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A newer version of this document is available.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

2.3.2. Pipelining and Latency

Adding pipeline registers within a path increases the number of clock cycles necessary for a signal value to propagate along the path. Increasing the clock frequency can offset the increased latency.
Figure 31. Hyper-Pipeline Reduced LatencyThis example shows a previous generation Intel FPGA, with a 275 MHz fMAX requirement. The path on the left achieves 286 MHz because the 3.5 ns delay limits the path. Data requires three cycles to propagate through the register pipeline. Three cycles at 275 MHz calculates to 10.909 ns requirement to propagate through the pipeline.


If re-targeting an Intel® Hyperflex™ architecture FPGA doubles the fMAX requirement to 550 MHz, the path on the right side of the figure shows how an additional pipeline stage retimes. The path now achieves 555 MHz, due to the limits of the 1.8 ns delay. The data requires four cycles to propagate through the register pipeline. Four cycles at 550 MHz equals 7.273 ns to propagate through the pipeline.

To maintain the time to propagate through the pipeline with four stages compared to three, meet the 10.909 ns delay of the first version by increasing the fMAX of the second version to 367 MHz. This technique results in a 33% increase from 275 MHz.