitor on the gates of the cascode is connected to the wrong rail. Since the gates of the cascode are referenced to the power supply the decoupling capacitor should also go to the power supply. Always decouple to the rail that the node is referenced to.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by mixed_signal on May 24^th, 2014, 4:58am

Hi RFICDUDE amd tm123,

Thank you very much for your suggestions

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by RFICDUDE on May 28^th, 2014, 4:22am

Glad to be of assistance.

One thing to consider, if you design the source follower 1/gm to be 50 ohms then half of the output signal will be dropped across the internal resistance of the source follower (6 dB gain loss). This may cause the output buffer to limit the linearity performance depending on what linearity you expect from the circuit before the buffer.

If buffer linearity is a problem then you can improve linearity by making 1/gm << 50 ohms and then inserting a resistor in series with the source to make the output impedance 50 ohms (if matching is important). There will still be a 6 dB gain loss, but the linearity will be much improved because most of the gain loss is due to a linear resistor instead of the electronic output impedance of the source follower.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by Pravesh Kumar Saini on Jun 30^th, 2014, 11:10pm

Hi,
I am designing the buffer to drive the 100ohm differential probe. I am having nmos input common source loaded with 100ohm differential resistor and source degenerated with 100ohm resistor for linearity, Is this a correct approach?
Is it wrong if I use the 100ohm differential resistor to model the 100ohm differential impedance probe at this common source(CML) buffer output parallel to 100ohm differential load?

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by raja.cedt on Jul 1^st, 2014, 12:44am

@RFICDUDE,
I didn't understand about 6dB signal loss concern here(common for all matched load systems, 50% eff), you can design a source follower with 1/gm=50, this will only require lot of power but still your gain will be close to 1.I don't think you can improve the linearity by adding series resistance, because your are not changing loop gain. In case if you are interested in high linearity better go for multi loop gain source follower aka super source follower, which has higher loop gain.
One caution, if you are interested in good reflections over the large signal source follower may not work, due to the variation of gm over the time period.

CML driver (resistive loaded diff pair) with 50ohm load is the best out of though it draws huge power...

@Pravesh Kumar Saini:Your approach works but gain will be 0.5 instead of 1, if so use 25ohm degeneration

Thanks,
Raj.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by Pravesh Kumar Saini on Jul 1^st, 2014, 7:07pm

Hi Raja,
Thankyou very much for your prompt response, though I have jumped into this discussion but my problem is same as in this discussion here but I need to buffer the signal of BW ~2GHz and OIP3 > 15dBm, NF>15dB.
Incase of CML driver, can I boost the gm of input pair using super source follower approach, basically I want to make it more linear by doing I/gm<<degenerated_resistor so that gm,eff~=1/degenerated_resistor and gain load_resistor/degenerated resistor and independent of input pair's gm. Is there anything wrong in this theory.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by raja.cedt on Jul 2^nd, 2014, 7:23am

Hi pravesh,
May be you have confused with my previous post, what I am saying is if some one would like to deploy source follower based o/p buffer and if they are looking for high linearity then super source follower(SSF) make sense. However for a CML driver I have no idea about SSF.
In case of linear CML driver if you have slightly relaxed S22 requirements then I would suggest go some where between 50 and 75 ohm load res. IIp3 also will increase with gm*rs. Your noise factor depends on Rs more or less so I have no idea where you are going to land up, I mean you have very less flexibility.

Hope this helps,
Raj.

Thanks,
raj.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by Pravesh Kumar Saini on Jul 2^nd, 2014, 5:13pm

Hi Raja,
Thanks for your opinion, one last question, If my this 100ohm differential output impedance test-buffer drives 100ohm differential input impedance probe, do I need to put 100ohm differential load in simulation for it, I am confused because 100ohm is its characteristic impedance not 100ohm physical impedance? What's your opinion on this?

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by raja.cedt on Jul 2^nd, 2014, 5:30pm

Hi,
In theory you have to use a transmission line model with proper Zo could be either s-param model or spice model with 100 ohm load. But for your low frequency applications(since you asked abut 2Gz BW) I am pretty sure 100ohm load works, but take care about ac coupling. Some times it creates problem...

Thanks,
Raj.

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by Pravesh Kumar Saini on Jul 4^th, 2014, 7:48pm

Hi Raj,
Thank you very much for your opinions. It was really helpful for me.
Regards,
Pravesh

Title: Re: 50 ohm buffer 10MHz to 40 MHz (wafer probe)
Post by RFICDUDE on Jul 5^th, 2014, 6:11pm

Hi Raj,

The issue I was pointing out is that if you design the source follower such that 1/gm=50 ohms and the load is also 50 ohms then half of the output voltage has to be dropped across the 1/gm resistance of the source follower (-6dB gain). The gain will be closer to 1 when 1/gm is much less than 50 ohms which means the output is no longer matched to the load.

You can't have the output impedance of the source follower be the same as the load resistance unless you can tolerate the gain being -6dB.

Alternatively, a common source amplifier can have gain because you can set the amplifier output impedance with a resistor and compensate the gain by designing gm high enough to drive the total load.

I hope this clarifies my comment.

Best regards