Saturday 30 August 2014

Low Power Techniques : Clock Gating

Disclaimer : All the articles are written with an assumption that the reader has a basic knowledge of Boolean Gates and digital elements.

In this article, the key focus of discussion would be clock gating and its impact on the design on modern day SoCs. Clocks could be considered as the Arteries (or Veins so for the matter) of the entire SoC and clock source being the heart of any SoC. All the IP’s like CPU, Audio, Display, USB, etc require a clock to function. It is the “CLOCK” which gets to decide the maximum speed of operation for any SoC (Here I am not considering the throughput increase due to the various kinds of architectures employed in designing the SoC) and hence special attention should be given to this signal.

Another reason why this signal should be given more attention is due the fact that it is the highest toggling signal in any SoC and therefore consumes the largest fraction of entire SoC power. In a typical SoC for mobile application, the clock tree consumes 30-40% of the power (Yes, this number can be more depending on the Clocking architecture employed).

Clock gating is a technique to turn the clock OFF depending on the requirement or the use case. How this is achieved is the main focus of this topic.


There are essentially two ways to gate the clock. One through hardware control and one through software control.

The software control is through registers which are used to control the PLL or DLL or any other clock source that we may have. We turn the PLL/DLL/Source off (which is in itself at times a complex process) by following a specific sequence. Turning of the source is a method to gate the trunk of the clock tree (I’ll talk about the Industry lingo on clocks in an upcoming article. Kindly bear with us on terms like trunk, branch, etc).
It may happen that in some of the cases we want to shut only a branch of the trunk. This is the point where hardware control comes in picture.

This article (and some subsequent articles) will discuss on hardware control of the clock.

Below is shown a free running clock :



Fig 1. Free Running Clock
---------------------------------------------------------------------------------------
How can we actually stop it based on some control signal say “enable”?
---------------------------------------------------------------------------------------

Yes, you got it right, we need to AND the two signals and put the enable to be at “1” when we need the clock and “0” when we don’t want the clock output for any IP.
The figure below shows a simple AND gate being used as a gate.


                                       Fig 2. AND Gate as a Clock Gate and Problem of Glitch

So do we actually use this “AND” gate in our SoCs to gate the clock from reaching to any IP?

You guessed it right. No, this is not the standard cell which any modern day SoC use. The reason is clear from above figure itself. If the enable signal changes while the clock is high, this will lead to a runt pulse termed as “Glitch”. A glitch is harmful for any SoC as it can lead to unknown states or the states which we don’t want our system to be in. For now, I’ll leave the glitch to a-not-so-far-in future article.

I’ll discuss about “Clock Gating Cell” in the next article.


Till then, Enjoy!!! 

No comments:

Post a Comment