From Constraints to Confidence: Executing Timing Checks
You've defined your clocks, specified I/O delays, and declared timing exceptions. Your constraints are complete, your design is loaded, and OpenSTA is ready. Now comes the moment of truth: running timing analysis to discover if your design meets its timing requirements. This is where theory meets reality—where you learn if your chip will run at the target frequency or if optimization work lies ahead.
Running timing analysis isn't just executing a command and hoping for green checks. It's about understanding what the tool is reporting, interpreting slack values, tracing critical paths, and making informed decisions about optimization priorities. This module teaches you to run comprehensive timing analysis, read timing reports like a professional, and extract the insights needed for timing closure.
Setup Analysis: Checking Maximum Delays
What Is Setup Analysis?
Setup analysis verifies that data arrives at capture flip-flops before the required time. It checks the maximum delay paths to ensure signals propagate fast enough to be captured reliably on the next clock edge.
Setup Question: "Does data arrive early enough?"
Analysis Corner: Slow corner (SS) where delays are longest and meeting timing is hardest.
Running Setup Analysis in OpenSTA
Basic Setup Check
report_checks -path_delay max
This reports the worst (most critical) setup path in the design.
Multiple Paths
# Report 10 worst setup paths
report_checks -path_delay max -group_path_count 10
Specific Endpoint
# Check paths to specific flip-flop
report_checks -path_delay max -to [get_pins data_reg[7]/D]
Paths from Specific Source
# Check paths from particular register
report_checks -path_delay max -from [get_pins state_reg[*]/Q]
Unconstrained Paths
# Find paths without timing constraints
report_checks -path_delay max -unconstrained
This should ideally return nothing—all paths should be constrained.
Example Setup Analysis Command Sequence
# Complete setup analysis workflow
puts "=========================================="
puts "Setup Timing Analysis"
puts "=========================================="
# Check worst path
puts "\nWorst Setup Path:"
report_checks -path_delay max -group_path_count 1
# Check top 10 failing paths
puts "\nTop 10 Critical Setup Paths:"
report_checks -path_delay max -group_path_count 10 -slack_max 0
# Check all negative slack paths
puts "\nAll Setup Violations:"
report_checks -path_delay max -slack_max -0.001
# Verify all paths constrained
puts "\nUnconstrained Setup Paths:"
report_checks -path_delay max -unconstrained
Key Options:
-group_path_count N: Report up to N paths per path group-slack_max X: Only paths with slack ≤ X-slack_min X: Only paths with slack ≥ X-endpoint_path_count N: Report up to N paths per endpoint
Hold Analysis: Checking Minimum Delays
What Is Hold Analysis?
Hold analysis verifies that data remains stable after the clock edge long enough for the capture flip-flop to latch it. It checks minimum delay paths to ensure signals don't arrive too quickly.
Hold Question: "Does data stay stable long enough?"
Analysis Corner: Fast corner (FF) where delays are shortest and hold violations are most likely.
Running Hold Analysis in OpenSTA
Basic Hold Check
report_checks -path_delay min
Reports the worst (most critical) hold path.
Multiple Hold Paths
# Report 10 worst hold paths
report_checks -path_delay min -group_path_count 10
Hold Violations Only
# Paths with negative slack
report_checks -path_delay min -slack_max -0.001
Complete Hold Analysis
puts "=========================================="
puts "Hold Timing Analysis"
puts "=========================================="
# Worst hold path
puts "\nWorst Hold Path:"
report_checks -path_delay min -group_path_count 1
# All hold violations
puts "\nAll Hold Violations:"
report_checks -path_delay min -slack_max -0.001
# Critical hold paths (< 100ps margin)
puts "\nCritical Hold Paths (slack < 100ps):"
report_checks -path_delay min -slack_max 0.1
Setup vs. Hold: Running Both
Complete Analysis Script
# Read design and constraints
read_liberty lib/cells_tt.lib
read_verilog netlist/design.v
link_design top_module
read_sdc constraints/design.sdc
# ============================================
# Setup Analysis
# ============================================
puts "\n================================================"
puts "SETUP TIMING ANALYSIS"
puts "================================================"
report_checks -path_delay max -group_path_count 5
report_worst_slack -max
# ============================================
# Hold Analysis
# ============================================
puts "\n================================================"
puts "HOLD TIMING ANALYSIS"
puts "================================================"
report_checks -path_delay min -group_path_count 5
report_worst_slack -min
# ============================================
# Summary
# ============================================
puts "\n================================================"
puts "TIMING SUMMARY"
puts "================================================"
report_tns
report_wns
Understanding Timing Reports
Anatomy of a Timing Report
This is a real report_checks -path_delay max report from OpenSTA. The design is an 8-by-8 multiplier with registered inputs and a registered product, mapped to SkyWater 130 cells (sky130_fd_sc_hd, typical corner). The constraints set a 10 ns clock named sys_clk, 1.0 ns of ideal clock network latency and 0.2 ns of clock uncertainty. The flip-flop instances were renamed after the signals they drive, so a_r_1_reg holds bit 1 of input a and p_14_reg bit 14 of the product; the other cells keep the names synthesis gave them.
Startpoint: a_r_1_reg (rising edge-triggered flip-flop clocked by sys_clk)
Endpoint: p_14_reg (rising edge-triggered flip-flop clocked by sys_clk)
Path Group: sys_clk
Path Type: max
Delay Time Description
---------------------------------------------------------
0.00 0.00 clock sys_clk (rise edge)
1.00 1.00 clock network delay (ideal)
0.00 1.00 ^ a_r_1_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
0.46 1.46 ^ a_r_1_reg/Q (sky130_fd_sc_hd__dfxtp_1)
0.21 1.67 v _248_/Y (sky130_fd_sc_hd__nand4_1)
0.28 1.95 v _283_/X (sky130_fd_sc_hd__a22o_1)
0.18 2.14 v _288_/X (sky130_fd_sc_hd__and3_1)
0.41 2.55 ^ _294_/Y (sky130_fd_sc_hd__nor4b_1)
0.22 2.77 v _295_/Y (sky130_fd_sc_hd__o21ai_0)
0.41 3.17 v _322_/X (sky130_fd_sc_hd__maj3_1)
0.25 3.43 ^ _324_/Y (sky130_fd_sc_hd__o21ai_0)
0.13 3.55 v _470_/Y (sky130_fd_sc_hd__a21oi_1)
0.24 3.80 ^ _471_/Y (sky130_fd_sc_hd__o21ai_0)
0.14 3.94 v _473_/Y (sky130_fd_sc_hd__a21oi_1)
0.20 4.14 ^ _475_/Y (sky130_fd_sc_hd__o211ai_1)
0.14 4.28 v _477_/Y (sky130_fd_sc_hd__a31oi_1)
0.39 4.67 v _479_/X (sky130_fd_sc_hd__maj3_1)
0.16 4.83 v _488_/X (sky130_fd_sc_hd__xor2_1)
0.00 4.83 v p_14_reg/D (sky130_fd_sc_hd__dfxtp_1)
4.83 data arrival time
10.00 10.00 clock sys_clk (rise edge)
1.00 11.00 clock network delay (ideal)
-0.20 10.80 clock uncertainty
0.00 10.80 clock reconvergence pessimism
10.80 ^ p_14_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
-0.13 10.67 library setup time
10.67 data required time
---------------------------------------------------------
10.67 data required time
-4.83 data arrival time
---------------------------------------------------------
5.84 slack (MET)
Breaking Down the Report
Header Section
- Startpoint: where the data is launched, here the flip-flop
a_r_1_reg, with its type and clock - Endpoint: where the data is captured, here
p_14_reg - Path Group: the group the path is reported in, by default one per clock
- Path Type:
maxfor a setup check,minfor a hold check
Data Path Section
Delay Time Description
---------------------------------------------------------
0.00 0.00 clock sys_clk (rise edge)
1.00 1.00 clock network delay (ideal)
0.00 1.00 ^ a_r_1_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
0.46 1.46 ^ a_r_1_reg/Q (sky130_fd_sc_hd__dfxtp_1)
0.21 1.67 v _248_/Y (sky130_fd_sc_hd__nand4_1)
0.28 1.95 v _283_/X (sky130_fd_sc_hd__a22o_1)
0.18 2.14 v _288_/X (sky130_fd_sc_hd__and3_1)
0.41 2.55 ^ _294_/Y (sky130_fd_sc_hd__nor4b_1)
0.22 2.77 v _295_/Y (sky130_fd_sc_hd__o21ai_0)
0.41 3.17 v _322_/X (sky130_fd_sc_hd__maj3_1)
0.25 3.43 ^ _324_/Y (sky130_fd_sc_hd__o21ai_0)
0.13 3.55 v _470_/Y (sky130_fd_sc_hd__a21oi_1)
0.24 3.80 ^ _471_/Y (sky130_fd_sc_hd__o21ai_0)
0.14 3.94 v _473_/Y (sky130_fd_sc_hd__a21oi_1)
0.20 4.14 ^ _475_/Y (sky130_fd_sc_hd__o211ai_1)
0.14 4.28 v _477_/Y (sky130_fd_sc_hd__a31oi_1)
0.39 4.67 v _479_/X (sky130_fd_sc_hd__maj3_1)
0.16 4.83 v _488_/X (sky130_fd_sc_hd__xor2_1)
0.00 4.83 v p_14_reg/D (sky130_fd_sc_hd__dfxtp_1)
4.83 data arrival time
Column meanings: Delay is the delay added by this line, Time is the running total from the launching clock edge, and the description names the pin and its cell. ^ marks a rising signal and v a falling one.
- Clock launch edge (0.00 ns): the reference point
- Clock network delay (+1.00 ns): the clock latency set in the constraints; "ideal" means no clock tree has been built yet, so the value comes from the constraints
- Clock-to-Q (+0.46 ns): the time the launching flip-flop takes to update its output, from its Liberty data
- Logic: one line per cell output. Each delay comes from the cell's tables for its input slew and output load, so the same cell type can have different delays on different lines, such as the two
maj3_1cells (0.41 ns and 0.39 ns) - Data arrival time (4.83 ns): when the data reaches
p_14_reg/D
Required Time Section
10.00 10.00 clock sys_clk (rise edge)
1.00 11.00 clock network delay (ideal)
-0.20 10.80 clock uncertainty
0.00 10.80 clock reconvergence pessimism
10.80 ^ p_14_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
-0.13 10.67 library setup time
10.67 data required time
- Capture edge (10.00 ns): the next rising edge, one period after launch
- Clock network delay (+1.00 ns): the same latency on the capture side
- Clock uncertainty (-0.20 ns): margin for jitter and skew, which makes the check stricter
- Clock reconvergence pessimism (0.00): a correction for clock path shared by launch and capture; zero with an ideal clock
- Library setup time (-0.13 ns): how long before the edge the capturing flip-flop needs its data
- Data required time (10.67 ns): the deadline for the data
Slack Calculation and Interpretation
The Slack Equation
Setup Slack
Slack = Required time - Arrival time
= 10.67ns - 4.83ns
= 5.84ns
Hold Slack
For a hold check the data must arrive after the required time, so the slack is arrival minus required. This is the worst register-to-register hold path in the same design:
Startpoint: a_r_1_reg (rising edge-triggered flip-flop clocked by sys_clk)
Endpoint: p_1_reg (rising edge-triggered flip-flop clocked by sys_clk)
Path Group: sys_clk
Path Type: min
Delay Time Description
---------------------------------------------------------
0.00 0.00 clock sys_clk (rise edge)
1.00 1.00 clock network delay (ideal)
0.00 1.00 ^ a_r_1_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
0.37 1.37 v a_r_1_reg/Q (sky130_fd_sc_hd__dfxtp_1)
0.14 1.51 ^ _487_/Y (sky130_fd_sc_hd__a31oi_1)
0.00 1.51 ^ p_1_reg/D (sky130_fd_sc_hd__dfxtp_1)
1.51 data arrival time
0.00 0.00 clock sys_clk (rise edge)
1.00 1.00 clock network delay (ideal)
0.20 1.20 clock uncertainty
0.00 1.20 clock reconvergence pessimism
1.20 ^ p_1_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
-0.04 1.16 library hold time
1.16 data required time
---------------------------------------------------------
1.16 data required time
-1.51 data arrival time
---------------------------------------------------------
0.35 slack (MET)
The required time is built from the same clock edge as the launch, plus the uncertainty and minus the flip-flop's hold time: the new data, launched at 1.37 ns, must not reach the capturing flip-flop before 1.16 ns. It arrives at 1.51 ns, so the hold slack is 0.35 ns.
Interpreting Slack Values
Positive Slack: the check is met. The 5.84 ns above means the data arrives 5.84 ns before its deadline, so the clock could be faster or the path slower by that much.
Zero Slack: met with no margin, which leaves nothing for variation not covered by the analysis.
Negative Slack: the check fails and the design will not work at that clock. With the clock tightened to 4 ns, the same design reports:
max_delay/setup group sys_clk
Required Actual
Endpoint Delay Delay Slack
------------------------------------------------------------
p_14_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.67 4.83 -0.16 (VIOLATED)
p_15_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.73 4.81 -0.09 (VIOLATED)
p_13_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.67 4.45 0.22 (MET)
p_12_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.68 4.35 0.33 (MET)
p_11_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.68 4.23 0.45 (MET)
wns max -0.16
tns max -0.25
Slack and Frequency Relationship
The worst path misses a 4 ns clock by 0.16 ns, so this netlist needs a period of at least 4.16 ns: about 240 MHz instead of 250 MHz. That holds only if nothing else changes: a faster mapping, a pipeline stage or a different netlist moves the number.
Worst Negative Slack (WNS) and Total Negative Slack (TNS)
report_wns
report_tns
wns max -0.16
tns max -0.25
WNS is the most negative slack in the design, -0.16 ns here. TNS adds up the negative slack of every failing endpoint: -0.25 ns from the two failing endpoints in the table above. A small WNS with a large TNS means many paths fail by a little; a large WNS with a TNS close to it means a few paths fail badly. Both are reported for setup by default; report_wns -min and report_tns -min give the hold values.
Path Details: Understanding the Journey
Startpoint Analysis
A startpoint is a flip-flop clock pin (data launched by a register), an input port (data from outside the block, timed with set_input_delay), or the output of a latch. The text in parentheses after the name says which, and which clock launches it.
Endpoint Analysis
An endpoint is a flip-flop data pin (setup and hold checks), an output port (timed with set_output_delay), or a check such as a flip-flop's asynchronous reset pin (recovery and removal).
Data Path Tracing
Reading the data path line by line gives the route the signal takes: from a_r_1_reg/Q through a NAND4, an AND-OR, an AND3, a NOR4 with an inverted input and a chain of OR-AND-INVERT, AND-OR-INVERT and majority cells to an XOR that produces product bit 14. The edge marks alternate where cells invert: _294_ is a NOR, so its output rises (^) while its inputs fall. A chain of carry and majority cells like this is typical of the critical path of a multiplier or an adder.
Clock Path Analysis
With an ideal clock, the clock path is just the latency from the constraints, the same on both sides. Once a clock tree is built, each flip-flop gets its own clock arrival time, and the difference between launch and capture is the clock skew. To show its effect, this run gives only the capturing flip-flop 0.2 ns more latency:
10.00 10.00 clock sys_clk (rise edge)
1.20 11.20 clock network delay (ideal)
-0.20 11.00 clock uncertainty
0.00 11.00 clock reconvergence pessimism
11.00 ^ p_14_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
-0.13 10.87 library setup time
10.87 data required time
---------------------------------------------------------
10.87 data required time
-4.83 data arrival time
---------------------------------------------------------
6.04 slack (MET)
The capture edge arrives 0.2 ns later, so the required time and the setup slack grow by exactly 0.2 ns, from 5.84 to 6.04 ns. Positive skew (capture later than launch) helps setup and hurts hold; negative skew does the opposite.
Practical Example: Complete Analysis Session
Example Design: 8-bit Counter
Setup
# Load design
read_liberty lib/stdcells_tt.lib
read_verilog netlist/counter_mapped.v
link_design counter
# Load constraints
read_sdc constraints/counter.sdc
# Verify setup
report_clocks
check_timing
Run Analysis
# ============================================
# Setup Check
# ============================================
puts "\n=== Setup Timing ==="
report_checks -path_delay max -group_path_count 3 -format end
# ============================================
# Hold Check
# ============================================
puts "\n=== Hold Timing ==="
report_checks -path_delay min -group_path_count 3 -format end
# ============================================
# Summary
# ============================================
puts "\n=== Timing Summary ==="
report_worst_slack -max
report_worst_slack -min
Output (8-bit counter with enable, mapped to SkyWater 130 sky130_fd_sc_hd cells at the typical corner; counter.sdc sets a 10 ns clock, 0.2 ns uncertainty, and 0.5 ns input and output delays):
=== Setup Timing ===
max_delay/setup group sys_clk
Required Actual
Endpoint Delay Delay Slack
------------------------------------------------------------
count_reg_7/D (sky130_fd_sc_hd__dfrtp_1) 9.72 1.19 8.53 (MET)
count_reg_6/D (sky130_fd_sc_hd__dfrtp_1) 9.68 1.04 8.64 (MET)
count_reg_4/D (sky130_fd_sc_hd__dfrtp_1) 9.68 1.01 8.67 (MET)
=== Hold Timing ===
min_delay/hold group sys_clk
Required Actual
Endpoint Delay Delay Slack
------------------------------------------------------------
count_reg_7/D (sky130_fd_sc_hd__dfrtp_1) 0.15 0.37 0.22 (MET)
count_reg_6/D (sky130_fd_sc_hd__dfrtp_1) 0.15 0.39 0.24 (MET)
count_reg_5/D (sky130_fd_sc_hd__dfrtp_1) 0.15 0.40 0.25 (MET)
=== Timing Summary ===
worst slack max 8.53
worst slack min 0.22
-format end prints one line per endpoint; drop it to see each path in full, as in the report at the top of this article. -group_path_count sets how many paths are reported per path group.
Interpretation:
- Setup: 8.53ns slack, a very comfortable margin at 10 ns
- Hold: 0.22ns slack, met but much closer (typical)
- Critical path: the
eninput through the increment logic (and4, nand4, xor2) intocount_reg_7, the most significant bit - Hold critical:
count_reg_7again, through its shortest path
Analyzing a Violation
Setup Violation Example
The multiplier from the reports above, run with a 4 ns clock, fails setup:
max_delay/setup group sys_clk
Required Actual
Endpoint Delay Delay Slack
------------------------------------------------------------
p_14_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.67 4.83 -0.16 (VIOLATED)
p_15_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.73 4.81 -0.09 (VIOLATED)
p_13_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.67 4.45 0.22 (MET)
p_12_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.68 4.35 0.33 (MET)
p_11_reg/D (sky130_fd_sc_hd__dfxtp_1) 4.68 4.23 0.45 (MET)
wns max -0.16
tns max -0.25
Analysis Questions:
- What is the path? The full report is the one at the top of this article:
a_r_1_regthrough the partial-product and carry logic of the multiplier top_14_reg, 3.83 ns of clock-to-Q and logic after the 1 ns clock latency. - Why is it failing? The data arrives at 4.83 ns, and with a 4 ns clock it is required by 4.67 ns (4 ns plus 1 ns of latency, minus 0.2 ns of uncertainty and 0.13 ns of setup time).
- How to fix it? Pipeline the multiplier into two stages; try a faster mapping or let place and route upsize the cells on this path; or run the clock at 4.16 ns, about 240 MHz, if the specification allows.
Hold Violation Example
The same netlist at the fast corner (ff_n40C_1v95), with the hold uncertainty raised to 0.5 ns:
Startpoint: a_r_1_reg (rising edge-triggered flip-flop clocked by sys_clk)
Endpoint: p_1_reg (rising edge-triggered flip-flop clocked by sys_clk)
Path Group: sys_clk
Path Type: min
Delay Time Description
---------------------------------------------------------
0.00 0.00 clock sys_clk (rise edge)
1.00 1.00 clock network delay (ideal)
0.00 1.00 ^ a_r_1_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
0.24 1.24 v a_r_1_reg/Q (sky130_fd_sc_hd__dfxtp_1)
0.10 1.33 ^ _487_/Y (sky130_fd_sc_hd__a31oi_1)
0.00 1.33 ^ p_1_reg/D (sky130_fd_sc_hd__dfxtp_1)
1.33 data arrival time
0.00 0.00 clock sys_clk (rise edge)
1.00 1.00 clock network delay (ideal)
0.50 1.50 clock uncertainty
0.00 1.50 clock reconvergence pessimism
1.50 ^ p_1_reg/CLK (sky130_fd_sc_hd__dfxtp_1)
-0.02 1.48 library hold time
1.48 data required time
---------------------------------------------------------
1.48 data required time
-1.33 data arrival time
---------------------------------------------------------
-0.15 slack (VIOLATED
Analysis: the path is a single gate, so the new data reaches p_1_reg 0.33 ns after the clock latency, before the 1.48 ns at which hold is satisfied. Slowing the clock does not help, because launch and capture use the same edge. The fix is to add delay on this data path, usually with delay or buffer cells inserted by the place and route tool, and to recheck setup on the same path afterwards.
Advanced Reporting Options
Filtering and Grouping
By Clock Domain
# Only paths in specific clock domain
report_checks -path_delay max -path_group sys_clk
Several Paths per Group
# Up to 5 paths in each path group (one group per clock)
report_checks -path_delay max -group_path_count 5
The older -group_count still works but is deprecated. Write options out in full: OpenSTA accepts abbreviations, so -group sys_clk is read as -group_count sys_clk and rejected because sys_clk is not a number.
Through Specific Logic
# Paths through critical module
report_checks -path_delay max -through [get_pins alu/*]
Format Options
Brief Format (summary only)
report_checks -path_delay max -format summary
Full Detail with Clock Paths
report_checks -path_delay max -format full_clock
The default format, full, shows the data path in detail and the clock as one latency line; full_clock adds the clock path lines, and full_clock_expanded expands them through the clock network.
Fields to Display
report_checks -path_delay max \
-fields {capacitance slew fanout}
Shows additional timing details like transition times and loads.
Output Redirection
Save to File
# Redirect output to file
report_checks -path_delay max > setup_timing.rpt
report_checks -path_delay min > hold_timing.rpt
Append to File
report_checks -path_delay max >> timing_all.rpt
report_checks -path_delay min >> timing_all.rpt
Summary
Running timing analysis reveals whether your design meets timing requirements:
- Setup analysis (
-path_delay max) checks maximum delays at slow corner - Hold analysis (
-path_delay min) checks minimum delays at fast corner - Timing reports show complete path details with incremental and cumulative delays
- Slack = Required_time - Arrival_time (positive = pass, negative = fail)
- Path components: Startpoint → Data path → Endpoint, with clock paths defining required times
Key skills you've developed:
- Running setup and hold checks systematically
- Reading and interpreting timing reports
- Understanding slack calculations
- Tracing signal paths through the design
- Identifying critical paths and violations
- Using report options for targeted analysis
With timing analysis mastered, you can now verify designs, identify bottlenecks, and guide optimization efforts. Every timing report tells a story about your design's performance—learning to read that story is essential for successful digital design. In the next module, you'll learn to fix timing violations and optimize for timing closure.