Ott, Gilbert, Sauer, Harvey, and Fong discuss the new Spice model for transmission lines
Paul Rako - November 18, 2011
The previous blog post talks about Roy McCammon’s method to do Spice simulation of transmission lines. Before blogging about it, I sent the links to his original EETimes article to several analog aficionado pals. Their responses, praise, and repartee are printed below. Henry Ott, a former member of Bell Labs working for the Longlines division is a treasured source for anything about transmission lines. He commented about McCammon’s transmission line Spice method:
- A novel approach to modeling the terminal properties of a transmission line. Of course, when I was at Bell Labs we did not have Spice to model our transmission lines, we just used Heaviside’s equations (listed as Eq. 1 & 2) in Part 1 of the referenced article) and a slide rule or later a desktop calculator. In those days an hp desktop calculator was the size of a desktop computer today. Ain’t technology great!
Barrie Gilbert a fellow over at Analog Devices noted:
- Roy’s work on TM is intriguing and extremely useful and I have added a few recipients to the cc list that I didn’t see in a quick scan of the original. It is really a great shame that RAP (Bob Pease) isn’t around any more - to rant about the pure idiocy of attempting to cram the quart of quantum infinity into the finite golden pot of recursion. I’m too busy for a while to take this model for a test drive, but after I get around to finishing some other serious spicing of my soups (the sort that magically create revenues for my company) I certainly will. All three articles in this series are now safely stowed in a directory on my desktop, awaiting exploration.
- Roy: Thanks so very much, for showing us how to exploit a simple recursive approach to model a transmission line, rather than attempting to pursue infinitesimals. In the past I’ve done similar things to model a variety of continuous RC networks - such as the Tektronix ’scope probe cable that Paul briefly mentioned. You have added some valuable ideas to our toolboxes! Indeed, the problem-solving potential of methods of this sort - and it even appeals to the seemingly distant domain of fractals - is a little-known secret weapon.
Barry Harvey, IC designer turned director over at Intersil commented:
- Well, nice treatise on the subject. The greatest variable of the lumped circuit is estimating how many lumps you need to deal with the highest frequencies in the simulation. The finer the lump, the higher the spectral frequency content is reasonably simulateable. Even 300 sections is reasonable in a simulation since the primitive devices converge and simulate so efficiently. Of course, the Rs really varies as the square root of frequency (skin effect), so the simple model is not so good for loss issues. I found lumped circuits worked quite well for transient simulations as with pin drivers. Cadence also has a lossy line model that is based on Laplace, but it requires storing past data to represent delay, and it can work poorly when transient timesteps vary from very small during transient events to longer during quiet periods. You have to force the timestep way down to make the combination of quick and sluggish events coexist accurately. When you do, the simulation time increases until you get frustrated and go back to 300 lumps. This is like saying that FFT or Laplace transforms work best over limited min/max spans of input or output spectrum/time ratios. Perhaps 100:1 is about it, generally. The lumps work well from DC to some maximum frequency.
Don Sauer, semi-retired National Semi IC designer noted:
- It may not be possible to generate one model for all applications. But trying to get Spice to behave like the real world implies that you have the real-world data in the first place. Why not input actual real-world stuff as well? Native Spice provides the means to import data in text format as a piecewise linear file. My web site shows how stock market history can be imported into Spice to do FFT (fast Fourier transform) on it. A spreadsheet was recruited to format the text data in this case. I don’t have the resources to fairly compare all the flavors of Spice that are out there. But you would think that someone by now would have developed a much more professional method of importing actual real world behavior into their flavor of Spice.
Which got Barrie Gilbert to respond:
- What do you mean by “real world data” in connection with a transmission line? Roy’s cleverness lies in treating what is really a continuous, distributed structure as a recursive one. It is up to the user of such solution methods to input parameters that, as nearly as possible, represent “reality”. So, for example, consider an R-nR ladder of “semi-infinite” extent. The only user-controllable data here is the variable n. The “characteristic impedance”, looking this network in the eye, is of interest; so also is the attenuation per step. One could create a Spice network using a large number of stages to “discover” the answer; but only a few lines of algebra in pursuit of a recursive solution suffices; and the answer in this case will be fully analytic and general. Simulation programs such as Spice are, already, replete with massive repositories of “real world” data. If they were not, one would be left with nothing but a blind, disoriented mechanism for fast matrix solution….Rather, a professional IC design company, or considerate foundry, instills into the mind of Spice innumerable ideas about reality. If that were not the case, IC designers such as myself, who daily depend on the accuracy of simulation and the generous insights it routinely provides, would have to be the worst and most pitiable of all fools. (Fortunately, I have a formidably more powerful simulation engine than Berkeley Spice, under my fingertips). I must be missing your point. It seems just to be about the utilization of lists of breakpoints, somehow gathered from actual measurements. Spot data are frequently unreliable in the extreme, most especially when one is attempting to emulate the behaviour of fully-distributed RF & microwave circuits, in which transmission lines play an ever-present, and indispensable, role. In this domain, EM modeling has the edge over “real world” measurements.
At which point Don Sauer elaborated:
- I will give an example. This might help you to see where Pease was coming from. Enclosed is a pdf file [NPN_Diode_TCtxt.pdf] containing a simple simulation on NPN temperature drift. This pdf file contains a table of some actual lab data taken off the transistor to be modeled. Now one time at National Semiconductor, the Cadence systems allowed two different Spice engines to be run. Neither flavor of spice came close to the lab data, or with each other. When you looked into the equations that the models said they were using, you wondered if someone had not thrown in a fudge factor or two. The suspicion for fudge factors came from debugging some cadence tech files that National’s CAD department had been developing, and finding out they where throwing in fudge factors to correct for fudge factors. Spaghetti code. You sound like you know a much different Spice environment. The point is, somewhere we need something that can be trusted. Would it be correct to say that reality is what should be setting the standards? When we all have agreed on the right way to measure reality, and when the measurements in the lab are giving all the same results, then maybe we might trust such “real world data”. Can you see why we tend to trust lab data with a good and long track record over Spice models? Being able to actually sanity check something has always seemed to prove itself worthwhile (At least at National). Now yes, sometimes you have to construct sort of a virtual reality to look at something. But still, isn’t some traceability back to reality needed to do that correctly? The feedback may not be perfect, but is there anything better? The point I was making is, would being able to import things like lab waveforms into Spice be useful “as well”? In the very least, it seems like it could aid in the verification and calibration of models.
Which sure got a rise from Barrie, who invoked Pease’s ghost:
- Don: You write: “Neither flavor of Spice came close to the lab data, or with each other.” This sounds a lot like Pease’s ghost speaking -the Anti-Spice as he has aptly been named! Oh, I wish we had him back, to taunt and irritate us. May you Rest At Peace, old man. But now……The days are long gone when analog simulation produced seriously “wrong” predictions of circuit behavior - whether merely a little bit misleading, or profoundly differing from product reality. We in fact DO “import reality data”, as a matter of necessity - and routinely: during the long and crucial period of IC Process Characterization. It is here that one’s general-purpose matrix solver (take your pick) is transformed into a specialist: one who knows all the arcane model equations by heart, and can readily cope with nonlinearity, either of which are practically impossible for you or I to conquer, equipped with only a clear head and a pencil as our spear.
- Such deeply trained expert companions are far more reliable in dealing with numerous matters of detail than the use of hand analysis, or even (heaven protect us!) BREADBOARDS, so much favored by dear old Pease. Once, while both of us were waiting in a lunch line, I asked “So Bob, how DO you pursue design?”, he said: “Pencil and paper, pencil ‘n’ paper”. I responded “Don’t you ever use calculator?” He said “Nope! Can’t trust ‘em! They give wrong answers”*. One’s trust of today’s simulation tools does not mean that a lifetime of experience - the appeal to abstract, mathematical solutions, or to one’s intuition and insights, are no longer appropriate. Implying such would be ludicrous. In an e-mail exchange with Pease, some years ago, I wrote that “used in an analytical mode, ‘Spice’ could provide one with deep insights”. In my home lab I keep 96** HB pencils ever sharp and ready - always with a fresh pad of paper on the bench. In an emergency, there are a dozen or so slip-sticks, scattered around; place as well (I confess) as an HP calculator or two…. This isn’t my principle Far-Seeing Place; that’s the Library, a floor up. Nevertheless, you can see a great deal on those ancient, flickering green screens - often suggesting a return to the Sim Shop is called for….Clearly, one must necessarily “Verify Reality” by comparing the actual measurements acquired at first (or later) silicon with the predicted behavior that was so readily visualized while simulating. In the course of pursuing such rigor, over a period of endless decades, the “better species of Spice” have settled down - so much so that one cannot even hear one’s hard-disc rattle any more….When things DO “go wrong”, the most common cause is a layout indiscretion. It is here that the transformation from “thinking to things” is most likely to be corrupted. But even here, we have plenty of tools at our disposal to back-annotate parasitics - R’s and L’s as well as just C’s - and iterate our way through yet another set of sims. Of course, whenever necessary one can readily import tabulated data into a simulation study. A case in point might be a set of S-parameters in a microwave development. In fact, the closest I have come to your notions, Don, was to consider (as a neophyte at Plessey) equipping the simulator-de-jour with some sort of data-acquisition system which would allow one to plug in an actual, real, glass-enveloped transistor and “refer” to it as a source of model parameters on-the-fly; that is, at run-time. Crazy idea. Our pencils, slide-rules, our simulation machines, an arsenal of perversely expensive equipment on our benches are all, each in their own unique way no more than tools; and their individual limitations must, instinctively, be ever at one’s fingertips.
- * The occasion was Tutorial Day at ISCAS at San Diego, once. His presentation was entitled roughly “IS BREADBOARDING A LOST ART?” while mine was something like “CAN DESIGN EXPERTISE BE CAPTURED?”. I have both of these talks on VHS tapes, and perhaps should transfer them to DVD some time. His talk was entertaining; mine was maybe a bit too serious. From the audience, Yannis Tsividis complained passionately that: “Spice does not converge”. Yes, that’s certainly sometimes true, especially with today’s large and hugely nonlinear circuits, and one must rummage around in the toolbox.
- **That is, a box of 8 dozen, from Costco, $2.99
So directing this rant at Don was pretty funny since Spice is obviously a religious topic for some, and Don Sauer was the biggest and most ardent Spice jockey I knew at National Semi. So Don replies:
- I agree with you on pretty much everything. I still have a few ICs in production that were designed with slide rules, pencil and paper, and breadboards. It sure was nice when the cost of calculators came down. The youth are really missing out when it comes to breadboards. One really needs to play with something to get a good feel as to how things work. But today, it gets a little tricky to breadboard, when all your minimum transistors have stray capacitances in terms of femto-farads. Sounds like you know a professional Spice environment. Everyone should know such an environment. I am challenged as to how to allow engineers to be able to play in Spice like a breadboard. Computers can do some things that humans just can’t do, but computers aren’t yet intuitive. Its unfortunate that most analog engineers have not developed a passion for computer programming. They are missing out on some pretty powerful stuff.
- PS. And to think that Pease was envious of me for having a lab in my cubicle…
So Roy McCammon must have enjoyed this back-and-forth since he sent the correlation of his method with Belden Coax:
Some data on simulating Belden 8259 (RG-58) 50 ohms
Which got Barrie Gilbert to joke, “Hmm… perhaps the writers of the data sheet for this cable used your method!”, so Roy McCammon responded:
- Coax is probably the easiest to model because in some ways it is the simplest, it has been analyzed thoroughly and has some known solutions, at least in some regimes. Belden coax is especially easy because they give you a lot of information such as the diameter of the center conductor, center conductor resistance and shield resistance; you can deduce the outer diameter of the dielectric. You can calculate dc inductance and high frequency inductance.
And that got my buddy Edison Fong to chime in:
- I couldn’t agree with you more Roy. What’s even better is in HFSS where you can describe (draw) the exact coax in any coordinate you want (rectangular, cylindrical (ideal for coax), and even spherical) and it will do a pretty job of modeling it.
So there you are, a nice discussion among some analog greats about Roy’s spiffy transmission line Spice model. I always found it frustrating to get in between the flame wars of Barrie Gilbert and my dear friend Bob Pease. They were both right and they were both wrong. To use a great expression taught to me by yet another analog great, Mike Maida over at Texas Instruments, the two guys arguments were orthogonal. Barrie was right that the custom proprietary ADICE simulator developed for IC designers at Analog Devices Corporation did not lie. My IC tool design buddy Dave Tamura, also at TI, confided that analog companies will spend hundreds of millions of dollars to insure their Spice models matches the silicon processes. And my dear friend Bob was right too. Spice does lie, and you can forget semantics, when it gives you a wrong answer or doesn’t converge it betrays you. All board-level system guys like me are much more wary of Spice. We don’t have transistor level models running on ADICE. We might not have any models at all, and some of them, like the legacy ones at National Semi, were really crude, no noise, no PSRR, no million other things. All I know is that the “Spice always lies/Spice always is true” attitude of Bob and Barrie sure were extremist digital positions held by two analog geniuses that should know better. I know Bob Pease often put on his “Spice lies” attitude for effect. I did observe him with a more reasonable attitude. Maybe we can get Barrie Gilbert to come to the next Analog Aficionados party the Saturday before ISSCC. With a few drinks in his fine English frame, he might admit that Spice does, occasionally, misbehave, at least when it is not ADICE. I will try to have a recorder running.