realistic operation the part would always be inside the gray area. But if the device will be loaded with a drastically different load, e.g. a 10 ohm load, than a new model for this load condition has to be created.
BIRD 104.1 - AMI MODEL - NEW IBIS SUPPORT
Manfred Maurer, Siemens AG, Germany
Manfred began the presentation by saying that many customers and IBIS users had asked him about AMI, and he pointed out that there had also been a lot of presentations at different IBIS summits during the last two years. There was the impression that it might not be possible to create an AMI model without knowledge of MATLAB or C++. He showed the advantages of a classical IBIS model in ASCII format. The easy readability of that standalone IBIS model meant that the engineer could look at the model directly. Then he pointed out that there had been a lot of improvements for IBIS models like [Driver Schedule], [Add Submodel] and even [External Model].
With the appearance of the AMI model a new level of model has come into being. It looks as if there is a growth of applications for this kind of model. The reason for this AMI model came from SerDes devices operating at frequencies above 5 GHz and their need to model equalization, feedback, clock recovery etc., and the need for simulation of extremely long (> 10 Million UI) bit streams. A large list of parameters is needed and is defined inside the model. These parameters might be platform, or even compiler, dependant. Furthermore, there are user defined parameters that are not easy to define (e.g. jitter, Rx-clock-PDF).
His next slide showed that there are three function signatures, AMI_INIT, AMI_GetWave and AMI_Close, and writing these functions requires the knowledge of a lot of disciplines (software engineer, electrical engineer and mathematician). He explained that an AMI model contains many parts including the I/O-model, channel information, complicated source and sink circuits and control parameters. All this information will be packed in a DLL (dynamic linked library), so the user does not know explicitly if the model will run on his/her system with their specific version of operating system.
The AMI model consists of an electrical part and an algorithmic part. The electrical part contains the transmitter, receiver, and the channel that is characterized by means of an impulse response. The algorithmic part includes the equalization and the clock and data recovery and is connected by high impedance to the analog part. The modeling description is done by an executable code (MATLAB/C++). He ended with a summary of the advantages and disadvantages of the AMI model. Advantages include that the model shows maximal flexibility, protects proprietary information and shows a short simulation time for millions of UI (in comparison to any other traditional tools). The main disadvantage is that the model is no longer in ASCII-format, but is hidden in a DLL.
Manfred was asked if he had used the tool kit. He had used it and it was complicated, but in the end, he was successful. Multidiscipline knowledge was necessary. The next question asked about the time it took to get first results, and he said it was easy after he had talked to the toolkit suppliers. He was asked about the future of IBIS, and he said that during his attendance at the IBIS summit at DesignCon 2008, he got the answer from the experts. At the moment only SerDes will be modeled with AMI, but the rest of IBIS will stay.