The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl Design >> RF Design >> rules of thumb for EM interference https://designers-guide.org/forum/YaBB.pl?num=1289542964 Message started by louis de musset on Nov 11th, 2010, 10:22pm |
Title: rules of thumb for EM interference Post by louis de musset on Nov 11th, 2010, 10:22pm Helo every body, I have a question plz. I am a beginner in RF design and i have desjgned a 2,4 GHz LNA. The question is that at the layout level what is the rules of thumb to appky to avoid the electromagnetic coupling betwenn inducance-interconnetion-inductance ?? THanks for your answers Regards |
Title: Re: rules of thumb for EM interference Post by louis de musset on Nov 12th, 2010, 8:41am Helo every body, I am a beginner in RF design. I have designed a CMOS LNA at 2.4GHz and i have finished the layout. Unfortunately, i did not have access to an electromagnetic simulation tool and i must make the necessary precautions to prevent the magnetic coupling between inductances between them, and each inductor with other interconnects. So is there a method to avoid these couplings that can degrade the performances of the LNA after the fabriation. Or is there a design rule (rule of thumb) that i can follow to avoid this coupling. Thanks a lot Best Regards |
Title: Re: rules of thumb for EM interference Post by rfmagic on Nov 15th, 2010, 1:20am Hi, If you are asking about RFIC design It actually depends on many parameters but I will list the most dominant: 1. if you have an on-chip inductor - you prorbaly would want to isolate it as by it nature it radiate energy- thus I usually keep about 25um from each side of the inductor. 2. if you have 2 metal strips that carry high currents - they also acts like inductors and I ussalyl keep 10um between them. But the most important think to do is - simulate the suspected paths in electromagnetic simulator such as Momentum or HFSS to make sure that you dont hit radiation from other sources. |
Title: Re: rules of thumb for EM interference Post by louis de musset on Nov 15th, 2010, 2:56am Thanks for your response. The problem that i don't have access to an electromagnetic simulator that why i need some rules of thumb to avoid the electromagnetic coupling. I have a question concerning your response: if i have an input RF signal (power = -10 dBm) that come on the gate of a NMOS transistor (200um/0.25um) and continue from the drain of this transistor to the output stage. So here i can't make a distance between the two metals equal 10um like you note for me (distance between drain and gate is very short), so what should i do ?? Should i use other metal layer, or this the same things ?? The dc bias current of my LNA is 16mA. Thanks for all for helping me... It's very urgent!!! |
Title: Re: rules of thumb for EM interference Post by rfcooltools.com on Nov 15th, 2010, 9:49am louis, build yourself a couple of what if cases using two ideal inductors and resistance to de Q them to onchip metal equiv and a mutual "mind" to link them. As a coarse rule of thumb inductance is ~ 1nH per mm length and coupling coeficients will be quite a bit less than 1. For example two 100uM length thin metal spaces 10uM apart I would and 1uM wide I would try: Resistance = 100*dcmetalResistancePerSquare*2 (2 for skin effect) Inductance = 100pH Coupling k factor= .2 This is a ball park to get going. Try it and see if it makes a difference in the performance. http://rfcooltools.com |
Title: Re: rules of thumb for EM interference Post by rfmagic on Nov 15th, 2010, 11:30pm I agree with rfcooltools.com that you can include mutual inductance in your simulation to see how it impacts your performance but I would also incldue the coupling capacitor between the strips as in my experience may be more critical.you can do it by for first order simulation by adding a coupling capacitor where C=Eo*Er*a/d Eo=4 Er is the dialectric constant of your process (for 0.18um take 8.8e-12) a = the plate area d = the distance between the lines this is also a good starting point. but if you are looking for a "rule of thum" - when you have close nets like in your exaple of the NMOS transistor input and output - make sure that they cross in 90 deg - to minimize coupling and EM interference |
Title: Re: rules of thumb for EM interference Post by louis de musset on Nov 20th, 2010, 2:58am Thanks for rfmagic and rfcools.com for your response. So rfmagic you propose me to include the calculated value of the coupling capacitor in the netlist between each closed nets ?? For rfcools exuse me, can you plz re-explain me simply this phrase and what should i do: ''build yourself a couple of what if cases using two ideal inductors and resistance to de Q them to onchip metal equiv and a mutual "mind" to link them''. So if i take your example about the two metal lines, if i have these two metals lines in my layout. And if a DC bias current passes throught the first and a RF current passes throught the second and i want to simulate the coupling inductance between them using your rules what should i do?? Thanks very much for you both |
Title: Re: rules of thumb for EM interference Post by rfcooltools.com on Nov 20th, 2010, 12:09pm louis, You originally asked how to avoid coupling my response is does coupling really degrade the performance of your LNA's operation. One could approach this in several different ways. 1. using a RLCK extractor extract the layout and simulate (louis this is not an option correct) 2. Identify in the layout potential lines of coupling and try to model manually (this is my suggestion). Take your schematic and make a copy then insert the inductors and resistors based on manual measure measurements. basically for 2.4GHz a 100pH will likely be not of significance for noise and gain so you can start by ignoring 100uM to 200uM lines. But as the lines start exceeding that they can start becoming a matter of concern. So what I would do is identify the long runs from supply, input and output and try to model these. " And if a DC bias current passes throught the first and a RF current passes throught the second and i want to simulate the coupling inductance between them using your rules what should i do??" Put the inductors, resistors and capacitors (from rfmagic's suggestion) into your modified schematic and simulate the LNA over your original suite of tests. Does the performance degrade? Do you see peaking at high frequencies? If the answer is yes then try to figure out which new circuit element it is sensitive to. From my experience most of the issues of LNA circuit performance occur at the board to chip interface and typically on chip resistance alone is usually the performance killer. So spend some effort trying to model the bondwire coupling from supply and ground to input. http://rfcooltools.com |
Title: Re: rules of thumb for EM interference Post by louis de musset on Nov 24th, 2010, 10:42am Thanks a lot for your educational response and sorry for delay but a lot of work in this time. So i have understand the idea that you propose. for the inductor it must be trated in the same way as a metal wire ?? Thanks for you Louis |
Title: Re: rules of thumb for EM interference Post by rfmagic on Nov 24th, 2010, 9:59pm Actually, for inductros you should keep al least 25um form each side of the inductor an it is also recommended to add a guard ring to isolate it from substrate coupling. |
Title: Re: rules of thumb for EM interference Post by rfcooltools.com on Nov 25th, 2010, 12:14am Louis, Exactly you cannot span a distance with a conductor without incurring inductance. Therefore conductor be it a metal trace on chip, PCB trace , or bondwire will have self inductance. Two conductors will have both self and mutual inductance. Designing circuits at higher frequencies requires a comfortable understanding of electromagnetic. Rfmagic, I have an interesting experience with your 25uM rule of thumb on inductive coupling. Several years ago I was introduced by a colleague the notion of a coupled tank filter. But before I go into this and observation that I witnessed a few years prior: I was to design an IC with a few PLLs and LC tank VCO's which could operate simultaneously at the same frequency. to experiment I had an existing IC modified so that the two vcos could both be operating at the same frequency. One of the VCO's was controlled by a PLL the other was free running and controlled by an external voltage source. My goal was to characterize the spurs associated with closely tuned resonant circuits. when I observed the first intersting artifact associated with what is called coupled tanks dtermine if the distance (1mm in this case) was sufficient enough to be safe for this application. while measuring spurs I was looking at phase noise of the PLL driven VCO and then for some reason I switched the input of the spectrum analyzer to the voltage controlled VCO. As I moved the center frequency of the open loop vco closer to the PLL controlled VCO I observed the the phase noise of the open loop VCO started to improve as it got near the PLL driven VCO. Just to put it into perspective the two VCO's were physically a millimeter apart. And yet I was observing this phenomena. So I I thought it may have something to do with the voltage source. Then something even more interesting happened. I set the open loop VCO above the PLL locked VCO (both have a positive KV) and then I disconnected the voltage source from the open loop VCO while I left the PLL controlled VCO locked and running. As the charge leaked out of the open loop VCO control line the VCO center frequency started dropping steadily until it was near the Locked VCO frequency until its frequency suddenly snapped to the locked PLL frequency, for a short time and during this time the the open loop VCO phase noise improved while the PLL locked VCO phase noise degraded around 3dB. It remained locked on the same frequency as the PLL locked VCO frequency for several seconds until it finally got released and continued the reduction in frequency trajectory that it was on before being captured by the PLL locked VCO. coupled tanks.... if two tank circuits are tuned to the same frequency and have a high enough Q then the coupling between them can be sufficiently coupled with even the weakest k factor. in fact if the the Q is around 100 or more and the tanks are identical then a coupling factor near 0.01 will be enough to transmit power at almost no loss. If you have access to a simulator such as cadence or ADS try it out for yourself. Take a port connected to a LC tank withe high Q then replicate this circuit and change only the port number then see what the coupling factor is needed to sufficiently couple the power delivered from one tank to the other, you will be surprised at the result. You will likely see a butterworth filter delivering perfect power with suprisingly little coupling. its real.. This is not a novel concept, ( maybe to you or I ) but if you crack open a satellite tv LNB case and take a look at the PCB board in it traces are not physically connected instead there are tiny resonant circuits that make up the spaced apart the act as filters. These structures are alive and well in the world. so my point is this... understand the nature of the medium your design in and rules of thumb are either irrelevant or un-encompassing. http://rfcooltools.com |
Title: Re: rules of thumb for EM interference Post by Frank Wiedmann on Nov 25th, 2010, 12:29am rfcooltools.com wrote on Nov 25th, 2010, 12:14am:
See for example http://home.sandiego.edu/~ekim/e194rfs01/filterek.pdf (section "Coupled Line Filters", starting on page 8). |
Title: Re: rules of thumb for EM interference Post by louis de musset on Nov 25th, 2010, 12:56am Thanks for you rfcool. What you say about designing high frequecy circuits is very corret and my problem that i don't have good understanding of the concepts of electromganetsim. I like to simulate you example in cadence, it is very interresting but how can i see what the coupling factor is needed to sufficiently couple the power delivered from one tank to the other?? what kind of simulation make this in cadence ?? Sorry for asking a lot of questions but i am a beginner in this domain and i don't have any one to helo me. Thanks a lot. Louis |
Title: Re: rules of thumb for EM interference Post by RFICDUDE on Nov 25th, 2010, 5:31am Hi Louis, You can use inductors with mutual inductance to model coupled tanks, but there is no easy way to relate this directly to coupling in a specific layout situation. EM solvers are the only type of tools that can provide answers to your questions. EM solvers take a physical layout input (say two inductors), physical description of the metal and dielectric layers and definitions of ports that are the electrical points that you want the solver to provide a solution. There are a variety of EM solvers depending what you are trying to model, the frequencies of interest, the access you have (or don't have) to different solvers and computing power. Nearly all of the solvers break the physical problem down into many little sections where Maxwell's equations can be numerically solved. The output of the solver is usually a N-port matrix at the N ports you defined in the layout; although, static solvers provide a L-R-C matrix that is usually easily converter to a L-R-C SPICE netlist. The N-port matrix (s-parameter) can be used in circuit simulators such as Spectre, ADS, etc., but if you want an actual circuit model then you have to make a model yourself to fit the s-parameter results. Making accurate circuit models is time consuming, so it is the engineer's decision how best to spend the time. So the flow is Layout into EM tool EM simulation with N-port output N-port simulation element and/or extract compact model Evaluate in circuit Make changes to the layout (if necessary) Repeat There are simpler analysis techniques for problems like bond wire coupling, and you can always use some basic estimation of coupling from different effects. But in the end, if you want or need to account for complexity or all distributed parasitic effects then you need a more general EM solver. Some of the EM Tools are Static Solver - Fast Henry and Fast Cap - Ansoft has some static solver tool Planar 2D and 2.5D solvers - Agilent Momentum - Sonnet - There are others ... Full 3D - Ansoft HFSS - Agilent EMPRO - There are others ... |
Title: Re: rules of thumb for EM interference Post by Frank Wiedmann on Nov 26th, 2010, 12:18am For a start, you might want to try the free Sonnet Lite (see http://www.sonnetsoftware.com/products/lite/). |
Title: Re: rules of thumb for EM interference Post by louis de musset on Dec 7th, 2010, 4:39am Thanks for all your comments and for your help. I have maked models for the interconnects in my layout and inserring the models in the schemtic and re-simulate to see the effect of the parastics of critical interconnection specially in the path RF. but i just have a problem to understand how can i model the mutual indictance between two inductances. for example if i have in the same layout two inductors (3.4nH and 600pH) and i want to know the mutual inductance between them and i don't have electromagntic tool so how can i built the model of the mutual inductance between these two inductors?? i don't think it 's the same thing as two parallel line no?? thanks for your reply Using ASSURA - RLCK ?? i don't think that this tool tell us about the coefficient coupling between two inductors no ?? plz i make mistake tell me Thanks |
Title: Re: rules of thumb for EM interference Post by vp1953 on Dec 7th, 2010, 5:24pm Hi Rfcooltools, Code:
Your experiment is very interesting and i am very much intrigued by your observed results. 1. the node where your VCO's control voltage line was connected to the external voltage source had a large capacitance ? 2. were there any circuitry between your PLL and the VCO (in the intervening 1mm spacing) or was it all just blank silicon except may be metal fill layers for passing DRC density checks. From your description, it sounds as though that you are ascribing the slow change and the pause in the oscillation frequency of the VCO to coupling between inductors - is that correct? Most inductors today have a patterned ground shield at the bottom and also a shield on the sidewalls so much so that these layers effectively block all flux lines and very little leakage results. I would have thought that the coupling factor between two inductors would have been very small as a result - would be very interested to know what sort of coupling factors result if you have done any calculations on this |
Title: Re: rules of thumb for EM interference Post by louis de musset on Dec 7th, 2010, 5:31pm If you want i will change my question. The idea is that i want to make electromagntic coupling between two inductors, i want to ''sense'' the power in the main inductor (in the LNA) electromagnetic coupling. So i want to place another indcutor (one pin to the ground ana the other one on the air) and measure the voltage in this second indcutor and this voltage will be proportionnal to the existing current passing in the first inductor in the LNA. SO IS my idea is reasonnable?? I think that this is the idea of the transformers??? no?? |
Title: Re: rules of thumb for EM interference Post by rfcooltools.com on Dec 11th, 2010, 1:34pm Louis, What you are asking is difficult to get accurate without a em simulation tool since coils have their own tightly coupled mutual inductance. As a result the coupling to a second coil will be different than two separate parallel conductors as you correctly stated. An yes RLCK will not give the coupling between inductors if the inductors are a parameterized cell (pcell). This is due to the way the LVS sees the inductor as an RLCK ignore region usually defined by an inductor recognition layer defined by the fab in the LVS deck. Without an em tool you could do the following: 1. if you have rlck then flatten both inductors in the layout and remove the inductor recognition layer and place a metal resistor before the pin to keep the LVS from seeing the inductor as a short (without the inductor recognition layer LVS will see the inductor as one piece of metal shorting out two or more pins and thus you will have to trick it into being LVS clean.) . After all this RLCK will be inaccurate in predicting the inductor values due to limitations in how it approximates coupling, but might give you insight into what the coupling between two inductors might be. 2. the reason I gave a lengthy discussion on coupled tanks was to explain that coupled resonators at the same frequency of resonance need only week coupling to transfer power. Why is this important? because if the resonant frequency of the two tanks are sufficiently far apart then it doesn't matter as much as to what the coupling is between. Check this by biasing up the entire circuit and placing an AC current through the first of the two inductors. Where does the AC voltage peak for that inductor? Remove the AC current source and do the same for the second where does it peak. Then try to get a bounds on what mutual inductance will give you perceptible performance degradation. Off frequency resonances will need much stronger coupling to produce undesired results. Try the experiments above and report the numbers or wave results and we can see if they can be interpreted further. Also include Q vs frequency of each inductor as well as inductance vs frequency of each inductor. and dimensions and spacing of each inductor. http://rfcooltools.com |
The Designer's Guide Community Forum » Powered by YaBB 2.2.2! YaBB © 2000-2008. All Rights Reserved. |