In Paul Rako’s recent Voices article my pals Larry Lerner (Agilent) and Greg Edlund (IBM) give their perspectives on the state-of-the-union in signal integrity. Greg highlights the importance of using not only SPICE but also EM solvers.
The electromagnetic behavior of interconnect systems is becoming more difficult to subdivide into independent components. Signal-integrity engineers need to understand how components interact with each other so they don’t miss important effects by making the models too small. In the future, successful signal-integrity engineers will use 3-D field solvers in the same way they used SPICE in the early days: to enhance their understanding of system behavior.
For mission-critical design work, you’ll want to step up to a supported, commercial tool. Working for Agilent, obviously I’m biased. But here’s a link to EMPro anyway! Click here to request an evaluation license.
In the white paper on EM simulator techniques mentioned in my previous post, the authors note:
Even more than for MoM- or FEM-based solvers, the popularity of FDTD-based solvers has been facilitated by recent advances in the speed and memory capacity of computer hardware. FDTD is an inherently parallel method and therefore lends itself very well to the processing capabilities of the most recent advances in CPU (general purpose processors) and GPU (graphics processors) hardware.
Given that a computer configured with EDA software is valued in the tens or even hundreds of thousands of dollar, the incremental cost of a couple of thousand dollars on an NVIDIA Tesla card is worthwhile, given the massive speed up and the value of getting the results early and often.
…free with registration, by my Agilent colleagues Jan Van Hese, Jeannick Sercu, Davy Pissoort, and Hee-Soo Lee.
The growing number and complexity of high frequency systems is leading to an increased need for electromagnetic (EM) simulation to accurately model larger portions of the system. There are several different technical approaches to EM simulation, and while no method is generally superior to the others, each one of them is aligned with one or more application areas. This article will discuss the three most established EM simulation technologies: method of moments (MoM), finite element method (FEM), and finite difference time domain (FDTD), linking the simulation technology to solving specific applications.
Now that my company, Agilent, has three EM solvers, I’m often asked about the pros and cons of each technique, so this white paper really helps. As for our products, here is a are some links:
EMPro platform has a 3D drawing environment, our FDTD engine and, as we announced recently, with EMPro 2009, our FEM engine (EMDS) will be an available add-on.
EMPro and ADS are sold as separate platforms. But if you purchase both, they are integrated
Our other two EM engines are both integrated directly inside our ADS platform:
Despite rumors from some Web 2.0 fanatics that it isn’t necessary, you can still actually meet people “IRL” (in real life) not just virtually via the web. My Agilent colleagues and I will be staffing Booth #305 at DesignCon 2009, in Santa Clara CA, Feb 2-5 2009. Stop by and say hello!
“Day in the Life of an Agilent EEsof EDA Chiphead®”. Is there really an upcoming ISO 1984 standard? Find out in this (hopefully) hilarious parody of a famous series of YouTube videos.
PPS: Here are the Agilent papers and panels. Example: “7-TA2″ means track 7, Tuesday AM, paper 2.
7-TA2: “Practical Analysis of Backplane Vias for 5 Gbps and Above”, Eric Bogatin (Bogatin Enterprises - BeTheSignal.com), Sanjeev Gupta (Agilent EEsof EDA), Mike Resso (Agilent instruments)
8-TA1: “The use of Optimization in Signal Integrity performance Centric High Speed Digital Design Flows”, Brahim Bensalem (Intel), Sanjeev Gupta (Agilent EEsof EDA)
8-TA3: “Analysis of Random Noise and the Effect of Band-Limited Noise on Stressed-Eye Receiver Tolerance Test”, Ransom Stephens, Marcus Mueller (Agilent instruments)
13-TA4: “Verify your signal integrity margins: De-embedding of fixtures and probing in a real time digital oscilloscope”, Jim Choate (Agilent instruments)
12-WA2: “VNA Characterization of Cable Assemblies for Supercomputer Applications”, Greg Edlund (IBM), Mike Resso (Agilent instruments)
12-WA3: “Characterizing Non-Standard Impedance Channels with 50 Ohm Instruments”, Julian Ferry (Samtec), Mike Resso (Agilent instruments), OJ Danzy (Agilent instruments)
7-WA4: “BER Performances for High-Speed Serial Link System Estimated by using Quasi-Analytical Method”, Ding-qing Lu (Agilent EEsof EDA)
12-THA2: “A Comparison of Fixture Removal Methods for Characterization of Differential PCB Channels”, Weiping Hou (Huawei), Quan-Li Li (Agilent China)
TF-MA4: “Fixturing and Calibration Techniques for Obtaining Wide Bandwidth Measured Data for Time Domain Simulations and Measurement-Based Modeling”, Heidi Barnes (Verigy), Sanjeev Gupta (Agilent EEsof EDA), Mike Resso (Agilent instruments)
TP-MP: “The case of the closing eye - Addressing the Industry’s Next Gen Serial Data Design Validation Challenges”, Panelists include Karl Kachigan (Agilent instruments)
TP-WP: “Do it right or do it over? Signal integrity engineer in the era of highly compressed project schedules”, Panelist include Larry Lerner (CTO, Agilent EEsof EDA), Wh
There are many practical difficulties that have to be solved in a signal integrity project, but two difficulties have their roots deep in the fundamentals of Maxwell’s equations.
One is obvious: each 0 or 1 pulse is actually an electromagnetic wave that travels at finite speed. The speed is named after as that famous electromagnetic wave: light. In vacuum, this universal speed limit is about one inch per 84.7 picoseconds, or one centimeter per 33.4 picoseconds. For striplines in FR4 board it’s about half that. Although the wave is guided by the copper trace, the wave travels in the insulating material surrounding the trace. The speed is set by the dielectric properties called permeability and permittivity:
c = (μ0μrε0εr)−0.5
(The wave does penetrate the copper somewhat, but it dies away exponentially from the metal surface, with a decay constant equal to the skin depth.)
But another fundamental limitation is also related to these dielectric material properties. The relative magnitudes of the electric field E and the magnetic intensity H are called the impedance, defined by Z=E/H. The impedance for free propagation is:
Z = (μ0μr/ε0εr)0.5
For vacuum, this is about 377 Ω and for FR4 it’s about half that. (Hence the retro graphic at the head of this post: the old color code for a five-band 377 Ω resistor was orange-purple-purple-black) For a guided wave, the impedance is generally less than that for free propagation. You can divide the free propagation impedance by a unitless geometrical form factor that is a function of various dimensional ratios of the cross-section of the wave guide. Often the function contains a logarithm of the geometric ratios and its value varies only slowly with aspect ratio. For practical geometries, the form factor is a smallish number say between 1 and 10. FR4 and a form factor of 4 would lead to an impedance of about 50 Ω.
All very well, but why should you care about impedance?
Just as light partly reflects and partly transmits at the impedance change between air and window glass, so your signal will partly reflect of off impedance changes along your trace. (Actually, optical folk usually talk about refractive index, n = Z0/Z, rather than the impedance itself.) The reflected energy will bounce up and down the line and arrive at your receiver later than the main, direct pulse. Like a latecomer at the theater, this energy will corrupt the flow of information. Communication engineers call this multipath effect “inter symbol interference” or ISI. To minimize reflections and interference, you have to match the output impedance of the transmitter to the interconnect impedance, keep the impedance constant throughout the length of the interconnect, and match the interconnect impedance to the input impedance of the receiver.
When Dorothy arrives in Oz, she says to her dog, “Toto, I’ve a feeling we’re not in Kansas any more.”
Likewise, digital electronic engineers like yourselves are being thrown from your world of ones and zeros into a microwave “Oz” – with its splendid but puzzling reflections, impedances, and electromagnetics — and you’re saying “Toto, I’ve a feeling we’re not in binary any more.”
You’re right.
Hopefully this blog will prove a more reliable guide than the Lion, the Tin Man and the Scarecrow. We’ll try to help you become the Wizard of Electromagnetism.
In this first post, I’ll reference the posting that inspired this blog. It honored James Clerk Maxwell, born June 13th, 1831 in Edinburgh, Scotland, UK (also the birthplace, by the way, of Alexander Graham Bell). Maxwell synthesized previously unrelated observations, experiments, and equations of electricity, magnetism, and optics into a consistent theory and set of equations—Maxwell’s equations—demonstrating that electricity, magnetism and even light are all manifestations of the electromagnetic field.
James Clerk Maxwell with one of his colour wheels
Maxwell’s work has stood the test of time. Maxwell’s equations are consistent with the Lorentz transformation, and inspired Einstein’s special relativity. In this view, magnetism is not a separate force, but the simply the dynamics of the electric force with space-time distortion from charged bodies in relative motion (electrodynamics). If the electric field is interpreted as the probability of observing a photon, the equations are consistent with quantum mechanics (quantum electrodynamics). QED has even been extended to cover the weak nuclear force (responsible for beta decay) in the electroweak theory. Thus extended, the theory encompasses in principle all physical, chemical, and biological phenomena except gravity and the strong nuclear force (quarks and gluons and such). The bottom line is that you can take the solutions to the bank: unlike SPICE, there are no approximations.
EM solvers are now a standard signal integrity analysis tool. Although the equations are computationally expensive to solve, the results accurately reflect the distributed nature of multigigabit per second serial links, where the wavelength of the highest frequency component is shorter than the physical size of the backplane. This is in contrast to nodal solvers like SPICE, which use a lumped element approximation, and which ignore crosstalk due to magnetic induction in conductor loops (curl(E) = -dB/dt).
So this blog will pick up the thread from the post on my signal integrity blog to drill down on practicle applications of electromagnetism at microwave frequencies to digital electronics.
