5
\$\begingroup\$

It is bad idea to add logic gates in clock signal path. How is clock gating achieve in FPGA and ASIC designs and how does it prevent glitch in the output signal i.e the gated clock as it is enabled or disabled?

\$\endgroup\$
5
  • \$\begingroup\$ In FPGA it usually isn't, you usually design with an enable input to a continuously clocked block. ASIC is a different matter. \$\endgroup\$
    – user16324
    Commented Nov 15, 2021 at 23:05
  • 1
    \$\begingroup\$ This is a few lines that asks for mountains of effort in reply. The site is not for free personal tutoring. This information is readily available to be researched on the Internet. Please edit your question and detail your situation and what you already know, show all that you discovered for yourself on the subject. \$\endgroup\$
    – TonyM
    Commented Nov 15, 2021 at 23:26
  • \$\begingroup\$ How the clock enable is implemented in a particular FPGA or ASIC is probably considered company IP (intellectual property), and so details are not readily available. Though I bet if you just do some web searching you can find design ideas. \$\endgroup\$
    – SteveSh
    Commented Nov 16, 2021 at 1:15
  • \$\begingroup\$ Look all I want to know, for specific FPGA e.g Intel Cyclone 10, or Xilinx Spartan 6, how is clock gating achieved physically using the hard IP that exists inside it. \$\endgroup\$
    – quantum231
    Commented Nov 16, 2021 at 13:01
  • 1
    \$\begingroup\$ Look, all you have to do is read the user manuals then. Put the effort in yourself instead of demanding personal reports from the community here. Unsurprisingly, this is being voted to closed. \$\endgroup\$
    – TonyM
    Commented Nov 16, 2021 at 13:41

3 Answers 3

12
\$\begingroup\$

Is it a bad idea to gate clocks? It depends.

In the ASIC there’s well-understood timing for clock paths, so it’s reasonable to instance a standard cell on the clock tree to gate a sub-region’s clock. On ASIC then, not only is clock gating not ‘a bad idea’, it’s widely used as a means to save power.

Not so much with the FPGA. In fact, it’s never a good idea to create gated clocks directly out of FPGA fabric logic; the synthesis tools will warn you about it if not outright forbid it. Why? The resulting inserted skew becomes impossible to manage at higher frequencies, even if the gated clock doesn’t glitch (which it will without careful design.)

This brings up a common issue: modeling ASIC clock gating on FPGAs. It isn’t really feasible to just define the clock gate in HDL and hope for the best. It needs special handing.

You can model ASIC-like gated-clock behavior in your FPGA using clock-enable flops for your synchronous blocks. This can be dealt with as a synthesis option in your flow, which will identify the gated clock domain and convert its flops to FDCEs. Vivado example: https://support.xilinx.com/s/article/982650?language=en_US

Some FPGAs do support clock gating, using dedicated clock gate resources with predictable timing and glitch-free behavior. More here from Xilinx (look for BUFGCE): https://www.xilinx.com/support/documentation/user_guides/ug572-ultrascale-clocking.pdf ; other FPGAs will be similar.

Note that with BUFGCE, you’re still obliged to meet clock setup/hold from clock enable to clock rising edge. Still, that’s an easier constraint to meet than making an asynchronous clock gate out of fabric logic. If you’re modeling an ASIC you have to account for the difference between BUFGCE and whatever standard cell you’re using in the ASIC.

Finally, you asked how clock gating is actually done. Tl, dr: the enable is latched to prevent disturbing the clock pulse with a glitch. More here: https://anysilicon.com/the-ultimate-guide-to-clock-gating/

\$\endgroup\$
10
\$\begingroup\$

It is bad idea to add logic gates in clock signal path.

Do not add logic gates to your signal path. Strategically placed clock-gating and clock-divider are sometimes exceptions, but should be used minimally if ever. The exceptions are more likely to be acceptable for large ASICs where power is a concern. FPGAs and small ASICs generally discourage any logic added to the clock.

How is clock gating achieve in FPGA and ASIC designs and how does it prevent glitch in the output signal i.e the gated clock as it is enabled or disabled?

On ASICs clock gating is achieved with a negedge flip-flop and an AND gate (low when disabled) or posedge flip-flop and an OR gate (high when disabled). Glitches are avoided because the toggle to the enable/disable happens when there is no impact on the output. If you find an FPGA that supports clock gating, then it likely does it the same way.

If you are to use clock gating, it needs to be done to large modules; not individual flip-flops. Adding any logic to a clock creates additional timing offset between gated-clock circuits those using the clock directly. If your synthesis tool does not have great clock gating support, then there is a chance it will mess up your clock tree and will require async FIFOs between modules of different gated-clocks even if the are derived from the same clock.

\$\endgroup\$
4
  • \$\begingroup\$ FPGAs that support clock gating are far from rare these days, they do it using clocking buffers and the synthesis tools recognize the same HDL constructs that you describe for ASIC gating. Clocking buffers are basically the same gates you describe optimized for glitchless transitions between having the clock enabled and disabled. \$\endgroup\$
    – DonFusili
    Commented Nov 16, 2021 at 8:38
  • 4
    \$\begingroup\$ Do not add logic gates to your signal path You mean 'Do not add logic gates to your clock path'. \$\endgroup\$
    – TonyM
    Commented Nov 16, 2021 at 9:25
  • \$\begingroup\$ Last time I worked on this it was a half-latch rather than a whole flipflop, but the principle of latching the enable during the active half of the clock is the same. \$\endgroup\$
    – pjc50
    Commented Nov 16, 2021 at 16:10
  • \$\begingroup\$ Agree with pjc50, using TSMC standard cells (28nm, 16nm, 7nm) I’ve only seen clock gates implemented as latch + AND (I think) gate. \$\endgroup\$
    – Michael
    Commented Nov 16, 2021 at 17:34
8
\$\begingroup\$

It is bad idea to add logic gates in clock signal path.

In an FPGA, not only is it a horrible idea, the design software should not let you do it.

how does it prevent glitch in the output signal i.e the gated clock as it is enabled or disabled?

It doesn't. What you do instead of gating the clock is to enable the logic which the clock drives. Don't worry about gating the input logic to a flip-flop, either. Just use a flip-flop with an enable input and drive that.

\$\endgroup\$
1
  • \$\begingroup\$ Will using flip flop with clock enable, or an enable input actually result in power saving? \$\endgroup\$
    – quantum231
    Commented Mar 22, 2022 at 20:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.