In January, I reviewed the Dennis Fitzpatrick book Analog Design and Simulation using OrCAD Capture and PSpice for EDN. I've also provided three excerpts from the book for EDN (Here’s the third, published last week.)
I shall not repeat all the points I made there about Fitzpatrick's book. However, given the context of this forum, I did want to discuss the chapter on mixed-signal simulation. If you're going to integrate analog and mixed-signal functions, you'd better get them right.
That's always easier said than done. Anyone remember Mentor's LSim? I was the technical lead for that project for several years after Mentor acquired it from Silicon Solutions back in the early 1990s. One of the interesting aspects of LSim was the way it handled mixed-signal simulation. All the solvers (analog, transistor timing, gate, behavioral) were unified in the same kernel. The basic component models were written using its own behavioral language, so it was at both the top and bottom of the hierarchy, so to speak. If you wanted to connect an analog net to a transistor, you just went ahead and did it. It would become a mixed node, and it was possible to define aspects such as the digital rise and fall slopes and threshold voltages. During simulation, the two subnets were evaluated and combined in a hierarchic manner.
Chapter 19 of Fitzpatrick's book describes the way this is done in PSpice. Interface nodes (defined as those that contain both analog and digital signals) are automatically separated into one analog node and one or more digital nodes by inserting analog and digital interface subcircuits — either analog to digital or vice versa. These subcircuits also have their own power supply.
An analog-to-digital converter is nothing more than a threshold detector. It could also present a resistance and capacitance to the driving circuit. A digital-to-analog converter works with ideal digital components that have just a few parameters associated with them: voltage, resistance, and capacitance. We could make very minor adjustments to the gate output voltages and rise times based on the load that the analog circuitry provides to adjust the slope of the output signal, but none of this is beyond what LSim provided. If you had a very special case in LSim, it was possible to model the node using the behavioral language.
Unfortunately, the chapter does not provide any more guidance on how this works, so I resorted to the PSpice user's manual. It shows that the interface models are basically the IO characteristics of the logic family (TTL, ECL, etc.), and that up to four levels of accuracy are provided for each. LSim provided a sampling rate on the interface that allowed any number of accuracies to be defined.
It seems as if there has been no change in the basic way mixed-signal interfaces are defined over the last 20 years — perhaps, if anything, a little simplification has occurred.
A few days ago, I talked to some folks from Analog Rails (about a completely different subject). They said that analog designers are stuck in the old ways of doing things, and that they had a very different way of handling analog design and simulation that would increase productivity.
Having been away from the analog world for two decades, I have to say that nothing much seems to have changed. What changes have you seen in the past 20 years? What changes do you want to see in terms of both simulation and functional integration?