Standard Clock Gate Cell

Moving ahead from my last article on Clock gating(You can read the previous article here : Low Power Techniques - Clock Gating), this small article introduces the standard “Clock Gate Cell” This cell is available to all the designers in semiconductor industry along with the other standard cells in their libraries. It is good to recognize early that clock and power are mingled up and clocks provide a good deal of avenues for saving power. So knowing more about clocks is always a plus to person interested in designing "Power Aware Circuits".

In the last article, we saw that there was a glitch in “certain cases”, when we gated the clock. These “certain cases” led to the development of this cell. If you look more closely at the waveform given in last article (reproduced below for convenience), the glitch will always be there whenever the enable signal goes down and the clock is high (high as in logic 1) .

Fig 1. AND Gate as a Clock Gate and Problem of Glitch (Reproduced from Low Power Techniques- Clock Gating)

The above observation is simple enough to propose a solution for the glitch problem in AND gate. Fig 2. shows the clock gate cell. 

Fig 2. Clock Gate Cell

This cell stops the clock without any glitch in the output.  And how does it do that? It utilizes a LATCH for this purpose. 

-----------------------------

Latching Mechanism

-----------------------------

A latch is a level sensitive device in which the output is same as input till its Enable signal is at a particular logic level. Generally logic 1 is the chosen level for this matter. The output traces the input till the Enable signal is logic 1 and as enable is switched to logic 0, the output retains the last value it captured from the input till the next enable comes where it starts tracking the input again.

------------------------------------------------------------------

Latching Mechanism as used in Clock Gate Cell

------------------------------------------------------------------

For clock gating purpose the enable signal of latch is tied to the inverted clock signal (Clk_in) so that output changes only when the clock is at a particular logic level (which is logic 0 in our case). So it is easy to see that we can pass Enable signal and the changes will be recorded only when the logic level is low (or logic 0) and then AND this with the clock signal. This will ensure that that the Enable low is recorded only when the Clk_in is low and hence no glitch in the output.

Note that we have inverted the clock before sending it to the latch so as to ensure that the latch is sensitive to logic 0 of Clk_in (since the latch itself is sensitive to logic 1). Before I proceed further, I would like to point out that a latch is a very powerful tool available to designers and it is used in multiple circuits (it will be a hot topic in some of the coming articles). With above cell Fig 1. modifies to Fig 3. as below :

Fig 3. Timing Waveform for Clock Gate

In the above figure notice how the enable edges which happened during the clock high have been translated or rather delayed in the clock low domain.

We can use this cell to TURN OFF an entire clock domain or TURN OFF only a few flops. It is entirely on designer that how much power he wants to save and according to that we have levels of clock gating in any SoC. Also note that these Enable signals are controlled by Finite State Machines (FSMs) which trigger based on inputs from other FSMs in SoC.

A small point to all the above information is : Clock gate is a HW means to control the clock signal there are software means as well in which we program the PLL itself to control the entire clock tree. 

I conclude this article here with these lines of Dr. Thomas Fuller :

"Let not thy will roar, when thy power can but whisper"

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!!!