Issue link: https://iconnect007.uberflip.com/i/1295812
14 DESIGN007 MAGAZINE I OCTOBER 2020 change will affect signal quality and timing on a relative basis. We want tools that let system designers make more informed decisions on their own because the SI expert's involvement generally has to be scheduled, but the need to evaluate design trade-offs is constant. Giving system designers good first-order analysis tools means the design will be in better shape when the SI specialist looks at it, which helps reduce design spins. Shaughnessy: Are you talking about having an analysis run in the background during the lay- out process? Westerhoff: Not necessarily. Analysis is more useful as a "checkpoint" process, particularly at the interface level. There's not much point in running analysis on half of an interface; it's better to get to a point where the design has reached some level of completeness and then produces results quickly. After each check- point, you continue to refine the design and add detail, gradually predicting the system operating margin more and more accurately. We refer to this process as progressive design and analysis. We're trying to match analysis to the way design normally takes place. You start by figuring out the big stuff and then add details as you go, checking your design each step of the way. For this to work well, you need several things: 1. A repeatable analysis process that produces a numerical metric for system operating margin. 2. An established pass/fail threshold for system operating margin. 3. An ability to dial design details and physical effects in and out of the system model. Let's take a closer look at each of these indi- vidually. Analysis Process First, the analysis process and results have to be absolutely repeatable. Simulation isn't much good if you can't run the same analysis twice and get the same results. When you're assessing design trade-offs, you need to know that the simulation process you're using is rock solid, or you won't have much confidence in the design decisions you make. You want that process to be entirely automated so that you just define simulation parameters and the analysis process runs by itself, always produc- ing the same results in the same format. Pass/Fail Threshold Second, you need well-defined criteria for whether a design passes or fails. System oper- ating margin can be specified in millivolts, picoseconds, decibels, or something else, but it needs to be specific, defined, and numeric, and it needs to match what the analysis pro- cess actually produces. That way, there's no ambiguity about success or failure and no ambiguity over which design alternative pro- vides more margin. Variable Modeling Accuracy Third, when we add more detail into a sys- tem interconnect model, the model takes lon- ger to generate, and the resulting simulation runs slower. We don't need ultimate accuracy early in the design process when we're just roughing the design in; it slows the process down without adding much value. Matching the model detail to the task at hand will make the process more efficient. Equally important, being able to dial design details and physical effects in and out of the system model allows us to numerically quantify the impact they have on system margin. It doesn't make sense to spend a lot of time optimizing something that has a negligible impact, and the things that matter most will change from design to design. During active design, we want to focus on what matters most and optimize the design accordingly. Progressive design and analysis means we use basic models at the start and add detail as we go. The level of detail in our model increases as the design progresses. Because we know there are things we're not modeling at any given point, there's an ambiguity, or "error band," in our analysis that's large at the start