Altera PowerPlay Analyser

Hi,

I have a design with two identical 16 taps filters runing in parallel on the same inputs. A switch allow to disable one of the two filters.

I also have two test bench runing on the same inputs : - one has the two filters enabled. - one has only one filter enabled.

I used post-synthesis files (.vho) to simulate the two test bench and record all signals into .vcd files (values changed dump files)

I use these .vcd files to perform Power consumption estimation for the two test bench.

In the case where only one filter is running, I only have 45% of the nodes toggled during simulation. In the case where the two filters are running, I get 85% of the nodes toggled during simulation.

In the first case, the powerplay power analyser inidcates a low power estimation confidence. In the second case, the powerplay power analyser indicates a medium power estimation confidence.

Can I consider the power estimation confidence in the first case medium as in the second case, knowing that if the estimation is low, it's just because I have intentionally not used half (1 filter out of 2) of the design ?

Hi,

Yes, that's exactly it. We had a tough design decision on how to handle zero-toggle signals, and we will be changing the functionality in the next release of the software. But first some background.

The PowerPlay Power Analyzer's power estimation confidence metric is intended to indicate our confidence in the signal activity information; a power estimate is only as good as the toggle-rate information provided by the user. While the Power Analyzer can guess at what happens on nets based on the activity on other nets in the design, the highest quality power esimates are obtained when all activity is derived by post-fit timing simulations.

In Quartus II 5.1, we decided to treat nodes that have a zero toggle rate from simulation the same as nodes for which there is no simulation information at all. The reason is that this could indicate the user didn't fully exercise their design. For example, if you didn't set your reset signal to the right value, nothing in your design would toggle and we'd say we have low confidence.

In Quartus II 6.0 (which will be released at the end of April), we are changing this behaviour. We will treat any signal read from simulation to be of good quality, even if it does not toggle. This means you (the user) need to be careful that when parts of your design do not toggle in simulation, you really meant for them not to toggle; Quartus won't warn you about it. But that means designs like yours where circuitry doesn't toggle for a reason will no longer result in a warning. A more common example of this is when a design contains start-up or error-handling circuitry which normally does not do anything.

BTW, you can obtain additional information on how Quartus comes up with a confidence metric by looking in the Confidence Metric Details panel of the Power Analyzer report. This report lists the various signal activity sources and how many signals came from each.

Regards,

Paul Leventis Altera Corp.