Gap 7 MMCM jitter cleaner + CIC comb CREG pipeline + XDC clock-name fix
MMCM (Gap 7): - Add adc_clk_mmcm.v: MMCME2_ADV wrapper (VCO=800MHz, CLKOUT0=400MHz) - Modify ad9484_interface_400m.v: replace BUFG with MMCM path, gate reset on mmcm_locked - Add adc_clk_mmcm.xdc: CDC false paths for clk_mmcm_out0 <-> clk_100m XDC Fix (Build 19 WNS=-0.011 root cause): - Remove conflicting create_generated_clock -name clk_400m_mmcm - Replace all clk_400m_mmcm references with Vivado auto-generated clk_mmcm_out0 - CDC false paths now correctly apply to actual timing paths CIC CREG Pipeline (Build 18 critical path fix): - Explicit DSP48E1 for comb[0] with CREG=1/AREG=1/BREG=1/PREG=1 - Absorbs integrator_sampled_comb fabric FDRE into DSP48 C-port register - Eliminates 0.643ns fabric->DSP routing delay (Build 18 tightest path) - +1 cycle comb latency via data_valid_comb_0_out pipeline - Move shared register declarations above ifndef SIMULATION (iverilog fix) - Update golden data for +1 cycle CIC pipeline shift Build scripts: build19_mmcm.tcl, build20_mmcm_creg.tcl Regression: 18/18 FPGA pass, 20/20 MCU pass Build 20 launched on remote Vivado (pending results)
This commit is contained in:
Jason committed
1 parent
f3bbf77ca1
commit
c6103b37de
10 files changed
+9038
-7438
No files matched your search
@@ -0,0 +1,164 @@
|
||||
# ADC Clock MMCM Integration Guide
|
||||
|
||||
## Overview
|
||||
|
||||
`adc_clk_mmcm.v` is a drop-in MMCME2_ADV wrapper that replaces the direct BUFG
|
||||
on the 400 MHz ADC data clock output with a jitter-cleaning PLL loop.
|
||||
|
||||
### Current clock path (Build 18)
|
||||
```
|
||||
adc_dco_p/n → IBUFDS → BUFIO (drives IDDR only — near-zero delay)
|
||||
→ BUFG (drives all fabric 400 MHz logic)
|
||||
```
|
||||
|
||||
### New clock path (with MMCM)
|
||||
```
|
||||
adc_dco_p/n → IBUFDS → BUFIO (unchanged — drives IDDR only)
|
||||
→ MMCME2 CLKIN1 → CLKOUT0 → BUFG (fabric 400 MHz)
|
||||
→ CLKFBOUT → BUFG → CLKFBIN (feedback)
|
||||
```
|
||||
|
||||
## Expected Timing Improvement
|
||||
|
||||
| Parameter | Before (Build 18) | After (estimated) |
|
||||
|-----------|-------------------|-------------------|
|
||||
| Input jitter | 50 ps | 50 ps (unchanged) |
|
||||
| Output jitter (MMCM) | N/A | ~20-30 ps |
|
||||
| Clock uncertainty | 43 ps | ~25 ps |
|
||||
| WNS (setup, 400 MHz) | +0.062 ns | ~+0.08 to +0.10 ns |
|
||||
| WHS (hold, 100 MHz) | +0.059 ns | unchanged |
|
||||
|
||||
The improvement comes from reduced clock uncertainty in the Vivado timing
|
||||
analysis. The MMCM's PLL loop attenuates the input jitter, so Vivado deducts
|
||||
less uncertainty from the timing budget.
|
||||
|
||||
## MMCM Configuration
|
||||
|
||||
```
|
||||
CLKIN1 = 400 MHz (2.500 ns)
|
||||
DIVCLK_DIVIDE = 1
|
||||
CLKFBOUT_MULT_F = 2.0 → VCO = 800 MHz
|
||||
CLKOUT0_DIVIDE_F = 2.0 → Output = 400 MHz
|
||||
BANDWIDTH = HIGH (maximum jitter filtering)
|
||||
```
|
||||
|
||||
VCO at 800 MHz is well within the Artix-7 -2 speed grade range (600–1200 MHz).
|
||||
|
||||
## Resource Cost
|
||||
|
||||
| Resource | Count | Notes |
|
||||
|----------|-------|-------|
|
||||
| MMCME2_ADV | 1 | Was 0/10, now 1/10 |
|
||||
| BUFG | +1 (feedback) | Was 4/32, now 5/32 |
|
||||
| FFs | 0 | No additional fabric registers |
|
||||
|
||||
## Integration Steps
|
||||
|
||||
### Step 1: Modify `ad9484_interface_400m.v`
|
||||
|
||||
Replace the BUFG instantiation with `adc_clk_mmcm`:
|
||||
|
||||
```verilog
|
||||
// REMOVE these lines (65-69):
|
||||
// BUFG bufg_dco (
|
||||
// .I(adc_dco),
|
||||
// .O(adc_dco_buffered)
|
||||
// );
|
||||
// assign adc_dco_bufg = adc_dco_buffered;
|
||||
|
||||
// ADD this instead:
|
||||
wire mmcm_locked;
|
||||
wire adc_dco_mmcm;
|
||||
|
||||
adc_clk_mmcm mmcm_inst (
|
||||
.clk_in (adc_dco), // From IBUFDS output
|
||||
.reset_n (reset_n),
|
||||
.clk_400m_out (adc_dco_mmcm), // Jitter-cleaned 400 MHz
|
||||
.mmcm_locked (mmcm_locked)
|
||||
);
|
||||
|
||||
// Use MMCM output for all fabric logic
|
||||
wire adc_dco_buffered = adc_dco_mmcm;
|
||||
assign adc_dco_bufg = adc_dco_buffered;
|
||||
```
|
||||
|
||||
### Step 2: Gate reset on MMCM lock (recommended)
|
||||
|
||||
In `ad9484_interface_400m.v`, modify the reset synchronizer to require MMCM lock:
|
||||
|
||||
```verilog
|
||||
// Change the reset synchronizer input from:
|
||||
// always @(posedge adc_dco_buffered or negedge reset_n) begin
|
||||
// if (!reset_n)
|
||||
// reset_sync_400m <= 2'b00;
|
||||
// else
|
||||
// reset_sync_400m <= {reset_sync_400m[0], 1'b1};
|
||||
// end
|
||||
|
||||
// To:
|
||||
wire reset_n_gated = reset_n & mmcm_locked;
|
||||
|
||||
always @(posedge adc_dco_buffered or negedge reset_n_gated) begin
|
||||
if (!reset_n_gated)
|
||||
reset_sync_400m <= 2'b00;
|
||||
else
|
||||
reset_sync_400m <= {reset_sync_400m[0], 1'b1};
|
||||
end
|
||||
```
|
||||
|
||||
This ensures the 400 MHz domain stays in reset until the MMCM has locked and
|
||||
the clock is stable. Without this, the first ~10 µs after power-up (before
|
||||
MMCM lock) could produce glitchy clock edges.
|
||||
|
||||
### Step 3: Add constraint file
|
||||
|
||||
Add `constraints/adc_clk_mmcm.xdc` to the Vivado project. Uncomment the
|
||||
constraints and adjust hierarchy paths based on your actual instantiation.
|
||||
|
||||
Key constraints to uncomment:
|
||||
1. `create_generated_clock` (or verify Vivado auto-creates it)
|
||||
2. `set_max_delay` between `adc_dco_p` and `clk_400m_mmcm`
|
||||
3. `set_false_path` between `clk_400m_mmcm` and other clock domains
|
||||
4. `set_false_path` on `LOCKED` output
|
||||
|
||||
### Step 4: Add to build script
|
||||
|
||||
In the Tcl build script, add the new source file:
|
||||
|
||||
```tcl
|
||||
read_verilog adc_clk_mmcm.v
|
||||
read_xdc constraints/adc_clk_mmcm.xdc
|
||||
```
|
||||
|
||||
### Step 5: Verify
|
||||
|
||||
After building:
|
||||
1. Check `report_clocks` — should show the new MMCM-derived clock
|
||||
2. Check `report_clock_interaction` — verify no unexpected crossings
|
||||
3. Check WNS on the `adc_dco_p` / MMCM clock group — should improve
|
||||
4. Check MMCM locked in ILA during bring-up
|
||||
|
||||
## BUFIO Compatibility Note
|
||||
|
||||
The BUFIO path for IDDR capture is **not affected** by this change. BUFIO
|
||||
drives only IOB primitives (IDDR) and cannot go through an MMCM. The BUFIO
|
||||
continues to use the raw IBUFDS output with near-zero insertion delay, which
|
||||
is correct for source-synchronous DDR capture.
|
||||
|
||||
The re-registration from BUFIO domain to BUFG domain (lines 105-108 of
|
||||
`ad9484_interface_400m.v`) now crosses from the raw `adc_dco_p` clock to the
|
||||
MMCM-derived clock. Since both are frequency-matched and the MMCM is locked
|
||||
to the input, this is a safe single-register transfer. The `set_max_delay`
|
||||
constraint in the XDC ensures Vivado verifies this.
|
||||
|
||||
## Simulation
|
||||
|
||||
Under `SIMULATION` define (iverilog), the module passes the clock straight
|
||||
through with a simulated lock delay of ~4096 cycles. This matches the
|
||||
current testbench behavior — no changes to any testbenches needed.
|
||||
|
||||
## Rollback
|
||||
|
||||
To revert: simply restore the original BUFG in `ad9484_interface_400m.v` and
|
||||
remove `adc_clk_mmcm.v` + `constraints/adc_clk_mmcm.xdc` from the project.
|
||||
No other files are affected.
|
||||
Reference in new issue
Block a user