# Estimating digital operation power consumption

I am wondering if there are generally accepted methods of estimating the power consumption of digital operations as they are implemented in custom logic (ASIC). From this previous questions, it seems that people have done it with simulator tools once the RTL is written.

What I am looking for, however, is a back-of-the-evelope method along the lines of this previous question. Perhaps this is an attempt at taking the question a bit further. In my case I am interested to see if there is a way to estimate a modular operation by itself.

For example when calculating the mean of n 8-bit numbers I would like to say that the power cost of such an operation would be:

P_mean8 = n*P_add8 + P_divide8

I realize that:

• Placement and routing (fanout, etc.) will play an important role in the final power number but I am not interested in getting bogged down in the details of such. I am more interested in getting a rough estimate assuming reasonably good practices.
• Technology nodes (90nm vs. 12nm) will change the number but I am hoping there is a method that will take a few parameters from the technology node and scale the answer accordingly
• Voltage times total gate charge times frequency times some factor. – Ignacio Vazquez-Abrams Sep 14 '16 at 18:20
• There are a lot of ways to implement divide in hardware, each with its own power requirements. For that matter, there are many ways to implement addition! – Cort Ammon - Reinstate Monica Sep 14 '16 at 18:25
• @IgnacioVazquez-Abrams For total gate charge I would have to assume a number of gates for each operation correct? Are there an accepted numbers used for low-level operations (i.e. add, multiply, etc.)? – Daniel V Sep 14 '16 at 18:28
• @CortAmmon Agreed. My question is, assuming you pick your favorite addition implementation, how would you go about estimating it's power consumption? – Daniel V Sep 14 '16 at 18:33
• So I am not a hardware designer, so I can't talk to how they should estimate their work. However, if nobody gives you an answer, that may suggest that it is not reasonable to seek such an estimate without getting into some of the details. It may be wise to sit down with a hardware person, discuss your particular problem with them, and see if they can give you guidance. An agile approach such as that will likely be far more enlightening than some RoM estimate. – Cort Ammon - Reinstate Monica Sep 15 '16 at 18:53