I'm resending my response to Weston, since the formatting seems to have gotten all messed up on the previous response. Weston, Fear is a good thing. If used properly, it is a good motivation for change. We should always seek to understand the limitations of the tools we use, the approximations which are involved, and perform some sort of engineering analysis as to whether they are appropriate to the problem we are solving. Signal integrity is really about margins: timing margins and noise margins. The question you have to ask yourself is whether your approximation of the problem (as embodied by modeling and simulation) have enough accuracy to insure that you can meet your margins. (Remember that there is a difference between precision and accuracy. A precise answer delivered by a simulator may not be an accurate one.) So, yes, do use the tools that you have ... but ... understand their fundamental limitations and do not believe their results implicitly. As a friend of mine often tells me, there is a fundamental operating principal in the the universe: the Conservation of Misery. The laws of physics do not change, but we do often bend them for our own convenience, either analytically, computationally, or politically. Whenever we do this, misery is conserved in the form of incorrect assumptions, errors in approximations, or just plain wrong answers. To borrow from your carpenter's analogy, you either use the right tool for a job, or you suffer the consequences. I'm sorry, board level EDA signal integrity tools can no longer claim to solve the customer's problems when they do not state the assumptions used, the compromises made, and show correlation to real circuits in real systems, and the actual errors introduced by their methods. With sub 150 ps edge rates, single ended busses operating at 200, 300, 400, 500 MHz, and differential busses running at 2.5, 3.125, 5, and 10 Gbps, all these "little things" begin to (and do) matter. If they didn't, why would we have chipset vendors establishing trace length matching rules of +/- 10 mils. So, what could a board level signal integrity tool vendor do? Well, for one, publish the known limitations of their products to there customers. Actually specify the accuracy of the algorithms used for field solutions and simulation. State known problems in the extraction of signal and power/ground conductors. State known problems and errors in the modeling of packages and connectors. Show correlation to real measurements, and explain why differences exist. State the accuracy of via modeling solution. Then ... provide easily supported and automated ways in which the user can import externally derived models for pads, vias, connectors, transmission lines, black boxes into the modeling and simulation engine from external tools in useful formats like Spice and S-parameters. Yes, in some cases, 3D field solvers are necessary and must be used to obtain accurate results. These can be used in areas where there are obvious limitations of board level tools, such as vias, non uniform trace geometries, package breakout regions ...etc. A good, accurate 2D field solver that handles lossy conductors and dielectrics, that can predict the frequency dependent inductance along with resistance and conductance are an absolute must for anyone serious in knowing what real signals really do on boards. Then, a field solver is useless without a method to actually simulate the lossy circuit in the time domain. And ... wouldn't it be nice if the methods have been shown to have some level of correlation with real measured results in the frequency and time domains? For pre route characterization, it is very useful to be able to have all of these tools and methods at your disposal. For post route characterization, wouldn't it be nice if a board level eda tool were able to provide a heuristic for finding the worst cases for a bus on a board, and then automatically extract them for further refined analysis using more advanced modeling techniques? The board could be passed through the conventional signal integrity and timing scans to find the "big chunks" and then pass off the final verification to a simulation and modeling environment that is more capable and accurate, to provide the final simulations for sign-off. The conditions are many and varied. In general, if you follow the rule that anything that is 1/20 of a wavelength of the highest frequency transmitted or 1/10th of the edgerate, or 1/10th of the timing margin that you are trying to achieve, then it ought to be modeled accurately. Given those very loose rule of thumb, for a bus operating at a frequency of 500 MHz, with driver rise times of 150 ps, and a design margin of 100 ps, in FR4 stripline, then we have: 100 ps modeling error for 500 MHz 15 ps modeling error for 150 ps rise time 10 ps modeling error for 100 ps design margin. In this case, the modeling error required to achieve 100 ps margin is the limiting factor and requires a 10 ps modeling accuracy. This is equvalent to accurately modeling every feature that is longer than about 50 mils in length. This would then include all traces, vias, pads and connectors and packages. As you can see, had the timing design margin not been so severe, then the same design could have been modeled to an accuracy of 15 ps (about 80 mils) and if slew rate controlled drivers were used to limit the risetime to 500ps, then modeling accuracy could be reduced to 50 ps (about 300 mils). So, for most common busses, it is either the rise time of the drivers, or the timing margin on the busses that is the limiting factor. For differential signals, timing margin is usually not as severe, and losses begin to dominated. In these cases it is usually the 3rd harmonic of the fundamental frequency of the fastest data pattern that is the limiting factor. So, for a 2.5 Gbps serial link that uses encoding and clock recovery, the third harmonic of the bit rate is 3.75 GHz, which is a 266 ps period. 1/20 th of this is 13 ps. So any feature longer than 13 ps should be considered critical and modeled accurately. This results in a feature size in FR4 of about 70 mils, again, the size of some pads, and definitely with the range of size for vias and plated through holes. In any of these calculations, stubs have a multiplication factor of 4 on feature size, since they form 1/4 wave resonators and require the signal to traverse the stub 4 times. Thus, in the 2.5 Gbps differential example above, stub lengths longer than 17.5 mils in FR4 are critical. This means that we need to especially look at things like square pads (which form stubs), test points not in line with the signal traces, and vias, which almost always have stubs. best regards, scott -- Scott McMorrow Teraspeed Consulting Group LLC 2926 SE Yamhill St. Portland, OR 97214 (503) 239-5536 http://www.teraspeed.com -- Scott McMorrow Teraspeed Consulting Group LLC 2926 SE Yamhill St. Portland, OR 97214 (503) 239-5536 http://www.teraspeed.com ------------------------------------------------------------------ To unsubscribe from si-list: si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field or to administer your membership from a web page, go to: //www.freelists.org/webpage/si-list For help: si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field List archives are viewable at: //www.freelists.org/archives/si-list or at our remote archives: http://groups.yahoo.com/group/si-list/messages Old (prior to June 6, 2001) list archives are viewable at: http://www.qsl.net/wb6tpu