Problem about resistor on chip

In recent research or application, a lot of RF works had been done in CMOS Logic process.

It is really cool to enjoy the low cost of fabrication. However, some obstacles seemed to be unable to overcome.

1. how to fab accurate 50 ohm matching resistor in amplifier ?
Since every RF system had to be matched to each other, the accurate resistor will be necessary. Also, the resistor load of the amplifier, such as limiting amplifier, will have to be accurate to obtain stable DC operation point.
But the process tolerance is too large.( >15% ) How to resolve this problem ? My personal idea will be use multi-parallel resistors to reduce the interface resistor and make the ultimate resistor value more controllable.

2. when there is no 'clean' ground in the circuit, how can we improve it?
My personal idea will be add more substrate tap to make the voltage potential on the whole chip to be closer. However, this will inject more noise into the substrate, does any people
had better idea?

best regards.

Bi Han

just a thought

weeeell, i was just thinking, about the 50 ohms load thingy... if its the input to the chip you can just put the 50 ohms in series on the outside of the chip (with a nice 1% resistor). If it is leaving the chip, then you can do the same. If it is inside the chip, between blocks then perhaps you can use some sort of matched impedance matching that uses fets that have matching instead of resistors. If you have a very good voltage source (bandgap) and a very good current source (bandgap going to offchip "nice" resistor) you can compare an on chip resistance to your ideal one (R=I/V), then digitally tweak it. Lets say by putting some resistors in parallel with switches so that the circuit turns off the switches until the resistance seen there matches the ideal one. It could be run once on startup.

well, just some ideas.

mmmm...

It looks that no way to avoid bandgap to obtain anything accurate in CMOS. Mmm... I had to say that things should not be that complex.

Any better idea ? Does all companies deal with the load of amplifier in that complex way? Will that cause the high frequency performance to degrade?

Thanks to electronlover, that will be a veru good idea and feasible way in
low speed design, however, any idea about circuit, working up to 5 GHz?

Hi,

Input matching can be achived by using an inductively degenerated common-source LNA or by usign a common-gate input stage. In the former, matching is performed by means of an inductor connected to the source of the LNA transistor. This inductor, multipied by the transitor ft gives the 50Ohm matching. An inductor in series with the gate of the input transistor is neede to cancell the imaginary part of the input impedance. Tunning is made by adjusting the transistor ft (current) and finally adjusting the gate inductor (normally external) to make the input matching net resonate. Matching in this case is narrowband, though by using a low Q source inductor, the pass band can be broaden a little.

In the case of a common-gate input stage, 50Ohm matching is achieved by the value 1/Gm, being Gm the transconductance of the input transistor. In this case, tunning is performed by varying the transistor current. Matching in this case is broadband.

In CMOS, you try to use architectures with a minimum number of external componets to try not to have the necessity to have 50Ohm output impedance (high current consumption). Instead, you try to work with architectures suitable for a high level of integration, so matching can be done at higher impedance values, once you have passed the input LNA (the ellement that influences the most the receiver noise).

In some cases where you have no choise than going out your chip, you may addapt the the impedance seen by the driving ellement by means of capacitors and/or inductors (impedance transformers).

I hope to have given to you a quick idea of what you do in CMOS RF.

Regards

Humungus

yes, you are right to have inductors and capactance in match

however, still there are needs for resistor as amplifier loading , what about them?