Case Study: “Your Mileage May Vary:” Creating Reliable Comparisons of IP Cores

An attractive attribute of licensable processor cores is the flexibility chip designers have to adapt these cores to their chosen fabrication process, cell library, tool flow, logic synthesis goals and other conditions.  In other words, chip designers can tune the core to the needs of a particular application and to their preferred chip design methodology.  An unfortunate side effect of this flexibility is that it can be extremely difficult to make apples-to-apples comparisons between licensable cores.

In general, each supplier uses its own unique set of conditions and assumptions to generate published performance, silicon area, and power consumption data for its processor cores.  Further confusing matters, suppliers rarely disclose the full set of conditions and assumptions underlying their published data.

One solution to this problem is for chip designers to do trial implementations of candidate processor cores, to evaluate the results they’re likely to get under their chosen conditions.  But doing several trial implementations requires a significant amount of work on the part of the chip designer.  Early in the design process, when there may be many candidate cores and when there is time pressure to make a core selection and get started with the chip design, this approach is typically not practical.  An alternative approach is for the chip designer to ask the core suppliers to do this work instead.  But the core suppliers probably don’t have access to precisely the same tool flow, libraries, and other items used by the chip designer, making it difficult for them to produce results identical to what the chip designer would obtain on his own.

To enable chip designers and core suppliers to make quick, initial apples-to-apples comparisons among competing cores, BDTI has defined a set of uniform conditions that vendors use when reporting results for licensable processor cores on BDTI’s benchmark suites. Currently, these conditions are specified for 130 nm and 90 nm process nodes, for each of the BDTI benchmark suites. They ensure that all processor core benchmark results for a particular process node use the same well-defined conditions, and that reported speed, area, and power metrics for all cores are obtained in the same way.

To learn how to effectively use BDTI’s uniform conditions for comparing licensable cores, contact Jeremy Giddings at BDTI (giddings@BDTI.com).