Intel® Hyperflex™ Architecture High-Performance Design Handbook

ID 683353
Date 10/04/2021
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

8.2.1.1. Synchronous Resets Summary

Synchronous clears can limit the amount of retiming. There are two issues with synchronous clears that cause problems for retiming:

  • A short path, usually traveling directly from the source register to the destination register without any logic between them. Short paths are not normally a problem, because the Compiler retimes the positive slack to longer paths. This retiming improves performance. However, short paths typically connect to long data paths that require retiming. By retiming many registers along long paths, registers push down or pull up this short path. This problem is not significant in normal logic, but becomes significant when synchronous clears have large fan-outs.
  • Synchronous clears have large fan-outs. When aggressive retiming pushes registers up or down the synchronous clear path, paths can clutter until they cannot accept more registers. This situation results in path length imbalances, and the Compiler can pull no more registers from the synchronous clear paths.

Aggressive retiming occurs when the Compiler retimes a second register through the ALM register.

Figure 138. Aggressive Retiming

Intel® Hyperflex™ architecture FPGAs have a dedicated Hyper-Register on the SCLR path, with the ability to place the ALM register into bypass mode. This ability allows you to push and pull this register. If you push the register forward, then you must pull a register down the SCLR path and merge the two. If you push the register back, then you must push a duplicate register up the SCLR path. You can use both of these options. However, you create bottlenecks when multiple registers push and pull registers up and down the synchronous clear routing.

Use resets in a practical manner. Control logic mostly requires synchronous reset. Logic that may not require a synchronous reset helps with timing. Refer to the following guidelines about synchronous resets:

  • Avoid synchronous resets in new code that must run at high speed. This limitation generally applies to data path logic that flushes out while the system is in reset, or logic with values that the system ignores when coming out of reset.
  • Control logic often requires a synchronous reset, so there is no avoiding the reset in that situation.
  • For existing logic that runs at high speeds, remove the resets wherever possible. If you do not understand the logic behavior at reset, retail the synchronous reset. Only remove the synchronous clear if a timing issue arises.
  • Pipeline the synchronous clear. This technique does not help when you must pull registers back, but can help when you need to pull registers forward into the data path.
  • Duplicate synchronous clear logic for different hierarchies. This technique limits the fan-out of the synchronous clear, so that the Compiler can retime the clear with the local logic. Apply this technique only after you determine that an existing synchronous clear with large fan-out limits retiming. This technique is not difficult on the back-end, because the technique does not change design functionality.
  • Duplicate synchronous clear for different clock domain and inverted clocks. This technique can overcome some retiming restrictions due to boundary or multiple period requirement issues.