Keys to Control Noise, Interference and EMI in PC Boards - Hartley

[Music] if we were here for a week I don't think I could teach you everything you need to hear so I would encourage you to read this first book is I believe the the most important book published in the last 30 years it talks about how energy actually moves in circuits as opposed to how we think energy moves in circuits and I'm going to talk about that to some level today I'm going to go through some of the material in this book and give you an idea of what we're talking about because in order to control am I you really have to understand where the energy is in a circuit board and it's not where most people think it is so we're gonna discuss that at length but this is a really good book I'm going to pass it around just have a look at it one of the in the introduction to the book this by the way the book is obviously by Ralph Morrison who sadly passed away about a month and a half ago he was 94 he lived a full life he published twelve or fourteen books in his life an interesting man and the truth is for the first probably 10 or 12 years of the time I was teaching I was teaching people the rules that I had learned and some of the rules worked and some of them didn't and when I met Ralph I understand to understand the truth about what's going on and I stopped trying to teach rules and started trying to teach people why things happen if you know why stuff happens then you can figure out how to avoid it under any circumstances you know you don't have to everything we're gonna go through some rules today don't get me wrong we're gonna go through a lot of rules but all the rules that I'm gonna talk about are based on what is in Ralph's book I once I figured out what's really going on suddenly I stopped having problems altogether I haven't had a signal integrity area my problem in the last fifteen to sixteen years and the reason is because I understood raus teachings believe me it's an incredibly important book in the introduction to the book he talks about going to help some people in a facility and I forget the details it's been so long since I read it their facility was a metal building it had metal walls and the metal walls stopped about a foot above the ground and there was concrete blocked the rest of the way down every time there was a lightning strike all the machinery in this building would reset every time and occasionally people in the building would feel the energy move through the building I mean that would never hurt anybody but they said they could actually feel it so Ralph went there to talk to him and by that time they had added multiple lightning arrestors around the perimeter of the building and they had a big honking like triple lot wire going from the Lightning arrestor to a stake in the ground but it didn't solve the problem at all and the reason is very simple the energy when it hit that building those metal walls were a much lower impedance to the ground than the path of the Lightning arrestor even though the metal walls didn't go all the way to the ground the capacitive coupling from the end of that wall into the ground down through the air was a lower impedance path than the heavy wire from those lightning arrestors and so the energy still hit the building and every time it would get down to that floor level as it was coupling in to the ground the energy would spread out in a giant wave that went across the floor of the building and that's why all the machinery was said and once Ralph explained to them what was going on it was suddenly boom light bulbs went off and they knew what to do to fix anyway it's a great book I would encourage you to read it PCB designed for real world EMI control Bruce arson boat probably about a 10 15 year old book Bruce is one of the brightest people I know he worked for IBM for many years he was he's retired from there he's a senior fellow with the I Triple E as well as at IBM the man is brilliant and this book is a genius piece of writing it's it's filled with real world real world issues and how to solve them principles of power integrity for PDN design simplified by Larry Smith and hi there and dr. Eric bogus and I'll trip at some point you guys will get a nice laugh it should be fun and Eric bogus and everything Eric writes the word simplified is at the end because he writes stuff in a simple understandable manner he's a great guy and anytime you have a chance to hear Eric speak I would encourage you to do so he'll be here this week doing a keynote and marvelous guy this books about a year and a half or two years old it's fabulous I mean it's absolutely fabulous the future of emi control and signal integrity control as we move into the gigabit domain in printed circuit boards the key to controlling many of the issues will be in proper power bus design and i'm going to go through some of the things that are in this book today and some some of many of the things that are in this book are things i've understood for years so i'm gonna go through them not necessarily because they were in the book but because the book simply verified that the stuff i understood was basically right but it's it's a great book digital designs for interference specification by david Terrell and our Kenneth Keene and Ken Keene and was running around the United States in the 1980s telling people how to solve EMI problems long before most people understood at all what EMI even was much less how to solve it and he he and David Terrell ran a test facility in New Jersey for a number of years and they helped people for many years 20 30 40 years figure out why their units were failing EMI testing and figuring out what to do about it and in the process they gathered a lot of information about why things go wrong and how to fix them and that's what this book is it's a compilation of their thirty years of running an EMI test facility and it's a phenomenal book electromagnetic compatibility by Henry OTT Henry wrote the first book on my control back in 1976 I forget the title of it when he wrote this book he basically took that first book and rewrote it to bring it up to current thinking and he added it was about an inch thick book originally and he added another inch to it so it's about a two inch thick book it's a that makes a great doorstop when you're done reading it it's over a boat anchor it's it's a heavy book but it's a great book and the information that it is phenomenal Henry did the keynote in 2014 at the I Triple E EMC symposium and during the keynote he talked about in 1989 he and the I Triple E sent a questionnaire out to colleges and universities across all of North America Mexico the United States and Canada and they asked us all the engineering schools they asked a simple question do you teach anything about EMI control to the people in your classes and they got the aunt then the answer yes came back from five schools out of thousands and thousands of schools five and in 1980 or in 2014 when he did the keynote he said we sent another questionnaire out about a year ago and it's not much better now there just aren't that many universities that teach anything about EMI control and honestly I think part of the reason is I'm not sure that all the professors who teach at these universities understand it you have to understand something to be able to teach it and I'm not sure many of them do we'll talk more about that this is a great book though and I asked Henry first time I met him about 12 years ago he attended one of my sessions and I asked him after the session I said Henry you're credited with writing the first ever book on EMI how did you learn and he said two ways we were stupid often and we learned from our mistakes we made a lot of them at Bell Labs and at AT&T and we learned from them and we figured out why things went wrong not just what but why and secondly by reading Ralph Morrison's 1967 book same guys the first book grounding and shielding he said I learned more about how to control fields through that book than anything I had done now there's others pardon me other stuff today that picks up for a Ralph's row Ralph's book left leaves off but believe me it's a spectacular book it's it's a good book for understanding as his his first book they're excellent treatises so food for thought read books not ICF notes I start every single presentation I do with this with this qualifier circuit application notes produced by IC manufacturer should be assumed wrong until proven right that is a quote from Lee Richie I first heard him make that quote about 25 years ago in a class at PCB West and when he said it I snickered and he heard me and he went who was that and I said I said me he said do you disagree with me I said no I agree with you entirely that's why I laughed because I thought I was the Lone Ranger because none of the engineers I work with believe that they all believe oh it's an icy good company they must know what they're talking about yeah sure they do if you talk to Dan beaker who was here last year and the year before I don't know if he was in Germany last year he was here the year before if you talk to Dan beaker who works for NXP semiconductor he is an app engineer and asked him are nxp zapped notes do they fall into this same cube he'll say absolutely and you absolutely shouldn't use them without proving that they're right why Dan what is wrong with the app notes don't engineers understand circuits of course they do they understand circuit theory what they don't understand is proper circuit board layout they haven't a freaking clue pardon my language about circuit board layout 99% of the engineers I've met do not jet board layout though they read things like no 90-degree corners and they think oh boy that's dangerous I know not that you should use 90-degree corners but they aren't the problem that people think they are there are many things that app notes talk that are wrong split the ground here here and here I've seen app notes that split ground down into three or four sections around an eye see what the hell are these people thinking we'll talk about that as we go through the material the agenda the impact of circuit frequency on design methods and strategies circuit transmission lines and electromagnetic fields is the second section concerns and types of electromagnetic interference PC board design methods to control Lea my power distribution to control the signal integrity and EMI PC boards stack up to control signal integrity and EMI and the i/o structure designed to help control EMI how should you set up the i/o structure one of the one of the leading causes of interference as anybody who's ever run down an EMI problem can tell you our cables how do you keep cables from resonating and radiating and we're going to talk about we're going to talk all through this stuff about how to do that but within this section we're going to talk about how to set up the i/o to help even further prevent the problem and then we're going to finish off the date going through just a few mechanical issues that tend to exacerbate EMI and what you can do about it we will start with the impact of circuit frequency and I'm going to go through some of this early stuff quickly because it's it's fairly just routine from the 50s to the middle 80s or beyond circuit boards could be laid out in any manner that you chose up until the mid 80s everybody I know was connecting the dots on the circuit boards I know that's exactly how I laid out circuit boards I started I was actually a EE originally and I was asked to help design some circuit boards at a company where I work because I had a little bit of experience with it when I say a little bit I mean a little bit but I helped out and the truth is I didn't understand correct board layout but back then you didn't need to as long as you could design a board that was manufacturable and could be assembled without problems that was pretty much all you really needed to worry about today things are much deeper than that yet seldom had problems the reason because frequencies were so low back then that everything was a lumped element regardless of the size of the board or how or how long the lines were I did a circuit board in 1984 that was for a vmebus it was 14 by 16 inches had 200 ICS on it was hand taped by the way and that particular layout had traces that ran all over the board and we never had the slightest problem with it because everything on that board based on frequency was a lumped element things that we put on circuit boards are a lumped element at low frequencies and there and I'm going to define what these two are they're a distributed element at high frequencies we will talk in a minute about what that means back in the 70s we actually did stuff like this and they worked in fact a company I worked for in the 70s had an equation that they'd come up with based on the number of square inches of the board how many jumpers you could put on before it was more cost-effective to go to two layers and we did this stuff all the time and the amazing thing is these works chances are likely there is a ground trace wandering its way around this board connecting all the ground points and in spite of that it worked you do that today chances are it won't even while chances are it won't do anything but anyway we even got away with stuff like this back in the 70s when we breadboard circuits this is exactly how we did it we either soldered wires on or we wire wrapped the things and this was a common practice back then the very first board I said I was initially a EE I was actually initially a technician for several years then they became a EE and then eventually a printed circuit designer in my early years is an R&D tech the very first circuit I ever built look just about like this didn't have any tube sockets on it but aside from that it looked just about like this one and it worked amazingly enough so thinking in terms of propagation time and rise in fall time or slew rate of IC outputs when a output an output is a lumped circuit when you have long rise times compared to propagation time in other words when the time it takes to move down the transmission line is short compared to rise time you have a lumped element circuit and the reason is because the energy moves into the transmission line and it will travel the entire length of the line before it's risen very far at all and the end result is that the energy in the line tends to look almost like the water in a bathtub as it rises and any impedance discontinuities you have don't matter because the reflections off of them just get absorbed by the energy crosstalk is a non-issue which we'll talk about why as we go through this nothing really mattered back in the day because everything was a lumped element distributed circuits have long propagation time compared to rise time long propagation time compared to rise time meaning that the circuit will rise to its peak level before it can propagate the line and the end result of that is that the energy becomes a wave that's moving down the line at will talk about how fast in a moment moving down the line at a very high speed and when it hits impedance discontinuities or hits the end of the line if it's an unterminated load or whatever happens you get energy bouncing around on the line creating ringing or reflections you also this is the point where you start to see crosstalk and EMI and everything you can imagine starts to go wrong once we become a distributed length element and it all comes down to rise and fall time clock frequency has absolutely nothing to do with it clock frequency when it comes to problems is totally completely irrelevant okay Rick and why do we see emi failures and harmonics of the clock because that's where the peak energy levels are in a digital square wave and we're going to talk about that but it's not because of the clock that we see the problem it's because of rise and fall time engineers it's a it's a it's about the rate of change of energy with respect to time and engineers refer to this as di rate of change of current with respect to time DT di DT and DV DT rate of change of voltage with respect to time now that's not actually true either which we're going to go through but that's how people discuss this di DT DV DT rate of change of energy level with respect to time that's what causes problems I was in Australia in 2006 doing up I've been there a number of times but doing a public presentation similar to this and during the morning break a young gentleman came up to me and he said Rick I worked for an elevator controller company and we have always designed our circuit boards a certain way and they've always worked we clocked by the way he said at 500 kilohertz 1/2 a megahertz which is almost you can only see that I mean we're talking about slow right and and we've never had problems and suddenly we build a batch of boards a few months ago we are having trouble getting them to work and we're failing am i testing and we don't have a clue why because they ignored rise time they thought clock frequency was the important factor the eye sees on their board went through a die shrink and went from some rise time to a faster rise time and as that happened suddenly the lines on their circuit boards became went from lumped to distributed length elements and they started seeing ringing in crosstalk and coupling of energy into Io that caused EMI issues they had an endless array of problems and he said what can we do about it and I said just pay attention for the rest of this class and you'll leave here knowing exactly what to do the next time I went to Australia the same guy was in one of my classes and I asked him were you able to solve the problems he said absolutely he said we solved them within a matter of a week on every board we had and we're having no problems at all I ran into the same thing when I first went to the telecom world in 1996 I had been into the company maybe three days and then a double II came to my office and said Rick people tell me you know a lot about signal integrity and I said yeah little and he said would you come with me to the lab I want you to see something and on he was either a logic analyzer and oscilloscope he had a picture of a waveform that was ringing like you can't even imagine and I said wow that's pretty severe ringing and he said what's causing it and I said chances are you haven't impedance discontinuity a major line and it's not terminated properly and he said terminated what the hell does that mean I'm thinking oh boy turned out the same thing happened to them that happens the elevator company except this was many years before they purchasing got went out to buy some particular IC and FPGA or a process or whatever and the distributor they bought it from said oh by the way there's been a dice ranked and the first thing guy said what's that mean and the guy at the other end said I don't know but they tell me it's a lot faster part is that okay and the guy said I don't know let me check with the engineer so he got the new data sheet for the part gave it to the engineer the engineer realized that everything in the IC was going to go faster quicker prop time through the IC all of these things were going to be reduced he thought oh this is good because they were right on the edge of a race issue anyway and all of a sudden with a shorter prop time man life is good what he didn't think about was rise and fall time and they put these things on that board and they built 50 of them and not one of the 50 worked and we patched one together enough to make it work and then did some in my testing and it failed EMI testing miserably the moral of this story is by the way it took us several weeks to completely redesign the board to make it so it would function properly by the time we got the prototype up and running and through debug and made additional changes to the layout and then got back into production this was in the early fourth quarter that this happened we did this particular board was used in every unit this company shipped whether it was a high-end router or a switch or whatever it was went into the telecom or networking office every unit had this processor board in it and we shipped zero product in the fourth quarter this company was a darling of the telecom industry prior to this event when news hit the market in 1997 that we had zero shipments and zero profit in the fourth quarter of 1996 the stock fell from $65 a share to $6 a share the company struggled to get money after that and they struggled to stay alive and finally in about 2003 or four they died completely now the drop of the telecom world in 2002 didn't help him at all trust me that was a big part of why they failed but they failed because they never the biggest reason is they never recovered from that one event the point is everything you design you should design as controlled impedance whether you buy it as controlled impedance or not and we're going to talk about that as we go through this stuff today every board should be does you don't have to buy a controlled impedance board but have it ready so that if you have a problem you can add termination devices which we're also going to talk about in the appropriate places and eliminate whatever problems you're having we'll discuss this much more as we go through this how extreme is the effect in the early days of five megahertz clocks and 100 nanosecond rise times a circuit board was a element until things on the board exceeded ten feet in length well I don't know about you but I haven't outed a lot of 10-foot traces in my life so we didn't worry too much back in those days about how things were gonna function even in even by the mid 80s when things were pushing 10 20 30 megahertz with tens of nanosecond rise times circuits had to be a foot and a half to three feet long before you really had a problem and the only people even in those days that really had issues god are you talking to us the only people in those days that really it can you guys see okay to take notes there you are right okay the only people in those days that really had problems for people that really and truly we're dealing with with very high frequency stuff today today's rise times almost every ICU by today process their FPGAs memory devices whatever have rise times in the 300 to 700 picosecond range five hundred picoseconds is a normal rise time for ICS today that's pardon me that's nominal for a lot of I sees at today's rise times things that are longer than 3/16 of an inch to two and a half two and three-quarter inches are considered distributed length elements 100 nanoseconds to two nanoseconds even if you have a microcontroller operating in the order of two to three nanosecond rise times microcontrollers tend to be slower if your lines exceed two and a half to three inches in length you still need to pay attention so it all comes down to line length versus rise time based on that actually limit one more slide and then we'll get to some information that's useful the problem actually begins when the rising edge when the time for the rising edge a quarter of the rise time sorry proper let's try this again when prop time exceeds 1/4 of rise time you're in harm's way that's when the circuit is going to start to see reflections and ringing that's when you're going to start to see an increase in crosstalk EMI and all of the things that bug us once the line is long enough that it's longer in terms of time then the rise time of the part if you've done nothing about it you can pretty much kiss it goodbye because you are gonna have problems big problems so you have to pay attention to rise time where do you get information on rise time is it in data sheets Rick of course not that would be helpful why would they do that it's in models ibis and spice models contain information on rise and fall time they have to I mean they can't simulate transmission lines without that data and they're just ASCII files you can open them up sometimes it's listed as a rate of change of voltage with respect to time dv/dt and then you have to calculate how high is it going to rise from zero to what level and then from that you can calculate time so you may have to do you know a small amount of math but you need to know rise time because you shouldn't design a circuit board without knowing what you have and what may be coming if you're designing stuff that's only gonna last a year before it fails like a computer and or two or three and then somebody's gonna throw it away and buy another one you know chances are you're not going to outlive the existing logic that you have but if you're designing products that are gonna be around for 5 10 20 30 years like avionics products or things of that type and there's a good probability that you're gonna see faster parts in the future because of dye strength then you need to pay attention and you really need to think about all these things speaking of all of that does anybody in here control line lengths in buses in other words match routing lengths and data buses and address yeah some of you you're not supposed to match lengths you're supposed to match time it's not about length it's about time and I would encourage you to start thinking about circuit boards and transmission line in terms of propagation time not length the reason is outer layers propagated at different rates and inner layers and if the line is heavily loaded it'll propagate at a much slower rate than one that's lightly loaded all this stuff impacts prop time and it's really prop time that you want to match we'll talk more about that as we go so how do we calculate when a line reaches a length that's a distributed element well you start by calculating prop velocity propagation velocity is the speed of light in free space divided by the square root of e sub R epsilon sub R of the circuit board in which the transmission line is traveling so if it's fr for for example and I'm going to use the number 4 for e sub R even though fr for Z sub R can be anywhere from three point nine to four point six or four point seven its there is no one number for fr for Z sub R or for any material Z sub R for that matter which is another thing you need to have information on you can't do accurate impedance calculations without accurately knowing what East of R is of the board you're designing so you need to know all these things to be able to do this correctly assuming because I can do the math in my head square root of four is two co2 genius I am greater two divided into the speed of light is half the speed of light propagation velocity in an fr4 board on an inner layer trace sandwich between dielectrics layers of fr4 is going to move at about half the speed of light speed of light is 11.8 inches per nanosecond 300 millimeters per nanosecond so propagation velocity is about just under six inches per nanosecond an fr4 or just under just about a hundred and fifty millimeters per nanosecond those are typical propagation velocities in a circuit board you add to that rise time and remember I said when when the length of the line exceeds a quarter of rise time you're in harm's so if we add the constant 0.25 and AD rise time to that equation we will get a number of how long a line can be on an inner layer before you need to pay attention to how it's routed its impedance how its terminated what it references all of these things speaking about what it references what I mean is is it referencing a ground plane or a power plane is it okay to reference a power plane sometimes it's okay under the right circumstances and we're gonna talk about those circumstances today sometimes it's just fine other times it's the worst thing you can do and we're going to talk about those conditions as we go through this on an outer layer you basically have the same hi there I am gonna trip over this you you have the same equation divided by the square root of effective relative permittivity effective relative permittivity is a blend of the e sub R of air above the trace and the high DK of the board material below and if you look at outer layer traces because some of the field energy is in air some of it's in the high DK of the board you end up with a blended e sub R called effective relative E sub R that's in the order of three to three and a half when dielectric constant of the material is in the order of four 4.1 4.2 something like that this one I used 4.3 as an approximation and when I looked on a field solver for what e sub R would be for a 50 ohm line it ended up being somewhere in the three to three and a half depending on all the characteristics of the line and you can imagine you take the square root of that divided into the speed of light it's going to be a vastly different number energy propagates on outer layers about 15% faster than it does on inner layers which means if you're matching length you should make those 15% shorter because what you should really be doing is calculating prop time and matching times it isn't about length it's about time yes the outer layer propagates at a faster rate correct no you need to make them longer sorry 15% longer yes thank you see I'm glad that somebody here can actually think yes 15 percent longer that's correct and it's not 15 percent it's in that vicinity yes that is exactly what I'm saying everybody hear that question if you have a slow clock frequency but have fast rise times with long traces do you need to pay attention and the answer is absolutely yes absolutely yes it isn't about the clock it's about rise time every single thing that goes wrong happens during the rising and falling as once a signal has risen and has leveled off nothing goes wrong during that period everything just sits there and hums along everything that goes wrong goes wrong here and here all of it all EMI all crosstalk all reflections all skin effect and loss tangent losses for those of you operating well up into the gigabit region everything that goes wrong goes wrong during the rising and falling edge yes [Music] yes and you you you would have to unless you could slow down the slew rate but then that would probably affect circuit performance so what he was saying is even at a hundred and fifteen kilohertz frequency rate they were seeing problems because the the rise times the edge rate if you will the slew rate of the signal was so fast that in longer lines they were having problems even though things were clocking very slow and that's exactly the experience the elevator guy said exact same experience yep thank you that's great that's great information thanks for sharing that this is an equation to calculate effective relative permittivity in case you don't have a field solver and we're gonna talk about field solvers in a little while I believe I've got information in here on them you would use this equation if the width of the trace divided this is an outer layer trace now if the width of the trace divided by height above the plain was less than one you would use this equation there's a simpler equation that you would use if W divided by H is 1 or greater and I can tell you right now if you're doing an fr4 design and you're targeting 45 to 60 ohm impedance W divided by H will be greater than 1 it will be greater than 1 which means if you want to calculate effective relative permittivity you would use this bottom equation it looks complex it's really really simple I know it looks a little bit hairy but it isn't at all trust me it's very simple and then you plug that into the equation previously shown and get this this is a table that I put together many years ago of how long lines can be based on rise time back in the days of 5 nanosecond rise times we could have lines that were seven and a quarter to eight and a half inches long enter an outer layer before we saw problems and that's why people got away even in the early part of this century very often people got away without having any problems at all because boards might be small enough that their traces were under those and they were lucky enough to have five nanosecond rise time devices as you go up and freak as you start to get faster and faster you can see these line links get shorter and shorter today just about everything we deal with is in this range right here these bottom three now they're not 110 100k echolert a you know there's CMOS devices but the reality is most devices that we use today processors FPGAs and memory fall into this category right here so you can see lines that are longer than a half inch to an inch if you're not paying attention and not doing something correctly with that line you're begging for troubles so understand the impact that that has because it's huge most memory devices today have typical rise times in the order of 250 picoseconds they're minimums are down under two hundred picoseconds and they're maxes are up around three they are blazingly fast devices these things are built for speed because most people today are using them in fast circuits if you happen not to be well you still have to treat it as though it is so food for thought what about analog circuits with analog circuits lumped versus distributed length is a function not of rise time but of the frequency of the circuit NDK meaning wavelength and DK it's 1/12 wavelength in the dielectric remember if for example if I have a 1 gigahertz circuit wavelength is going to be eleven point eight inches in free space in an fr4 material it's going to be half that about 5.9 inches so as long as my circuit as long as my line links are shorter than a twelfth of that then I would be okay the equation to calculate this it's called L critical and why is it frequency dependent with analog circuits because at any point in time an analog signal is a sine wave at any point in time a digital signal is not a sine wave and but an analogue signal is so it's all about the frequency of the wave at the time of the calculation so l critical is the speed of light divided by frequency times one over the square root of epsilon sub R board material decay times 1/12 and that gives you the lker the length the critical length for an analog circuit on an inner layer of an of a particular circuit board strip line means inner layer trace one that references two planes on an inner layer at a gigahertz for example a line can be up to four hundred and eighty five mils long before you really care about what's going on with it when I first joined the avionics world in 1990 I went to work for Goodrich aerospace and our chief RF engineer a wonderful guy whom I grew to love dearly named yeonu's but who I also found to be a pain in the butt Yanis told me everything you're right he was he was from Europe and he had very thick accent he said everything you route will be controlled impedance line no matter how long 50 ohms well guess what be honest we had lines Abba would get to that in a minute the reality is he was wrong and once we figured this out we were able to work around that but anyway a line that's longer than this you have to be attention to one that's shorter you're okay for a microstrip which is an outer layer trace basically it's the same equation but you plug in effective relative permittivity and for outer layers it can be as long as about five hundred and thirty or so mils somewhere in the order of 13 to 14 millimeters before you care about it what about a square wave if a analog signal is essentially a sine wave what is a square wave well the answer is it's a series of sine waves that are algebraically some they're called harmonics and they're algebraically summed to create a square wave shape if you look at a square waves amplitude at any point in time right here for example it is the sum of this energy plus this energy Plus this Plus this and all of that energy sum together creates that amplitude right no matter where you and there's I've only shown the first first third fifth and seventh harmonics here but the reality is they extend well beyond the seventh our mind and as you get faster and faster rise times guess what happens to the harmonics they increase in amplitude which means that if you have a fast edge rate you're going to see higher frequency energy that can and will exacerbate EMI problems well up into the gigahertz region that's why these elevator guys even though they even though they got one or two of their boards patched together enough to work it failed EMI testing same thing we ran into in the telecom world one even though we got one of the boards to work it wouldn't pass EMI testing because it was it was just a nightmare of distributed lines that weren't properly controlled anyway square wave is a series of sine waves the frequency is called the maximum pulse frequency the frequency of the rising edge is the one that's important the frequency of a transmission line is not the clock it's the frequency of the rising edge and it is a function of rise time divided into a constant point five if you look at doctor Howard Johnson's eye speed or a high-speed black magic handbook that he published in the 90s he will tell you this number should be 0.35 and the reality is if you're looking for the knee frequency of the square wave of the rising edge yes it's 0.35 the reality is though there is energy in the square wave beyond the knee that's detrimental to EMI so from an EMI standpoint what we want to do is calculate the maximum frequency of a square wave based on the car instant point-five and what I'm saying here is that from an EMI standpoint if you have a circuit that's clocked at 50 megahertz but has 0.5 nanosecond rise time devices 0.5 divided by 0.5 is 1 gigahertz that means you have energy in every transmission line every time they fire that goes from 50 megahertz to a gigahertz and all of that energy across that spectrum is capable of creating EMI problems but the frequency of concern in a transmission line is the rise time frequency 0.5 divided by rising when it comes to power bus design we need to design the power bus to be low and impedance from 50 megahertz to 0.5 divided by rise time whatever that number is so we're going to talk later about methods to do that why digital rise times get faster we've already talked about this because of dice rank and why does that happen because we're greedy because purchasing is allowed to make decisions worse decision any company ever made was giving purchasing any power at all we eliminated that problem when it came to the circuit boards in the division of l3 where I worked because we had such a disastrous problem because purchasing bought boards from the cheapest vendor they could find and we had oh my god problems in the field that were nightmarish and we ended up solving them by getting higher quality boards and replacing these cheap boards that purchasing had steered us toward once that happened and I was I was lucky enough to be the one to figure out it would trust me it was not a challenge but I was lucky enough to be the guy who they said yeah he solved the problem so the president of the company called me in her office and said Rick what do we have to do to prevent this from ever happening in the future and I said take away purchasing toys and she said meaning and I said put board house selection in the hands of the quality Department the engineering department I named a couple others I said when we go out to inspect a shop will take purchasing along they can go and they can have Danish and nod you know that's okay they're allowed to do that because they can't do anything else useful but it's our company's beating up I see companies that drove this problem in the first place it's got to be cheaper got to have lower prices got to be cheaper it's got to be cheaper cheaper cheaper cheaper well they can only take so much fat out of the hog and then they've got to find another way to make it cheaper and they discovered that first off there are two reasons they did it one because they needed faster rise times because of increased frequencies people's clock rates were going up and obviously there's a relationship between rise time and clock frequency you don't want rise time to exceed more than about fifteen percent of the total clock period because if you do you're going to start to have timing problems potential timing problems so they were shrinking died partly for that but a lot of the eye sees they were shrinking they were shrinking purely to get a smaller die so they could get a lower cost more died per wafer lower cost die lower-cost I see to shut purchasing up for maybe six months and then they'd start squawking again the producing department at l3 where I worked was so bad about this the company that did all of our board assembly was in the same city as the headquarters of this division of l3 the purchasing department told them at the end of every single year you have to take 10% out of the cost of all the board assemblies for next year well there's only so much 10% you can take out and then you've removed all of the fat and now you're removing lean meat there's only so much you can do and what they did was drove poorer quality because these guys started taking shortcuts and that's what fab shops that are cheap do they take shortcuts they push drill life longer they do everything they can to reduce costs so they can reduce price and still make a profit there are no free lunches cheap words are not the way to go so for wedding source I'll get off my soapbox I'm done thank you very much okay let's talk about Circus transmission lines and electromagnetic fields a transmission line is any pair of conductors used to move energy from point A to point B often of controlled size and in a controlled dielectric to create a controlled impedance I asked one of the engineers at l3 right after I joined the company I was getting ready to start a new design and I said to the guy what is the typical impedance of transmission lines in your boards and he said oh we don't have impedance and I thought hmm I said you mean you don't control impedance yeah yeah that's what I meant that's what you have an impedance whether you control it or not because you have inductance and you have capacitance and you have resistance and you have conductance and those four elements constitute impedance and whether you control it or not you have an impedance in every board so for what it's worth anyway the equation for impedance of a transmission line is R plus J Omega L divided by G Plus J Omega C R is resistance of copper J Omega is a frequency function which we'll talk about later L is inductance G is conductance conductance of what the board material it's shown in this model as a resistor it's not a resistor is an impedance to the movement of energy in the transmission line it's called conductance the inverse of conductance is lost tangent and that's why high frequency circuits have losses in material because their frequency is high relative to line length the material is not fully conductive and as you move energy down the transmission lines some of the energy is lost in the medium and we'll talk a little bit more later about that but that's what G is C is capacitance below a gigahertz R and G are really don't care they're don't care issues below a gigahertz unless you're dealing with with analog analogs another matter but I'm talking in this case digital in a digital circuit below a gigahertz R and G are for the most part of wash and you can throw them away and still get a fairly accurate impedance calculation in fact well oh let's talk about this first because of that we're gonna focus right now on defining inductance and capacitance and we'll come back a little later and talk more about R and G but for now let's define inductance and capacitance because everybody is affected by those two inductance is inductance is an impedance to change in current flow caused by the mass or the inertia of the magnetic field meaning what meaning if I have a transmission line it's going to have some defined inductance and it's going to have a magnetic field surrounding the transmission line and the size of that magnetic field will determine the inductance of that line what its flux basically correct more or less so the the mass of energy in the field creates an inertia so if I attempt to change level of energy in the transmission line I have to change the voltage and change the current to do so as I attempt to change the current I have to overcome that massive magnetic field to move it to a different size and shape and as you do that that causes a delay in that change of current and therefore causes a voltage drop across the inductance of that circuit I learned in college that inductance is a function of our size and to a degree it is to a degree it is but let's talk about real numbers for a minute let's say that we have a two-layer circuit board and I've got a trace and it's ground trace returns return trace I don't like the word ground but I use it because that's what everybody insists on using I've got the trace and a reference trace on a circuit board on the same layer of the board separated by some distance let's say these are ten mill wide traces okay and I've got them separated by five hundred mils a half an inch there's going to be an electric and a magnetic field in this dielectric space between the two and the size of that field is going to be large because I've got these two separated by a large by a long distance and the result is I'm going to have high inductance because of the size the amplitude of the magnetic field I learned in college that an easy way to reduce inductance is to increase wire size what about increasing trace width if I increase these ten mill traces and make them wider this direction in other words keep them a half-inch apart but make them wider away from each other do you know how wide I have to make them to reduce the inductance by 50 percent I have to increase them from 10 mils to five hundred mils wide to reduce the inductance by 50 percent if on the other hand I move them closer together without changing their width at all I can reduce the magnetic field well below what would be would reduce here by making the traces wider and I can probably reduce the inductance by 60 or 70 percent just by bringing them close together if I put one above the other then it create an even more favorable geometry to contain the magnetic field which means it shrinks even further same amount of energy in the field but its size its amplitude its inertia has changed and by changing its inertia we have lowered inductance it's all about proximity you want to have vias that are delivering power on a decoupling cap you want them to have a low inductance don't put them at the ends of the cap put them side by side as close as you can physically physically get them without upsetting the manufacturing people and near the part where the trace is routing to them are extremely short that's how you lower inductance is proximity it's all about proximity now again size matters but it's a secondary factor compared to proximity and that's really what this slide is talking about when you when you if we had this trace sitting here relative to ground traces we would have some inductance if I simply put a plane under it I'm going to drastically reduce the inductance without changing line size at all without having any impact on line size at all and this goes on it capacitance is the same thing remember on the previous slide inductance is the square root of inductance divided by capacitance so I want to get low impedance I want low inductance and I want high capacitance how do I get high capacitance it's the same way I get low inductance close proximity keep the trace above a plane not separated by a half an inch two traces that are a half inch apart are gonna have a really low capacitance you bring them closer together it'll go up ah you'll lower impedance because you'll lower inductance raise capacitance here it's even better keep that in mind and that's all this slides talking about an extremely important aspect of grounding is to establish a reference or a reference for a return current in transmission lines this is an important this is important for both signal integrity and for EMI return references will be a very important part of this discussion today as an example if we have a circuit board that has a trace on layer 1 and a plain on layer 2 I will get some performance out of it if I alter the plane on layer 2 2 4 different configurations I'm going to get 4 different results when it comes to radiated emissions off of this board dr. Wu from the National Taiwan University presented this slide in one of his presentations at the I Triple E emcee symposium some years back and he asked the question there were a roomful of about a hundred and twenty almost all double E's and he asked the question which of these will have the best e of my signature and which will have the worst and what will the two in between be I want you guys to just look at it for a few seconds and then we'll move on give it some serious thought the dashed line is a tres sorry I should have explained this better this dash line is a trace on layer one and the gray area is a point plain on layer two and that's true here and here and here the gray is a this is a crosshatch plane both of these across that's planes this is a plane with a split in it and this is a solid plane so they're essentially all have a reference on layer two but they're different references so I hope you guys have an idea in your mind know which ones you think will function so I'm going to come back to that later I'm gonna leave you hanging for about an hour the term ground refers to the earth we first used ground as a return path for lightning to keep from burning down buildings they figured out about 300 years ago oh if you put a big piece of metal on the roof of the building and a big honking wire down the side of the building if it's a wood building that when it gets wet will conduct duh when it gets wet it will conduct electricity we don't want that to happen because that usually sets the building on fire so put a big metal rod and a nice long piece of wire deep into the earth and you'll have a path to divert lightning great idea and it worked really well and it still works today lightning rods are a very key ingredient with a lot of buildings this is a perfect use of the earth as a return path it was soon realized that the earth could actually be used as a return path for other kinds of circuits which is by the way where the word ground came from when they used to use the earth for returns in a lot of cases one is telegraphy in the 1830s when they were stringing telegraph lines around the world in Europe and in the United States mostly for a long time they used two wires because they had to have a transmission line with a forward and return path so they had two wires on the pole and they put transceivers in every so many miles because as you push energy through these lines because of loss tangent you lose some of the energy over distance right you dissipate the energy in the signal so you have to reboost it so every so many miles they would have a transceiver that would take the energy boost the signal back to its initial level and send it on its way one point in time they were hooking up they were actually called transceivers because they were receivers transmitters they were hooking up two transceivers and they strung the wires across the poles and they turned everything on and it worked and all of a sudden somebody noticed hey we didn't hook up the ground wire it's there but it's not hooked up at one end and somebody said but the damn thing's working why is it working well the reason the circuitry in this transceiver this was a big metal box heavy metal box the circuitry inside had a return for all the signals and that return was attached to the chassis of the box in multiple places heavy attachment of ground to the enclosure the enclosure was sunk deep into the ground and attached to a concrete pad below the frost line so that in winter you wouldn't get heave that would cause the thing to block tip Oh or have some other kind of problem or a tornado would pick it up and blow it away you know ten miles away or whatever they wanted it to stay there so they had a good solid attachment of the circuitry into the earth and the end result was that the returned energy went through the earth well once they figure this out and realized we don't have to use two wires obviously wire was expensive still is it was ridiculously expensive in the 1800s because they were just starting to really use it a lot and they realized wow we need to we need to do this on a regular basis I was watching a movie about six months ago that was filmed in the 1940s and it was filmed in Mexico and there were these guys who were bandits and they were running from the law and of course as in any good western film let's cut the telegraph line so they can't you know communicate to the extent and there was a wire on the telegraph line because that's how it was done once they figured out you don't need to earth is not a great return path but it will work if the signal strength is large enough and the frequency is low enough earth and sometimes chassis of a system can be used as a return path based on frequency and amplitude another example would be some equipment on factory floors motor generator sets and things of that type where voltage was low and frequency was low or voltage was high in frequency was low or wiring in automobiles and so on automobiles as far as I know to this day still used the chassis as the return side of the battery they didn't certainly when I was young and used to actually work on cars well back then I could figure out how to work on them today I don't have a clue but anyway they use the chassis as a return for the battery and for a lot of the circuitry in the car because it made a perfect return voltage was high frequency was low frequency is lea already said all this from the activities used we spawned the use of the word ground and boy has that created a lot of confusion I'll get to that in a minute we'll talk more about that in a minute for safety reasons ul requires when you're dealing with a 3-yr AC line that you take the green wire in North America in Europe it's either gonna be yellow or yellow green you take that wire and attach it to metal chassis at one end and attach it to the earth at the other end why for safety reasons if any high voltage in the Box shorts against the box shorts to the box is particularly the AC input shorts to the box you've now got a path for return current back to the earth which the AC is reference to that will cause a high current and cause the fuse to blow so that you won't walk up and go there and be harmed severely by it it's all about safety and that's why they do this safety is his only role in its only value in fact if you don't hook that third wire up properly by adding some inductance to it at the box level it's probably going to become a radiator of EMI if you couple energy on to it from your box you need to pay attention to how you hook that up it carries no value with respect to EMI absolutely none there are a lot of people who think you have to attach a system to earth for EMI testing or to pass EMI testing absolutely untrue this has no attachment to earth and yet it passed in all of yours past EMI testing just fine the avionics products that we designed and built at l3 and put in aircraft there was no wire coming out of the plane attached to earth somewhere to help this thing pass EMI testing and yet they always did you don't have to attach anything to earth in fact if you do you have to do it with care EMI personnel I've often heard referred to the chassis as chassis ground earth ground it is not ground it is a chassis that functions as a Faraday cage to control and contain energy that is the metal chassis function when we design it poorly it won't contain the energy we'll talk about that later when we design it correctly it will but it is a Faraday cage and only a Faraday cage and we'll talk about ways to treat it to make everything function right earther chassis is not a good return path for anything at high frequencies in today's electronics earth our chassis should only be used as a safety point attachment point like we just talked about or for a place to couple ESD or lightening energy to divert it away from our circuitry but Rick what if that thing doesn't attach to Earth the fact is you will still dissipate that energy across the surface of the box and that's better than allowing it to be dissipated into your internal circuitry and that's why we attach it as the enlightening devices to the chassis never to the internal ground big mistake when you do that and I've seen people do it sometimes they get away with it even not a great plan but do the chassis it is not necessary I already said this ground on PC boards is often considered a region of zero volts and zero impedance absolutely untrue it's only close to true at DC due to impedance due to inductive and capacitive impedence earth or ground I'm sorry on circuit boards is not zero holes and it's not zero impedance ground is often thought of as a place to attach components to bleed off or filter noise as if it's somehow a sinkhole and I honestly have no idea how this concept ever crept into the minds of some engine but I know engineers who think that way I've heard engineers say oh attach a capacitor to that line-to-ground that'll pull the energy to ground and it'll disappear oh it will I'm sorry how is it gonna disappear you're attaching it to what is ground ground is the return side of everything it's the return side for transmission lines the return current travels in the ground plane it's the return for power routes and planes it's the return for everything that has a forward path it's the return path nothing bleeds off or goes anywhere I think this concept came about because of the water analogy for years I was engineers were taught in college that electricity is a little bit like water moving through pipes and you pull water from a reservoir move it through pipes it comes out of a faucet you do work with it and then it goes into the drain and back to the to the reservoir or the river and I think that's how some engineers envision circuitry so ground is a little bit like that drain it's not at all like that drain it is the return side of stuff and we're gonna talk a lot about that today the concept yeah in the 1992 is time frame I saw a survey that quite frankly changed my life it was in the e Times Magazine at least I believe it was EE times I mean we're talking 27 years ago and I'm old so there's no guarantee that that's actually what it was but I'm remembering that was 80 times and the survey basically said if you're a practising EE if you're a manager we love you if you're a technician we love you if you're this or that we love you but don't bother to fill out the survey we want double e's to fill this out and tell us what you think of your company your job the the the field of engineering we want to ask you some questions and like to have answers I read through the questions and there was one question that really caught my attention what was missing from your college education well by then I had explored enough noise problems in the 1980s we started having noise problems in circuits and I had explored enough noise problems because I I mentioned I was first attacked in a EE well in the process of getting from tech to EE I spent three years in field service there is nothing that will train you better to troubleshoot problems than laying on your back at ten below zero in a paper mill in Wisconsin in January trying to troubleshoot a circuit to figure out what in the heck is going wrong and by the way at that point in time every circuit board we had that didn't have the reference designators in chronological order on the board I wanted to grab the layout person by his scruffy neck and choke every ounce of life out of his miserable being because I was freezing and dying in this environment and I'm trying to find a component that's where is it a lot of Engineers think that the schematic should rule I'm sorry I think you're idiots but enough said anyway this survey well I saw what was missing from your education so I answered it the answers that came back in EE times three weeks later shocked me what I said to that question was I wished the college had taught me anything about understanding noise problems and how to prevent them not how to not how to to ban daid then oh I knew how to ban daid them you know it was like having a nail in your foot and the way we treated our noise problems in the 1980s we'd shove a bandage over the nail and give the guy crutches so he could get around without bleeding to death and killing his foot by stepping on it we didn't pull the nail we needed to pull the nail but we didn't know how so we ban dated problems and what I wanted to know is how to design to prevent them all together over 80% of the people who answer that survey answered that question the same way I did up to that point honesty I thought I and the engineers I worked with were the biggest group of idiots on planet Earth because none of us knew how that really solve noise problems none of us our college education was useless when it came to that and I suddenly realized wow it isn't just us there are a lot of idiots in the world so I decided to rank them in there I was gonna do something about it and I just when I started heavily buying books I bought my first book in 1984 when we had our first noise problem it was Henry aughts book and then I bought Ralph Morrison's book and you know yadda yadda about 1992 or three I started heavily buying books I was buying seven to eight books a year and reading all of them I now own over a hundred and twenty books and by the way I paid for him out of my own pocket because when I changed jobs I didn't want the company saying oh by the way those are ours no they're not they're mine I spent $20,000 on books over the years and it's been worth every penny of it trust me anyway I started buying books and studying and trying to figure out why noise problems occur in the first place why does energy get from circuit a in the circuit B we can see the effect of circa days energy on the input of devices on circuit B in particular on analog devices because they're most sensitive to noise why were we having and later as digital rise times got really fast we could see it happening in digital circuits why is this happening to us we didn't have a clue but I wanted to know desperately wanted to know so I started reading it took me about 10 years just to really get my arms around the problem starting in about 1993 ish and somewhere around 2003 or four I met Ralph Morrison and that man changed my life he sat in one of my two day classes at PCB West back when I used to do two-day classes there and at the end of the second day he came up to me and he said Rick loved your class good content lots of good information can we have dinner I thought God Ralph Morrison loves me this is phenomenal man the guy's gonna take me to dinner and tell me Rick you're the most wonderful guy on planet Earth and geez what a phenomenal presentation and we went to dinner and he said Rick I really did love your presentation had really good material however there was that word but and I knew what that word meant it meant everything before was a lie and I said however what and he said you're thinking of it incorrectly and I said meaning what and he said you talked about voltage and current it isn't about voltage and current and I said I'm sorry what he said it's not about voltage and current it's about fields it's about fields and where they travel in circuits that's what creates noise issues and creates EMI and even create signal integrity problems it isn't about voltage and current it's about fields he then asked me what is energy and I gave him some half-baked answer that I pulled out of my head that I was remembering from college physics and it was something to this effect this is a definition right out of a physics book it was something to this effect and then we talked about energy the fact that it's never created or destroyed you know that you convert energy from one form to another yada yada and that there are many kinds of energy there's mechanical sound light heat and so on and so on we can turn an electrical energy into light or heat we can do all sorts of things but we never lose or create energy by the way speaking of that just shout it out if you know the answer and please feel free what is the one and only difference between electrical energy and light energy thank you several people shouted it thank you very much that is the only women light and electricity are the you betcha they are exactly the same thing they are energy consisting of electromagnetic fields so then he said where's the energy in a circuit Rick is it in the voltage or is it in the current and I said well good question Ralph thinking in terms of power delivery we have a relatively constant voltage 3.3 volts and as we pull energy out of the power bus to drive transmission lines we are pulling current to drive these lines so we have high frequency currents with relatively stable voltage it's probably in the current Ralph he said no actually it's neither and I said it's neither and he said yeah no it's neither the energy is entirely in the fields the electric and magnetic field there is no energy in voltage or current zero zip none nada in voltage or current none at all in voltage or current all the energy travels in the fields then he said to me where are the fields in the circuit are they in the traces or are they in the plains now I'm really scratching my head going I don't know Ralph and he said I'm messing with you Rick it's neither the fields travel in the plastic of the circuit board between the trace and plane when you have a trace routed above a plane all of the energy travels through the plastic and fiberglass of the fr4 none of it travels through the copper it attaches itself to the copper because the copper is the lowest impedance path to move the energy from here to here if that weren't true radio wouldn't work radio works because we can broadcast waves into free space and the reason they will attach themselves to an antenna when they encounter one is because it's a lower impedance than the impedance of free space free space has an impedance of 377 ohms 120 times pi an antenna is gonna have a much lower impedance so as it goes past the antenna some of the energy is coupled into the antenna and into the input of the receiver capacitors wouldn't work if this weren't true you couldn't move energy from one plate of a capacitor to another if the energy were in the voltage recurrent the energies in the fields and that's where it always is in the fields you want to hear the best part there's no such thing as voltage I'm sorry wait a minute what there's no such thing as voltage when we put a voltmeter or an oscilloscope on a transmission line to measure voltage we are actually taking the integral of the e field from distance point 1 to distance point 2 with respect to time we are measuring the energy in the electric field over distance with respect to time that's all voltage is there is no energy there is no voltage there is energy in the fields period and that's the key to and once I understood that once Ralph got that into my puny little mind all of a sudden everything started to make sense because now that I realized that the energy traveled in the plastic of the board I realized why some transmission lines behaved well and others didn't I realized why a lot of the EMI problems I had had I've had and all of a sudden I knew how to take the nail out of the foot or prevent putting it there in the first place avoid stepping on the thing in the first place the energy travels in the plastic and fiberglass only in the plastic and fiberglass the energy in a transmission event is called a wave an electromagnetic wave the traces or trace and plane that make up the transmission line steer the energy from point A to point B because it's a lower impedance path it will energy given a choice we'll always take the path of least impedance always 100% of the time no exceptions ever we'll always take the path of lowest impedance so when we put a trace at the output of a driver and we push energy into that trace and its return path we shove it into the dielectric between the trace and its return path well what if we don't have a return path nearby well guess what's gonna happen folks that energy is going to spread out until it finds a return path and you probably won't like the results so you have to every time you route a trace you are routing half of a transmission line and you better pay really close attention to the other half of that transmission line because that's going to determine where the fields will be contained in the circuit the copper elements act as a waveguide a wave all I know about wave guides yes the RF engineers I worked with they talked about wave guys yeah RF people have wave guides yes so do you so does everybody remember the light bulb experiment that you were shown in elementary school and the teacher lied to inadvertently she was nice she didn't mean to lie lie to you and said yeah you hook up the wires to the battery and the light bulb and you get current flow from the positive side of the battery through the light that flows back to the battery and it's the energy and the voltage and current that cause the light to light up incorrect what actually happens is the fields are launched from the battery into that transmission line between the two wires and it travels in the air between the two wires and the closer you have those wires together the more you contain the fields but at DC who gives a rip right a DC who cares if they spread out big deal but it's not actually DC because the initial burst of energy is a wave so it isn't really DC it's just like the output of it of a driver driving a transmitter in line so the energy moves through that dielectric between the two wires and when it reaches the light bulb it lights it up but it happens so quickly because it moves at almost the speed of light that we don't really notice it that's how all circuits work I don't care if they're DC or 50 gigahertz all circuits work on that principle period what is this it's a waveguide what do you mean it looks like a tube to me it is it's a metal rectangular tube I first encountered one of these when I right out of college I went to work for Xerox in upstate New York worked in their R&D building in Philips New York or Webster in New York and after a couple years I wanted to be challenged more I mean I love Xerox a great place to work but was a big company and they were giving me one hat to wear for six straight months and I wanted to wear five hats a day you know I mean I wanted to learn and so I left and went to a small company and when I got there I was assigned to work in the R&D lab and as a technician and I went in the lab my first day and I saw a guy working with an amplifier and a receiver that were several feet apart and he had a tubular waveguide in between them and I said oh that's cool what's that and he said it's a waveguide and I said is it what like a conduit and he said no it's a waveguide and I said where are the wires in the waveguide he said there are no wires and I said how's that work and he said I don't know but it does I mean this guy was working with it and he didn't know how it worked how does it work if I apply an energy field now this isn't how you would do it I've drawn it this way for a reason if we apply some fields to this waveguide and if those fields have a wavelength whose half wave is this distance we get a standing wave inside this tube and that standing wave will transmit down the tube through the air in the tube creating a current in the wall of the tube as it travels from point A to point B why a halfway look at this sine wave if I'm right here a half wave later is right here same voltage opposite polarity no matter where you are on a sine wave when you're 1/2 wavelength away the voltage difference is always zero so you don't short this energy to the tube you create a standing wave in the tube that transmits down the tube this is also a waveguide what is this same thing this is a substrate integrated waveguide looks like you've shorted the trace to the ground plane yes we have to create a substrate integrated waveguide just like the air waveguide on the previous page and it functions in exactly the same way except the energy doesn't move at the speed of light it moves at the speed of light divided by the square root of the e sub R of the board material I got into an argument once with Yann who's the chief RF engineer at Goodrich we used to plate the edges of our RF boards that were operating at multi gigahertz because he was concerned about fringe fields off of the edges of the board's coupling out and creating one problems or to some of our communications information coupling off of the board and being able to be read by an enemy in a KN time of war for example so we would plate the edges of boards I contended for years that we should be able to create a via string with the vias close enough together that they will act like a wall of copper and Jana said nope no matter how close you put them they won't there will be leakage points well that's incorrect and we later proved him wrong and I'll tell you much later today how far apart vias need to be based on frequency to create a continuous wall of copper and it's not as close as you might think they especially well we'll talk more about it anyway so what they're doing here is creating a tube a rectangular tube that they can move the she threw if this is a waveguide what is this the microstrip trace above a plane it's a waveguide they're both wave guides the way we're moving the energy into this waveguide is through that why do we use a tubular waveguide so we can push one and only one frequency and that's the frequency whose half wavelength is equal to that distance around the tube all other frequencies get shorted to ground and essentially don't move through the waveguide and that's the purpose of it the tubular waveguide on the previous page was used to move millimeter wave frequency energy things that have really short wavelengths this one similar but not quite as high of frequency this is what the fields look like here in the microstrip versus in the waveguide you can see they're more well contained and controlled in the in the substrate integrated waveguide but they essentially look the same the fields in a micro strip though they spread out a little or essentially the same as the fields inside the waveguide they're both wave guides a trace referencing a plane is a waveguide and the energy will always move in the dielectric space so we have to be sure we set it up correctly and we have to know how to do that I've already said this already been through that so a transmission line I said earlier is any pair of wires or conductors used to move energy from point A to point B the voltage if you want to call it that is across these two copper features if you stick an oscilloscope across here you're gonna see a voltage waveform which is actually the energy in the e field but nonetheless we'll see what we call voltage and it'll be a square wave right coming out of the output the current is in the copper I've had many engineers tell me yeah yeah I know about current it goes into this trace goes down here and then returns to the driver incorrect that's not at all what happens the energy moves into this dielectric space from the driver for actually from the power bus the energy is pulled out of the power bus through the dielectric of the IC out of the output pin relative to the ground pin of the IC through that dielectric space and into this dielectric space and as the energy starts to move through this dielectric it establishes a voltage across the copper features and it will establish a forward current and a return current simultaneously there's no forward current return current it's at the same time there generated so when the energy has reached this point here there's voltage from here to here across this entire length there's current down both sides of this copper there's no voltage down here there's no current down here the energy hasn't gotten there yet that's how transmission lines work how we take care of this side of the line really really really really matters it really matters and we're going to talk a lot about that today here's a picture that was I drew this picture but I took it from a simulation from hyperlynx of how a circuit actually functions and I drew it because I want to show the thing to you in steps not as one continuous picture if I put a trace between the planes of a circuit board two layers two plane layers and I apply energy to this transmission line the first thing that happens the e-fields couple from this line to the plane above and the plane below the first thing that appears are the e fields establishing what we call voltage in the transmission line the existence of the e fields caused the movement of electrons in the copper and that movement of electrons causes what we call current flow to take place in the copper of the trace and in the copper of the planes notice that the current in the planes has spread out to a wider area than the current in the trace because the fields spread a little and as they spread they move current in a wider swath they don't spread far but they do spread right I've shown the skin effect condition here meaning what meaning at very high frequencies you don't penetrate all the way through the copper to create current flow at frequencies in circuit boards with circuit board thickness coppers half an one ounce coppers once you get above about seven 800 megahertz or so you won't use all of the copper in the circuit you'll only use part of it and you'll only get current movement through part of that copper structure why because at higher frequencies the fields the electric field doesn't have time it because the field appears and then at some point in time later it disappears and then reappears again later at high frequencies it doesn't have time to penetrate all the way through the copper so we only get current flow through part of the thickness of the copper this is referred to as skin effect and what ends up happening is we lose of the bowl if people will tell you that what happens is you get a resistive drop across the copper that takes away some of the energy from the signal that's not really quite what happens what actually happens is we the the higher frequency we have the harder the electric field has to work to cause current flow in the copper because it wants to move X amount of electrons right to create current flow and the only way it can do that is to work quickly so it has to exert more energy to move X amount of electrons and the fact that it exerts more energy means we lose some of the energy you always lose some of the energy as you move electrons in the copper even at low frequencies but the losses at low frequencies are so minimal we typically ignore them once you get above a gigahertz or so these start to become significant and people call it resistive loss but it isn't really it's how much work the fields have to do to move X amount of electrons in a thinner layer of copper it has to work harder so it gives up more of its energy to the copper that energy is reflected out of the copper in the form of a magnetic field that surrounds the copper and couples to the plane above and below people say it's the current flow that generates the magnetic field but it's actually the energy from the electric field generating the movement of electrons that creates a magnetic field around the trace the energy is in the fields and only in the fields people will tell you that rough copper if you have rolled a kneeled copper which is smooth versus electrodeposited copper that you get lower skin effect losses because with the bumpy copper they'll tell you that the electrons half the current has to move up and down this hilly thing and it takes longer and causes great resistive losses no it it takes more energy because it's harder still to move X amount of electrons in a rough surface than it is in a smooth surface that's the reason the resistive loss occurs by the way we mentioned loss tangent loss earlier its loss of energy in the dielectric all the energy is moving through the dielectric of course you're gonna lose some of it some dielectrics are lossy others are not there are low loss materials designed for high-frequency operation fr4 is one of the law seus materials there is there are times that's a good thing when it comes to power delivery that loss enos is actually a plus and we're going to talk about why when we get into power delivery so lossy materials certainly have their place as long as they don't cause so much loss of energy in the signal that it cripples the ability of the system to function and that can happen at moderately low frequencies in the analog world and will certainly happen at gigabit speeds in the digital world so be aware of that yes question well it would be more efficient in terms of loss tangent loss yes if you could if you could run a wire between two planes in air it would be way more efficient the losses would that's one of the reasons they use tubular wave guides because at millimeter wavelengths where losses are really an issue they want to be able to move the thing through air and that's exactly when you're absolutely right question of course was if you move it to air wouldn't it be more efficient the answer is absolutely air is the most efficient dielectric we have and in fact if you look at an outer layer trace and look at the fields on an outer layer trace you can see that there is less magnetic field energy and electric field energy in the dielectric some of it is moving through free space which is more efficient but unfortunately that causes greater skin effect losses because the majority of the current has to flow on the bottom of this trace instead of all the way around the trace so it increases skin effect losses but it decreases loss tangent losses so the question you have to ask yourself is which is more important to me and if if you can't get there with fr4 then you have to start thinking about higher higher grade materials due to the natural field spread of outer layers as as shown here when the need to avoid noise coupling into a line is critical we will often route distributed length lines that are critical on inner layers when you just can't tolerate losses typically you want to route high frequencies critical signals on inner layers there are exceptions to that when the lines are short enough that they're not a distributed length then you don't want to go to inner layers because why in the world would you want to go through a via which is going to generate some level of loss go through a via to another layer go 1/4 of an inch and then back up you see where I'm going with this that would be absolutely insane so even in critical RF circuits we never did that in all the RF circuits I ever designed in the aerospace world well there was a good richer l3 we always lumped all of the elements of a circuit an amplifier an oscillator circuit the local oscillator though whatever circuit that the D mocks all of the circuits each individual circuit all of the components were clumped together really tight so we could route everything with lump length elements and then we would dive into inner layers to route from the preamp to the input stage across an inner layer because that was going to be a distributed length so we'd put all the really serious distributed length elements on inner layers and anything we routed on outer layers just make sure we kept it to a lumped length you newse wanted me to make all of these lines when i first design I did at Goodrich make him all 50 homes well this was a 12 mill wide pad because of the dielectric thickness is that we used for RF circuits why would you use a thick dielectric in an RF circuit so you can use very wide traces and still hit a 50 ohm impedance why do you want to do that because there's a tolerance in the dielectric thickness which affects impedance so if you have a thin dielectric plus or minus 1 mil is a huge tolerance movement but if you have a 25 mil dielectric plus or minus a mil is almost nothing if I have a 25 mil dielectric I can make it trace 20 25 30 mils wide to hit a 50 ohm impedance and plus or minus a mil of of itching on that is very little tolerance so we made all of our RF stuff use thick dielectrics and very wide traces so that we had good control of impedance because all the drivers and all the receivers had 50 ohm outputs and 50 ohm inputs you needed to drive from that 50 ohm output into the transmission line without loss of energy movement because you're moving through a constant impedance down the line into the input of the receiver also a constant impedance we needed to match that 50 ohms as close as possible to prevent reflections and other problems he wanted me to make all these short traces 50 mils wide this is a 12 mil or 30 mils wide it's a 12 mil wide pad good luck with that so I told him meow John it's no problem I'll take care of that and of course I ignored him and didn't do it and thank God he didn't ask to see the layout before we sent it out and it worked perfectly this is one of the first layouts I ever did a good reached in 1991 it was an RF amplifier circuit all of these traces are lumped links bait it was a 2 gigahertz circuit and all of those linked all of those traces are lumped links at 2 gigahertz so who cares about their impedance what is this little guy and all it's a test point come on Rick you can't put test points and gigahertz circuits why not as long as they're a lumped length and don't hurt anything who cares we don't care about the width of these traces they're lumped elements and we don't care we put 100% test point coverage on every circuit board I've been involved with since 1991 because we found out the hard way what happens when you don't and bullet a is sometimes it's a challenge the last product I worked on at l3 was a little 3 by 4 inch circuit board it was six by what does that work out to three by four hell I can't think I was trying to think of millimeters anyway doesn't matter three by four inch board it had ten BGA's on it two of the BGA's were 1300 pins each and there were a total of 308 parts on this board I'm gonna tell you putting 100% test point coverage on that mother was a challenge we managed but boy it was not easy it took a long time we built the prototype without 100% test point coverage because the prototype once it was assembled they could they could test it with a flying probe tester and it was expensive but we could build prototypes wasn't cost effective but we could do it we needed to we needed the test points for production builds they need to be able to test these boards and know that all the solder joints are good all the components are good and everything is actually going to work when I said we found out the hard way we built a dozen boards once in the early days of good Rick's digital boards that had no test points so the assembly house couldn't test them and not one of them worked so our technicians had to sit and go through them one at a time finding out okay what solder joint is screwed up or what component is bad that's causing these not it took weeks and weeks that we lost from our delivery window test points are critical anyway food for thought microstrip style boards most of our RF stuff was what we're microstrip they're actually coplanar waveguide with ground which we'll talk about a little while but microstrip means the traces were on the outer layer there was a ground plane of two layer board ground plane and layer two and we had all the power routed on the outer layer and I'll tell you later why in the RF world there's no reason to use a power plane absolutely none whatsoever you can route power we used to route power up to and including 16 gigahertz circuits there's no need to have a power plane in an analog circuit period anyway enough said we were out of the power and we would pour all the open area with ground copper and then attach all that ground copper with lots of vias to the ground plate and that looked from the side something like this if you rotate this thing around so you're looking down on the top of it and if two of the ICS on this board happen to be in these corners and had a route that went all the way around the board where would the return current for this signal line be if that's a plane on layer 2 when I start working for the first time with an engineer whom I've never met before or another board designer or whatever I draw this on a white board and I asked him the question where's the return current and about 80% of them say it's from this via going to the ground plane straight across to that via going to the ground plane and I'm sure most or all of you know that is absolutely incorrect at DC the current would be this direct path because that's the path of lowest impedance because if you take the inductance out of the equation altogether the numerator is resistance it's R so the path of lowest impedance is that direct path at DC yes I'm gonna I'm gonna explain that yes the question was how about the electric field in that case and in fact what actually happens when the current flows across here the this distance you will actually get current flow in a square area or a square ish area and its width will be the same as this length so you'll get current flow through this whole area at DC and the fields will spread across that entire area to create that current flow because that DC that whole area is a path of really low impedance as you go up in frequency you get a little bit less current flow here and a little more out here and you keep going up in frequency and all of a sudden you're getting current flow away and by the time you get to in a board of this size they'll say this is a three by five inch border two-and-a-half by five inch board and this trace would be five ten maybe a foot long at that length by the time you get to one kilohertz all of the return current is going to be directly under the trace back to the driver from a kilohertz and Beyond and certainly by one megahertz once you're out of the audio range there will be no current flow anywhere except the signals here that the return current will be on the plane below directly below the trace and the energy will move through the dielectric forming current in both features at the same time so when it moves here there's current in both directions it moves here now you've got current in this whole span in both directions and so on all the way around until the energy reaches the load and that's how all circuits work all of the time so why does the return current take the path directly under the trace because it's the path of least impedance which is defined as for the most part the square root of L over C for the most part it is dominated by L&C the path directly under the trace is the path of lowest inductance because that's the path that generates the smallest volume of magnetic field and the path directly under the trace is the path of highest capacitance because that's the path that generates the smallest volume of electric field right makes perfect sense doesn't it L&C are dominated by the fields and the path of least impedance at frequencies above audio in a circuit board are almost always directly under the trace unless we screw things up and diverted elsewhere like putting splits in planes because Oh analog device that said that's what we're supposed to do their app note said that's what we're supposed to do so we should do that and I had a guy from analog devices attending this was about 15 years ago attending one of my classes and I made that statement about wasn't it it was linear technology so somebody else I was making fun of and this guy from AI came up to me and he said you know I read an Intel app note recently that said to do this this and this what do you think of that and I said I think they're incorrect completely incorrect and he looked at me and he said yeah but Rick it's Intel and I said let me tell you a story I said one week ago I was doing a class and a guy came up to me and said I've got an apt note from analog devices and they say to do this this and this and he said yeah but Rick gets analog too and the guy from a analog devices might say no more I get it because he knew their app notes were crap nobody needed to tell him he knew they were anyway ok then why does the low frequency current spread out the answer is simple it really has to do with the fields it's really comes down to skin effect the fields want to move X amount of electrons and the lower you go in frequency the more electrons it wants to move once it penetrates all the way through the copper it'll start spreading to find electrons to move so it DC it'll spread really wide as wide as it is long resistance is a per square measurement and current flow at DC is a per square measurement if you think about it in terms of this equation what is inductance at DC it goes to zero the frequency is zero inductance goes to zero the numerator is dominated by the resistance of the copper so the energy will take the path of lowest resistance which is the direct path from the ground to ground point there's a neat experiment you can run that will show you exactly how this works put a signal generator on a coax I first saw this done by dr. Tom Van Doren and later by dr. Todd huming both of Missouri University of Science and Technology in Rolla Missouri the first University in the world to do EMI Studies was um are they would take a 2 meter piece of coax about my top my height a little bit longer than I am tall and they loop it back till the two ends were within two or three inches of one another then they'd solder the shields together with a short really heavy piece of wire heavy enough to take its solder it without damaging the dielectric but as heavy as they could get it and still solder it properly to the shield hook up a signal generator to this end a resistor to that and calculate how much current is going to flow based on voltage and resistance write Ohm's law we know that's how it works put a current probe around this short wire put it set it to DC and turn this on and you'll find that at DC all of the return current is in this path right here at DC all of the return current is in that short path 100% of it as you go up in frequency this starts to become a high impedance path and by the time you get up to a few hundred Hertz there's much less current flowing in this wire by the time you get to a kilohertz there's way less even with this being two or three inches long and this being thirty times longer even with those conditions by the time you get to one megahertz there is zero current flowing in that wire and here's a graph that todd ubin put together that shows the current distribution in the wire based on frequency by one megahertz by in fact just above 100 kilohertz there's almost no current in that wire when you're talking about circuit boards that are three by five or four by eight or whatever when you're talking about those kinds of links by the time you're out of the audio range there will be no current flow anywhere except directly under the trace period all digital circuits today are high-speed some analog circuits are not so how do you keep current from spreading in critical low frequency analog circuits you know people have said to me man I I really look up to you that you've designed 10 and 20 gigahertz circuits man that must be tough actually it's not you know what's tough low frequency stuff that's what's tough because you have currents wanting to go everywhere and you have to channel them now at low frequencies it's less critical because we have a much wider noise margin at low frequencies and that's the good news but if you don't want circuits to couple into one another you still have to channel them the way we used to do it in businesses where I've worked over the years when we route a low frequency trace we'd route a return trace about three times wider directly below it from point A to point B we didn't have a ground plane there wasn't a ground plane we routed the return path with the trace you don't want field spread at low frequencies that's what it takes now if you don't care fine but if your low frequency analog circuits are sensitive and they're fairly near one another on a circuit board and you're concerned about circuit a coupling in the circuit B you don't want those fields to spread because then you may get field coupling into circuit be from circuit a food for thought I had a discussion with Ralph Morrison about a year or two after him and I met and he asked that very question he said I'm sure you've done low-frequency designs in your earlier years and I said yeah a lot of he said how did you keep the fields from creating problems at low frequencies I said we channeled the returns with defined return fast and he looked at me and smiled and said good boy Rick good boy you know good boy Rick well done back when this two layer board was designed in 1984 this was part of a robot and when this thing was designed we'll take a break in just a few minutes when this thing was designed in 1984 frequencies were low enough that we could do two layer boards we didn't need a ground plane anywhere in a board we had power routes and ground routes that were made wide for a reason which I'll explain in a minute and everything functioned properly anything that was shorter than 24 inches long based on a rise time back then had no problems well these circuit boards were only about this big I mean unless we were rather to trace 14 times around the edge and gonna be 2 feet long so we didn't have problems looking at this a little bit closer if I wanted to follow this trace right here based on the frequencies of many years ago of 1984 where would the return current from this trace be the answer is everywhere because at low frequencies this trace the fields are going to spread out across this entire region and it's going to couple a return current in one of these traces in low-frequency one and two layer boards the traces use each other constantly as a return path the reason we made the power and ground routes very wide was so the predominant so that the majority of the return current would be in those white traces we wanted the fields a couple most of its energy into the wide traces there was a second reason we made them wide it's called copper balance for manufacturability now I'm sure everybody in this room knows exactly what I'm talking about so that's how that worked at today's frequencies first off you wouldn't be able to make this work but if you attempted this at today's speeds where would the return current from this trace be well the answer is probably entirely in these two traces at today's frequencies the fields would couple so tightly to those other signal traces there'd be no return current in the power ground rails all the return current would be in the other traces back to the driver and that's why it probably wouldn't work let's take a break

Channel: Altium

Views: 16,985

Keywords:

Id: ZYUYOXmo9UU

Length: 119min 37sec (7177 seconds)

Published: Sat Nov 16 2019