Clock Primitives - 2023.1 English

Versal Adaptive SoC Hardware, IP, and Platform Development Methodology Guide (UG1387)

Document ID
UG1387
Release Date
2023-05-24
Version
2023.1 English

Most clocks enter the device through a global clock-capable I/O (GCIO) pin. In horizontal XPIO banks, these clocks directly drive the clock network via a global clock buffer or are transformed by an MMCM, XPLL, or DPLL located in the clock management tiles (CMTs) of the XPIO bank. For devices with columnar HDIO banks, the clocks drive the clock network via a global clock buffer or are transformed by DPLL located in the CMT of the HDIO bank for devices with this feature.

Each horizontal XPIO bank contains the following clocking resources:

  • Clock generation blocks
    • 1 MCMM
    • 2 XPLLs
    • 1 DPLL
  • Global clock buffers
    • 24 BUFGCEs/MBUFGCEs
    • 8 BUFGCTRLs/MBUFGCTRLs
    • 4 BUFGCE_DIVs/MBUFGCE_DIVs
Note: Clocking resources in XPIO banks that exist in the corner of Versal devices are limited and have inaccessible resources such as BUFGCTRLs and BUFGCE_DIVs. To use BUFGCTRLs and BUFGCE_DIVs for clocks sourced from a corner bank, it might be necessary to use a cascaded clocking topology, as described in Cascaded Clock Buffers. Clocking resources in XPIO banks that have unbonded I/Os are available for use. For more information on XPIO corner bank restrictions, see this link in the Versal Adaptive SoC Clocking Resources Architecture Manual (AM003).

For devices with columnar HDIO banks, each bank contains the following clocking resources:

  • Clock generation blocks
    • 1 DPLL
  • Global clock buffers
    • 4 BUFGCEs/MBUFGCEs
Note: DPLL is not available in HDIO banks for VC1902, VC1802, and VM1802 devices.

Each gigabit transceiver (GT*_QUAD) clock region column contains the following clocking resources:

  • Clock generation blocks
    • 1 DPLL
  • Global clock buffers
    • 24 BUFG_GTs/MBUFG_GTs

The following table provides a summary of the Versal device clock buffers.

Table 1. Versal Device Clock Buffers
Versal Device Clock Buffer Leaf-Level Clock Division Support Location Description
BUFGCE Yes using MBUFGCE XPIO and HDIO banks The most commonly used buffer is the BUFGCE, which is a general clock buffer with a clock enable/disable feature.
BUFGCE_DIV Yes using MBUFGCE_DIV XPIO banks The BUFGCE_DIV is useful when a simple division of the clock is required. This clock buffer is considered easier to use and more power efficient than using an MMCM or PLL for simple clock division.
BUFGCTRL Yes using MBUFGCTRL XPIO banks The BUFGCTRL can be instantiated as a BUFGMUX and is generally used when multiplexing two or more clock sources to a single clock network. As with the BUFGCE and BUFGCE_DIV, this clock buffer can drive the clock network for either regional or global clocking.
BUFG_GT Yes using MBUFG_GT GT*_QUAD columns When using clocks generated by GTs, the BUFG_GT clock buffer allows connectivity to the global clock network. In most cases, the BUFG_GT is used as a regional buffer with its loads placed in one or two adjacent clock regions. The BUFG_GT has built-in dynamic clock division capability that you can use in place of an MMCM for clock rate changes.
BUFG_PS Yes using MBUFG_PS Vertical clock column adjacent to PS The BUFG_PS is a simple clock buffer with one clock input (I) and one clock output (O). This clock buffer is a resource for the PS and provides access to the PL clock routing resources for clocks from the processor into the PL. There are up to 12 BUFG_PS buffers available.
BUFG_FABRIC No NoC columns The BUFG_FABRIC is driven by the PL and used for routing high-fanout, non-clock nets, which allows a signal from the PL routing resources to be brought onto the clock network. However, this clock buffer is not for global clocking.

MBUFGs are clocking primitives that allow you to take advantage of the leaf-level clock dividers driven by the local horizontal clock distribution tracks in Versal devices. The leaf-level division results in the use of only one global clock routing resource providing less clock track resource utilization, improved power efficiency, and improved skew between synchronous clock domains. The MBUFG primitives (MBUFGCE, MBUFGCE_DIV, MBUFG_GT, MBUFGCTRL and MBUFG_PS) have four outputs (O1, O2, O3, O4) that configure a clock divider setting of 1, 2, 4, and 8, respectively, for any leaf clock divider driving the clocking loads connected to the MBUFG primitive. For more information on how the MBUFG primitives can help to reduce the high skew between synchronous clock domains, see this link in the Versal Adaptive SoC System Integration and Validation Methodology Guide (UG1388).

To use the MBUFG primitives in your design instead of the standard BUFG primitives, select the MBUFG primitives when running the Versal adaptive SoC Clocking Wizard. The MBUFG primitives are only available when the output frequencies are multiple to each other by a factor of 2, 4, or 8. For more information, see this link in the Clocking Wizard for Versal Adaptive SoC LogiCORE IP Product Guide (PG321).

Note: When the BUFG primitives are already instantiated in the design, you can use logic optimization (opt_design) to transform some BUFG primitives into MBUFG primitives. Transformation using opt_design is only possible in certain cases and includes some limitations. For more information, see the Vivado Design Suite User Guide: Implementation (UG904).

The following figure shows logical and physical implementation views of a MBUFG-driven clocking network. The BUFDIV_LEAF primitives are not represented in the logical netlist but are route-through primitives that are configured by the Vivado router. The CLRB_LEAF input on the MBUFG primitives can be used to asynchronously reset the BUFDIV_LEAF dividers. A signal connected to the MBUFG CLRB_LEAF pin is automatically routed to its connected BUFDIV_LEAF CLR_B pin. Because only the 1x clock is routed on the global clock routing and distribution tracks, MBUFG-driven clocks preserve clocking resources. In addition, the common node for paths that are clocked by two output clocks of the same MBUFG is typically much closer to the driver and the load, reducing clock skew and easing timing closure.

Figure 1. MBUFGCE Logical and Physical View

On device startup, the BUFDIV_LEAF clock dividers are reset, and the MBUFG output clocks start up in the High state. In the following cases, special handling is required to ensure that the BUFDIV_LEAF dividers are reset to their startup state before the MBUFG receives an input clock or is enabled again:

  • If the MBUFG clock buffer or a clock modifying block driving an MBUFG is reset during device operation
  • If the MBUFG-driven clock network is part of a reconfigurable partition

To reset the BUFDIV_LEAF buffers, the CLRB_LEAF pin of the MBUFG must be asserted Low. To ensure proper device operation, the user logic must stop the MBUFG clock before the CLRB_LEAF pin is asserted Low and hold the clock inactive for a predetermined amount after the CLRB_LEAF signal is deasserted High. The time to hold the clock inactive after the CLRB_LEAF signal is deasserted must be greater than the maximum pin delay time that the router reports for routing the signal connected to the MBUFG CLRB_LEAF pin to the BUFDIV_LEAF CLR_B pins. The router reports this time in an INFO message as shown in the following example. A 10 ns delay is sufficient in most cases to satisfy the CLRB_LEAF net route delay.

INFO: [Route 35-3345] MBUFG*/CLRB_LEAF net route delay summary. Please ensure that 
the wait time between de-asserting the CLRB_LEAF signal to each MBUFG and enabling 
the MBUFG output clocks is greater than the delay listed in the table below.
+----------------------+--------------------+----------------------+--------------------+
|  MBUFG Cell          |  Site              |  CLRB_LEAF Net Name  | Max Pin Delay (ns) |
+----------------------+--------------------+----------------------+--------------------+
|  U_hwsim_engine/dcm  |  BUFGCE_DIV_X3Y0   |  U_hwsim_engine/p_1  |  2.252             |
|  _bufg_fx            |                    |  _out                |                    |
+----------------------+--------------------+----------------------+--------------------+

The following figure shows the timing relationship required between the MBUFGCE_DIV CLRB_LEAF and CE signal assertion after CLR has been asserted. The CE signal is held Low to stop the clock until the CLRB_LEAF signal has propagated to all of the BUFDIV_LEAF CLR_B pins. In this example, the CE_TYPE property of the MBUFGCE_DIV is set to SYNC.

Note: In the following figure, the MBUFGCE_DIV/CLRB_LEAF to BUFDIV_LEAF/CLR_B propagation time is due to routing.
Figure 2. Timing Relationship Between Asserting CLRB_LEAF and CE Signals when CLR is Asserted on an MBUFGCE_DIV

You can use the Clock Utilization Report in the Vivado IDE to visually analyze clocking resource utilization and clock routing. The following figure shows the clock resource utilization per clock region overlaid in the Device window. For more information on this report, see the Vivado Design Suite User Guide: Design Analysis and Closure Techniques (UG906).

Figure 3. Clock Utilization Report

For more information on the BUFGCE, MBUFGCE, BUFGCE_DIV, BUFGCTRL, and MBUFGCTRL buffers, see the Versal Adaptive SoC Clocking Resources Architecture Manual (AM003). For details on connectivity and use of the BUFG_GT and MBUFG_GT buffers, see the Versal Adaptive SoC GTY and GTYP Transceivers Architecture Manual (AM002) and Versal Adaptive SoC GTM Transceivers Architecture Manual (AM017).