Providing accurate toggle rates in the various XPE sheets is essential to get quality power estimates. This information, however, might not be readily available at the stage in the design cycle where you enter data in XPE. Activity might be refined as the design gets more defined. Following are guidelines you can follow to help you enter design toggle activity.
- For synchronous paths, toggle rate reflects how often an output changes relative to a given clock input and can be modeled as a percentage between 0–100%. The max data toggle rate of 100% means that the output toggles every active clock edge. For example, consider a free running binary counter with a 100 MHz clock. For the Least Significant Bit you would enter 100% in the Toggle Rate column, because this bit toggles every rising edge of the clock. For the second bit you would enter 50%, because this bit toggles every other rising edge of the clock. When data changes twice per clock cycle, enter 200% for the toggle rate.
- For non-periodic or event-driven portions of designs, toggle rates cannot be easily
predicted. An effective method of estimating average toggle rates for a given design
is to segregate the different sections of the design based on their functionality or
hierarchy and estimate the toggle rates for each of the sub-blocks. An average
toggle rate can then be arrived at by calculating the average for the entire design
or hierarchy. Most logic-intensive designs work at around 12.5% average toggle rate,
which is the default toggle rate setting in XPE.
It has been observed that designs with random data patterns as input generally have toggle rates between 10%-30%. However, designs with a lot of glitch logic can have toggle rates as high as or even higher than 50%. Glitch logic is generally classified as combinatorial functions which have a high probability of the output changing when any one input changes, such as XOR gates or unregistered arithmetic logic (i.e. adders). Functions that use large amounts of such logic, such as error detection/correction circuitry, might exhibit higher toggle rates due to this. Designs with large amounts of control path logic, such as embedded designs, on average have lower toggle rates due to large sections of logic being inactive at any given time during operation.
In summary, the primary factors that have an appreciable impact on the toggle rate of a design are:
- Input data pattern
- Random data pattern versus known patterns have an impact on the toggle rate.
- Control signals
- Use or lack of control signals such as reset and clock enables.
- Design logic
- High glitch XOR/CARRY logic, a highly pipe lined design, or an embedded design have an impact on the toggle rate.