External Memory Interface Handbook Volume 2: Design Guidelines: For UniPHY-based Device Families

ID 683385
Date 3/06/2023
Document Table of Contents

1.2.6. PLLs and Clock Networks

The exact number of clocks and PLLs required in your design depends greatly on the memory interface frequency, and on the IP that your design uses.

For example, you can build simple DDR slow-speed interfaces that typically require only two clocks: system and write. You can then use the rising and falling edges of these two clocks to derive four phases (0°, 90°, 180°, and 270°). However, as clock speeds increase, the timing margin decreases and additional clocks are required, to optimize setup and hold and meet timing. Typically, at higher clock speeds, you need to have dedicated clocks for resynchronization, and address and command paths.

Intel® FPGA memory interface IP uses one PLL, which generates the various clocks needed in the memory interface data path and controller, and provides the required phase shifts for the write clock and address and command clock. The PLL is instantiated when you generate the Intel® FPGA memory IPs.

By default, the memory interface IP uses the PLL to generate the input reference clock for the DLL, available in all supported device families. This method eliminates the need of an extra pin for the DLL input reference clock.

The input reference clock to the DLL can come from certain input clock pins or clock output from certain PLLs.

Note: Intel® recommends using integer PLLs for memory interfaces; handbook specifications are based on integer PLL implementations.

For the actual pins and PLLs connected to the DLLs, refer to the External Memory Interfaces chapter of the relevant device family handbook.

You must use the PLL located in the same device quadrant or side as the memory interface and the corresponding dedicated clock input pin for that PLL, to ensure optimal performance and accurate timing results from the Intel® Quartus® Prime software.

The input clock to the PLL can fan out to logic other than the PHY, so long as the clock input pin to the PLL is a dedicated input clock path, and you ensure that the clock domain transfer between UniPHY and the core logic is clocked by the reference clock going into a global clock.