Questions about RFIC design using LC folded inductor source

RFIC design question

I am currently designing a LNA by using Si with 0.18um technology. My target is to design an LNA that has NF < 2db, Gain =15db and OIP3 >-3, I<4mA.

I am using a LC folded inductor source degeneration topology. Below are some of my questions:

1. Does folded technique will provide better noise performance than cascode technique?

2. I get a gain of -50dB for my design(i know it is wrong...), could i know what went wrong with my circuit?

3. Any idea how to improve NF of LNA?

4. Does adding a capacitor at Gate-Source terminal at first stage NMOS will help to reduce the NF (especially gate induced noise)?


Thank you in advance.

Re: RFIC design question

There are no big differences in RF performances between cascode and folded cascode structures, but folded cascode allows for lower voltage operation.

Noise figure improves by minimizing the size of the input device.

-50dB of gain means that your LNA is not DC supplied.

RFIC design question

4: the parallel cap helps to increase the Cgs, the resonate Lg, the input Q, then improve the NF

comments of "There are no big differences in RF performances between cascode and folded cascode structures, but folded cascode allows for lower voltage operation. " is right, the power is doubled then.

Re: RFIC design question

neo said:

4: the parallel cap helps to increase the Cgs, the resonate Lg, the input Q, then improve the NF

Click to expand...

neo said:

Neo, could you explain further about how the parallel cap helps to improve NF? How is it relate to Lg and Q?

By the way could you provide any formula or intuition way to prove this?

Thanks!!

Click to expand...

Re: RFIC design question

electronics_sky said:

neo said:

4: the parallel cap helps to increase the Cgs, the resonate Lg, the input Q, then improve the NF

Click to expand...

neo said:

Neo, could you explain further about how the parallel cap helps to improve NF? How is it relate to Lg and Q?

By the way could you provide any formula or intuition way to prove this?

Thanks!!

Click to expand...

i wouldn't add a capacitance to gate-source. it worsens your input match, you won't be able to achieve broadband match, moreover the thermal noise due to the series resistance of Lg will still exist even if you add a capacitance to gate-source. so i don't believe that it will improve NF and your design will be more vulnerable to process variations.

to your question, the gate-induced noise is never the dominant noise source in LNA design, it is proportional to Cgs^2 and Cgs is mainly some tens of femto farads, therefore you can imagine what the impact of gate induced noise is.

15 dB gain and 2 dB noise figure should be easily achieved in 0.18 um process.

Click to expand...

RFIC design question

Hi estradasphere, How about if i only need a very narrowband (2.11G-2.17Ghz, BW = 60Mhz)?

I know Channel noise of first stage transistor is the dominat in LNA design, however if we ignore the gate induced noise in our calculation then it will become dominant. Therefore we need to add a capacitance to gate-source in-order to decouple the Q of first stage transistor in order to help reduce the gate induced noise.

By the way, what are your suggestion to improve the NF of a cascode LNA instead of add a capacitance to gate-source?

As for the cascode topology, i am thinking of using a inductor load to decrease the parasitic capacitance of second stage PMOS, Cgs. Anybody think this will help to improve the NF as well?

As i know, with the smaller channel length we will have better NF, could i know any formula will prove this? If we have a same W/L ratio from 0.18um process and 0.5um process which one will provide smaller NF?

Please comment and correct me if i am wrong.

Thanks

RFIC design question

electronics_sky, u can refered to "CMOS Low-Noise Amplifier Design Optimization Techniques", it may help.

If u control the input Q not to high, it will meet ur
matching requirement.

And the process variations, the input transistor itself will suffer from the corners, right?

But frankly speaking, estradasphere, i donnot like the adding Cgs neither, anyway, it is not a generic approach.

RFIC design question

Hi Neo, i am currently refered to "CMOS Low-Noise Amplifier Design Optimization Techniques", but i found that its approach is quite different as what was discussed by Thomas H. Lee in his "1.5v, 1.5Ghz LNA for GPS".

I am sorry, i don't understand what you mean by "And the process variations, the input transistor itself will suffer from the corners, right?"

Re: RFIC design question

Can you give an article which describes the LC fold technique ? (I dont mean to Folded cascode...but the LC-Fold technique).

B.R

Re: RFIC design question

electronics_sky said:

Hi estradasphere, How about if i only need a very narrowband (2.11G-2.17Ghz, BW = 60Mhz)?

hi,

if you want to make an "on-chip" narrow-band matching, then it is more critical, because you have to realize your input match exactly at 2.1 GHz (if you intend to make an off-chip matching with a package, then this is no more critical). i would run a simple s-parameter monte-carlo simulation (takes maybe 20 minutes) and see the variation of the input match. the lna should be as simple as possible and i would avoid to use too many elements in my circuit.

to your question; in my opinion, keeping the W/L ratio constant, the circuit with the smaller L would give the smallest NF, because, you get the same gain with less transistors, which means less noise and less parasitics. please correct me, if i'm wrong.

Click to expand...

RFIC design question

I do not agree with both estradasphere and neo. Actually, the C, which parallel to Cgs, will decrease the input side Q (check with Lee). The Q will affect NF, linearity of the LNA performance. The higher Q, the smaller NF. As to the paper "CMOS Low-Noise Amplifier Design Optimization Techniques", there is a mistake there. When the parallel C added, the effective wt will be changed to gm/(Cgs+C), it is not convention wt any more. It means it will degrade the NF at the same bias current. The equation (26) will not be valid.

Re: RFIC design question

noiseless said:

I do not agree with both estradasphere and neo. Actually, the C, which parallel to Cgs, will decrease the input side Q (check with Lee). The Q will affect NF, linearity of the LNA performance. The higher Q, the smaller NF. As to the paper "CMOS Low-Noise Amplifier Design Optimization Techniques", there is a mistake there. When the parallel C added, the effective wt will be changed to gm/(Cgs+C), it is not convention wt any more. It means it will degrade the NF at the same bias current. The equation (26) will not be valid.

Click to expand...

hi noiseless,

your theoretical point of view may be correct. maybe you'll get 0.2 dB lower noise figure with parallel C or maybe a little bit lower? but the main problem is the robustness of your design despite several process variations. a high input Q might be good for noise figure, but quite risky in terms of on-chip input matching (again, process variation) and stability of your lna, the input Q of cmos transistors in common-source topology is high enough even without adding any capacitances in parallel.