The ECP5’s global clock routing is split into a number of parts. A high level overview of its structure is contained in Lattice Technical Note TN1263.
From a point of view of clock distribution, the device is split into four quadrants: upper left (UL), upper right (UR), lower left (LL) and lower right (LR). Throughout this document, qq refers to any quadrant named in this way, s refers to a side of the device (B, T, L, or R).
This document is a work in progress! Some information is still incomplete and subject to change.
The mid muxes are located in the middle of each edge of the device, and take clock inputs from a range of sources (clock input pins, PLL ouputs, CLKDIV outputs, etc) and feed them into the centre clock muxes.
Depending on the location, mid muxes have between 12 and 16 outputs to the centre muxes.
Between each mid mux multiplexer output and the centre mux is a Dynamic Clock Control component (DCC), which provides glitch free clock disable.
Inputs to the mid muxes are named as follows:
- PCLKx_y for the dedicated clock input pins
- qqC_PCLKGPLLx_y for the PLL outputs
- s_CDIVX_x for the clock divider outputs
- qqQ_PCLKCIB_x and qqM_PCLKCIBx have unknown function, most likely a connection to fabric (TODO: check)
- PCSx_TXCLK_y and PCSx_RXCLK_y are the SERDES transmit and receive clocks
- SERDES_REFCLK_x are the SERDES reference clocks
Outputs from the muxes themselves into the DCCs are named sDCC00CLKI through sDCCnnCLKI.
The outputs from the DCCs into the centre muxes are named HPFE0000 through HPFE1300 for the left side; HPFW0000 through HPFW1300 for the right side; VPFN0000 through VPFN1500 for the bottom side and VPFS0000 through VPFS1100 for the top side.
The left and right muxes are located in a single tile, LMID_0 and RMID_0 respectively. The top side is split between TMID_0 and TMID_1, and the bottom side between BMID_0 and BMID_2V.
The centre muxes select the 16 global clocks for each quadrant from the outputs of the mid muxes and from a few other sources. They also contain a total of two Dynamic Clock Select blocks, for glitch-free clock switching.
There are four fabric entries to the centre muxes from each of the four quadrants. These connect through DCCs in the same way as the mid mux outputs (but in this case the DCC is considered as part of the centre mux rather than mid mux).
The fabric entries come from nearby CIBs, but the exact net still need to be traced.
Inputs to the mid muxes (and DCSs) are named as follows:
- HPFExx00, HPFWxx00, VPFNxx00 and VPFSxx00 for the mid mux (via DCC) outputs
- qqCPCLKCIB0 for the fabric clock inputs (via DCC)
- DCSx for the DCS outputs
- VCC_INT for constant one
Outputs from the centre muxes, going into the spine tiles, are named qqPCLK0 through qqPCLK15.
Centre muxes also contain CLKTEST, a component with unknown function used in Lattice’s testing.
The exact split of centre mux functionality between tiles varies depending on device, this is an example for the LFE5U-85F.
|MIB_R22C66||EBR_CMUX_UL||DCS0 config, CLKTEST||Shared with EBR|
|MIB_R22C67||CMUX_UL_0||UL clocks mux, DCS0CLK0 input mux, DCC2|
|MIB_R22C68||CMUX_UR_0||UR clocks mux, DCS0CLK1 input mux, DCC3|
|MIB_R22C69||EBR_CMUX_UR||CLKTEST||Shared with EBR|
|MIB_R70C66||EBR_CMUX_LL||DCS1 config, CLKTEST||Shared with EBR|
|MIB_R70C67||CMUX_LL_0||LL clocks mux, DCS1CLK0 input mux, DCC0|
|MIB_R70C68||CMUX_LR_0||LR clocks mux, DCS1CLK1 input mux, DCC1|
|MIB_R70C69||EBR_CMUX_LR||CLKTEST||Shared with EBR|
The outputs from the centre muxes go horizontally into each quadrant and connect to a few spine tiles (there are three spine tiles per quadrant in the 85k part). Spine tiles are located adjacent to a column of TAP_DRIVEs (see next section) and are combined with another function (EBR or DSP).
The purpose of a spine tile is to selectively connect the centre mux outputs to the vertical global wires feeding the adjacent column of TAP_DRIVEs through a buffer. There is one buffer per wire, with a 1:1 mapping - the only reason spine tiles are configurable is to disable unused buffers to save power, there is no other routing or selection capability in them.
The inputs to spine tiles from the quadrant’s centre mux are named qqPCLK0 through qqPCLK15.
The outputs are named VPTX0000 through VPTX15000, which feed a column of tap drives.
TAP_DRIVE tiles are arranged in columns throughout the device.
The purpose of TAP_DRIVE tiles is to selectively connect the vertical global wires coming out of the adjacent spine tile to horizontal wires serving tiles to the left and right of the TAP_DRIVE. A TAP_DRIVE will typically serve a row of about 20 tiles, 10 to the left and 10 to the right.
Like spine tiles, TAP_DRIVE tiles have a 1:1 input-output mapping and only offer the ability to turn on/of buffers to save power.
The outputs are named HPBX0000 through HPBX15000, with a net location on the left or right for the left or right outputs (signified as L_ or R_ in the Project Trellis database.
Non-Clock Global Usage¶
Inside PLBs, global nets can not only be connected to the clock signal, but also to clock enable, set/reset and general local wires. This does not seem to be commonly used by Diamond, which prefers to use general routing and the GSR signal.
Not all globals can be used for all functions, the allowed usage depending on net is shown below.