Compute a driver's load and two met2 wire delays by hand, then let OpenSTA check them.
Hardware context: Buffer driving a long met1 wire to a second buffer. Estimate a driver load and two wire delays by hand and check them against OpenSTA.
The block lwire has a buf_2 named drv that drives a long net w to a buf_1 named rcv. Two hand-made SPEF files route that net on met2, as 700 µm or 1 mm of wire. Each is a 10-segment RC ladder, with half a segment's capacitance at each end.
Work by hand from the data in the full specification. Give the total capacitance drv drives with the 1 mm wire. Then give the Elmore delay from drv/X to rcv/A for both lengths. The server runs OpenSTA on both SPEFs and compares your numbers with its report.
Requirements
| Requirement | Limit | Unit |
|---|---|---|
| Load error on drv with the 1 mm wire | ≤ 0.5 | fF |
| Wire delay error, 700 µm | ≤ 0.005 | ns |
| Wire delay error, 1 mm | ≤ 0.005 | ns |
Live analysis: Live OpenSTA at tt_025C_1v80 on lwire with the authored met2 SPEFs lwire-m2-700 and lwire-m2-1000, ideal 2 ns clock. Graded: the load of drv/X with the 1 mm SPEF, within 0.5 fF. Also graded: the net line from drv/X to rcv/A on the worst setup path, within 5 ps. The tool's calculator is not plain Elmore, and the 5 ps tolerance covers the difference at these lengths.
Full specification
| Quantity | Value |
|---|---|
| met2 resistance (platform setRC.tcl) | 1.22132 Ω per µm |
| met2 capacitance (platform setRC.tcl) | 0.136233 fF per µm |
| Ladder | 10 equal segments; each inner node holds one segment's capacitance, each end half of one |
| rcv/A (buf_1) rise_capacitance | 2.191 fF (the worst path rises at rcv/A) |
| Clock | create_clock -name clk -period 2.0 [get_ports clk]; set_clock_transition 0.1 [all_clocks] |
| JSON field | Allowed values | Meaning |
|---|---|---|
| load_ff | 0 to 1000 | Total capacitance on net w that drv drives, in fF. |
| delay_700 | 0 to 1 | Elmore delay from drv/X to rcv/A with the 700 µm wire, in ns. |
| delay_1000 | 0 to 1 | Elmore delay from drv/X to rcv/A with the 1 mm wire, in ns. |
Live OpenSTA at tt_025C_1v80 on lwire with the authored met2 SPEFs lwire-m2-700 and lwire-m2-1000, ideal 2 ns clock. Graded: the load of drv/X with the 1 mm SPEF, within 0.5 fF. Also graded: the net line from drv/X to rcv/A on the worst setup path, within 5 ps. The tool's calculator is not plain Elmore, and the 5 ps tolerance covers the difference at these lengths.
Numeric checks use inclusive bounds. A missing measurement always fails. Only the documented fields are accepted.
Examples
Example 1
Total R Times Total C- Input
Multiply the whole wire resistance by the whole capacitance
- Output
Load error on drv with the 1 mm wire: PASS Wire delay error, 700 µm: FAIL Wire delay error, 1 mm: FAIL
The first resistor charges almost everything, but the last one charges only the far end. On average each resistor sees about half the wire.
Example 2
The Pin Left Out- Input
Use only the wire capacitance
- Output
Load error on drv with the 1 mm wire: FAIL Wire delay error, 700 µm: PASS Wire delay error, 1 mm: PASS
The delays land close, but the driver also charges the receiver pin. The report's load includes it.
Example 3
One Lumped RC- Input
Use 0.69 times the wire's R times C
- Output
Load error on drv with the 1 mm wire: PASS Wire delay error, 700 µm: FAIL Wire delay error, 1 mm: FAIL
0.69·RC is the 50 % point of a single resistor and capacitor under a step. A distributed wire is not one RC.
Submission Rules
- Submit one JSON object with every documented configuration field. The supplied circuit and external environment stay fixed.
- Run measures your configuration. Submit applies the same requirements and records the result.
- The stated stage and cell corner define this exercise; passing does not establish full chip sign-off.
Hints
Hint 1
Write the Elmore sum: each resistor times all the capacitance beyond it, including the receiver pin.
Hint 2
For a uniform ladder with half a segment at each end, the sum closes to R·(C/2 + Cpin). R and C are for the whole wire.
Hint 3
Watch the units: kΩ times pF gives ns.
What is checked: Live OpenSTA at tt_025C_1v80 on lwire with the authored met2 SPEFs lwire-m2-700 and lwire-m2-1000, ideal 2 ns clock. Graded: the load of drv/X with the 1 mm SPEF, within 0.5 fF. Also graded: the net line from drv/X to rcv/A on the worst setup path, within 5 ps. The tool's calculator is not plain Elmore, and the 5 ps tolerance covers the difference at these lengths.
Run and Submit use the same measured requirements. Run is a preview; Submit records an attempt. The fixed circuit and allowed settings use the pinned OpenROAD/OpenSTA engine.
More problems on this topic