Just to add to what Scott said: Just because you have a SPICE model, does not mean that it correlates in any way (or in your area of interest) to reality. There are many SPICE models out there that have been "sanitized" to protect IP. These sanitized versions are usually not as complete or accurate over the full range of operating conditions as were the original physically correlated models they were derived from. In fact, this issue was one of things considered when IBIS was first being designed: protecting IP while giving the necessary data. (Not saying success was achieved, but it was the plan). So even when you have a SPICE model, you have to check to see how good it is. jon -----Original Message----- From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]On Behalf Of Scott McMorrow Sent: Tuesday, April 01, 2003 9:59 AM To: cclewell@xxxxxxxxxxxxxx Cc: jeff.loyer@xxxxxxxxx; twesterh@xxxxxxxxx; si-list@xxxxxxxxxxxxx; rhaller@xxxxxxxxxx Subject: [SI-LIST] IBIS, si simulators, modeling and other sources of correlation error Craig, To clarify a bit. If one is comparing Spice simultations of driver/receiver circuits to Ibis simulations of the same circuits, usually the most glaring error will be in the creation of the IBIS model. There are so many poor ones out there that the differences between the IBIS simulator and Spice are huge. Once the IBIS model is extracted correctly and correlated, then usually all sorts of other errors begin to pop out. Some easily seen, some hidden in the details of modeling. I can classify these in two areas. 1) differences in simulation environments. 2) issues with modeling. In area #1 board level signal integrity simulators are often poor at correctly modeling the entire system. For example, component mounting pads, their associated capacitance and inductance, are excluded from the electromagnetic extraction. Only the traces themselves are extracted. Another example in this category would be in the modeling and simulation of lossy lines. In Spice we have many alternatives, since we can create lossy line models in many different ways, each with it's own tradeoffs and errors. In board level SI simulators, there is usually only one way. Usually there is an embedded field solver which extracts the parameters very quicky (in a matter of seconds) and then a simulator which uses a standard lossy model for simulation. In general, this is often some variant of the w-element algorithm. Given that the field solution is taking only a few seconds per cross-section, there is no way that fully coupled lossy parameters are being extracted completely, which can give rise to significant errors. Usually frequency dependent inductance is not extracted at all. Often a TEM approximation is used to derive the inductance matrix from the capacitance matrix, which is totally incorrect when we are dealing with lossy systems. In some cases the algorithm being used is just plain wrong and gives incorrect results for lossy lines. Unfortunately, board level signal integrity tools rarely allow for other simulation or modeling methods to be integrated into their environments. What they will do is to export the circuit to external Spice simulators, or launch a Spice process. Unfortunately, what is being sent into Spice is often still the result of incorrect or incomplete modeling. A final example of a major case #1 problem is in the area of ground and power modeling. Most of these board level signal integrity tools do not correctly model ground or power. Ground is often treated as a node 0 infinite current sink that is connected to all ground pins everywhere. As you know, this invalidates coupled multi-line connector modeling, as it does coupled multi-line package modeling. Additional modes of propagation that result from including ground and power conductors into the simulation model are ignored by most board level SI simulation environments, causing minor simualation errors for low impedance ground and power structures and high errors for higher impedance ground and power structures. In area #2, there are a number of generally areas of modeling that are not well considered. Package models for semiconductor devices are usually poor approximations of the actual substrate. The problems can usually be categorized into the following areas. * insufficient bandwidth for the extracted package sections. * failure to correctly model the power and ground paths * mixed models with loop inductances for power and ground, mutual inductances for signals. * losses not modeled * resonance effects of package not accounted for * silicon redistribution layers not modeled * wirebonds modeled as single inductor, rather than coupled RLC circuit. * package vias and feedthroughs not modeled. * package plating tails not modeled. * wrong topology used for package spice model. Vias are rarely modeled and in most cases are not modeled well. Via-to-via coupling within tight pitch fields like BGA breakout areas are rarely modeled, yet have significant coupling effects in high performance systems. Connector modeling for edge card connectors always makes assumptions (that are unstated) about the board which will plug into the connector, and as such will differ from the actual circuit under many customer design conditions. The largest area of assumption will be in the placement of the plane under the finger card traces and pads. It is almost universally assumed that this plane is a ground plane, when quite often it is not. For well-grounded connectors, actual performance may be significantly different than modeled performance, due to very large power/ground return path loops and slots created when a designer uses different power reference planes on opposite sides of the connector. Incorrect assumptions and/or extraction methods are often used when modeling lossy board traces. So what if the simulator can simulate the lossy element if the modeling assumptions are not based upon reality and/or verfied by measurement. And unfortunately, those measurements are often poluted by poor measurement launch design, or interpretatation of the results, due to unforseen or uncontrolled resonance conditions. Connector model boundaries, assumptions and limitations are not well delineated. Rarely have I ever seen a connector model that takes loss effects into account. Loss simulations of backplanes will often be off by 1/2 dB due to lack of loss modeling in the connectors Interactions between the electromagnetics of the PCB and that of the connector are often excluded in the modeling process, leading to questions in how to complete the modeling of the underlying PCB structures such as pads and vias, and questions about the accuracy at the connector/board transition boundary. Craig, these are just a few things that I can think of to elaborate on my previous topics. best regards, scott -- 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 ------------------------------------------------------------------ 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