Constrain a two-clock block: time its ports, cut the raw reset and keep both synchronized releases timed.
Hardware context: Two-flop reset synchronizer releasing a data flop. Cut the raw reset input while keeping the synchronized release timed by recovery and removal.
The block rst2 has one raw reset input, rst_n, from a board button, and two clock domains. On clka (2.0 ns), the synchronizer ra1, ra2 releases the data flop fa. On clkb (3.0 ns), the synchronizer rb1, rb2 releases the data flop fb.
The starter SDC is empty. Data on da arrives 0.2 ns after a clka edge, and qa needs an output delay of 0.3 ns. On clkb, db arrives after 0.4 ns and qb needs 0.5 ns. Give rst_n an input delay of 0.5 ns against clka so that every port is constrained. Then make the reset timing honest in both domains.
Requirements
| Requirement | Limit | Unit |
|---|---|---|
| Clocks defined | = 2 | count |
| Input ports without an input delay | = 0 | count |
| Unconstrained endpoints | = 0 | count |
| Recovery slack of the release into fa (clka) | same as the reference ± 0.001 | ns |
| Removal slack of the release into fa (clka) | same as the reference ± 0.001 | ns |
| Recovery slack of the release into fb (clkb) | same as the reference ± 0.001 | ns |
| Removal slack of the release into fb (clkb) | same as the reference ± 0.001 | ns |
| Recovery checks reported from the raw reset into ra1 | = 0 | count |
| Removal checks reported from the raw reset into ra1 | = 0 | count |
| Recovery checks reported from the raw reset into rb1 | = 0 | count |
| Removal checks reported from the raw reset into rb1 | = 0 | count |
Live analysis: Live OpenSTA at tt_025C_1v80 with ideal clocks and no wire parasitics. Graded: the clock count and check_setup coverage. Also graded: the recovery and removal checks of both releases, and every such check from the raw reset. The release slacks must equal the reference, so any extra constraint that moves them fails.
The Netlist
Bundled design rst2. Submit also times your SDC on 3 hidden variants with the same instance and port names.
// rst2: one raw reset, two clock domains. rst_n clears both synchronizers at once; ra1/ra2 release fa on clka and
// rb1/rb2 release fb on clkb (T3.1.1: one synchronizer per clock domain).
module rst2 (clka, clkb, rst_n, da, db, qa, qb);
input clka; input clkb; input rst_n; input da; input db; output qa; output qb;
wire one, a1, a2, b1, b2;
sky130_fd_sc_hd__conb_1 tie (.HI(one));
sky130_fd_sc_hd__dfrtp_1 ra1 (.CLK(clka), .D(one), .RESET_B(rst_n), .Q(a1));
sky130_fd_sc_hd__dfrtp_1 ra2 (.CLK(clka), .D(a1), .RESET_B(rst_n), .Q(a2));
sky130_fd_sc_hd__dfrtp_1 fa (.CLK(clka), .D(da), .RESET_B(a2), .Q(qa));
sky130_fd_sc_hd__dfrtp_1 rb1 (.CLK(clkb), .D(one), .RESET_B(rst_n), .Q(b1));
sky130_fd_sc_hd__dfrtp_1 rb2 (.CLK(clkb), .D(b1), .RESET_B(rst_n), .Q(b2));
sky130_fd_sc_hd__dfrtp_1 fb (.CLK(clkb), .D(db), .RESET_B(b2), .Q(qb));
endmodule
| Graded number | Meaning |
|---|---|
| clocks | Clocks defined |
| input_missing | Inputs with no input delay (check_setup) |
| unconstrained | Unconstrained endpoints (check_setup) |
| rel_a_rec | Worst recovery slack at fa/RESET_B, ns |
| rel_a_rem | Worst removal slack at fa/RESET_B, ns |
| rel_b_rec | Worst recovery slack at fb/RESET_B, ns |
| rel_b_rem | Worst removal slack at fb/RESET_B, ns |
| raw_a_rec_paths | Number of recovery paths reported to ra1/RESET_B |
| raw_a_rem_paths | Number of removal paths reported to ra1/RESET_B |
| raw_b_rec_paths | Number of recovery paths reported to rb1/RESET_B |
| raw_b_rem_paths | Number of removal paths reported to rb1/RESET_B |
Full specification
| Graded result | What it reads |
|---|---|
| Clocks defined | clocks = 2 |
| Input ports without an input delay | input_missing = 0 |
| Unconstrained endpoints | unconstrained = 0 |
| Recovery slack of the release into fa (clka) | rel_a_rec: same as the reference ± 0.001 on every netlist |
| Removal slack of the release into fa (clka) | rel_a_rem: same as the reference ± 0.001 on every netlist |
| Recovery slack of the release into fb (clkb) | rel_b_rec: same as the reference ± 0.001 on every netlist |
| Removal slack of the release into fb (clkb) | rel_b_rem: same as the reference ± 0.001 on every netlist |
| Recovery checks reported from the raw reset into ra1 | raw_a_rec_paths = 0 |
| Removal checks reported from the raw reset into ra1 | raw_a_rem_paths = 0 |
| Recovery checks reported from the raw reset into rb1 | raw_b_rec_paths = 0 |
| Removal checks reported from the raw reset into rb1 | raw_b_rem_paths = 0 |
Live OpenSTA at tt_025C_1v80 with ideal clocks and no wire parasitics. Graded: the clock count and check_setup coverage. Also graded: the recovery and removal checks of both releases, and every such check from the raw reset. The release slacks must equal the reference, so any extra constraint that moves them fails.
Run times your SDC on the public netlist. Submit also times it on 3 hidden variants of the block and compares what OpenSTA reports.
Examples
Example 1
A Wildcard False Path on Every Reset Pin- Input
The clocks and port delays plus set_false_path -to [get_pins */RESET_B]
- Output
Clocks defined: 2 Input ports without an input delay: 0 Unconstrained endpoints: 0 Recovery slack of the release into fa (clka): not reported Removal slack of the release into fa (clka): not reported Recovery slack of the release into fb (clkb): not reported Removal slack of the release into fb (clkb): not reported Recovery checks reported from the raw reset into ra1: 0 Removal checks reported from the raw reset into ra1: 0 Recovery checks reported from the raw reset into rb1: 0 Removal checks reported from the raw reset into rb1: 0
check_setup is clean, but neither synchronized release has a recovery or removal check left.
Example 2
A Hold-Only False Path on the Raw Reset- Input
The clocks and port delays plus set_false_path -hold -from [get_ports rst_n]
- Output
Clocks defined: 2 Input ports without an input delay: 0 Unconstrained endpoints: 0 Recovery slack of the release into fa (clka): 1.9039 ns Removal slack of the release into fa (clka): 0.0037 ns Recovery slack of the release into fb (clkb): 2.9039 ns Removal slack of the release into fb (clkb): 0.0037 ns Recovery checks reported from the raw reset into ra1: 1 Removal checks reported from the raw reset into ra1: 0 Recovery checks reported from the raw reset into rb1: 1 Removal checks reported from the raw reset into rb1: 0
The removal checks from the button are gone, but its recovery checks are still in the report. -hold cuts only the min checks.
Example 3
One Clock for Both Domains- Input
One create_clock on both clock ports, with the same delays and false path
- Output
Clocks defined: 1 Input ports without an input delay: 0 Unconstrained endpoints: 0 Recovery slack of the release into fa (clka): 1.9039 ns Removal slack of the release into fa (clka): 0.0037 ns Recovery slack of the release into fb (clkb): 1.9039 ns Removal slack of the release into fb (clkb): 0.0037 ns Recovery checks reported from the raw reset into ra1: 0 Removal checks reported from the raw reset into ra1: 0 Recovery checks reported from the raw reset into rb1: 0 Removal checks reported from the raw reset into rb1: 0
Everything is timed, but the clkb domain is checked against a 2 ns clock instead of its own 3 ns.
Submission Rules
- Submit one SDC file for the supplied netlist. The commands and options are the ones Try it accepts; anything else is refused before the tool runs.
- Run times your SDC on the public netlist. Submit also times it on 3 hidden variants with the same instance and port names. A variant may use other cell sizes, extra logic or another corner.
- Grading reads the timing results: slack values, the endpoints and check types of the paths, and check_setup findings. The text of your SDC is never compared.
Hints
Hint 1
Each domain needs its own clock. Then ask which reset paths start at a clock edge and which start at the button.
Hint 2
A false path can name where a path starts, where it ends, or both. It can also be limited to setup or hold with -setup or -hold.
Hint 3
Run it and read the asynchronous path group. The releases into fa and fb must still be there, and nothing may start at rst_n.
What is checked: Live OpenSTA at tt_025C_1v80 with ideal clocks and no wire parasitics. Graded: the clock count and check_setup coverage. Also graded: the recovery and removal checks of both releases, and every such check from the raw reset. The release slacks must equal the reference, so any extra constraint that moves them fails.
Run times your SDC on the public netlist. Submit times it on the public netlist and every hidden variant, beside the reference constraints. The fixed circuit and allowed settings use the pinned OpenROAD/OpenSTA engine.
More problems on this topic