Advertisement
STA Sign-off & Constraints

SDC Constraints &
Timing Exceptions Studio

Static Timing Analysis is only as accurate as your design constraints. Synopsys Design Constraints (SDC) is the industry standard for specifying clock trees, I/O delay budgets, clock domain crossings, and timing exceptions across synthesis (Design Compiler, Yosys), STA (PrimeTime, OpenSTA), and place-and-route (Innovus, OpenROAD).

40 min read
Tcl / SDC 2.1 Standard
Interactive SDC Generator & Waveforms
Launch Interactive SDC Studio Run in Online OpenSTA Setup/Hold Calculator

What is SDC and Why Sign-off Depends On It

SDC (Synopsys Design Constraints) is a declarative, Tcl-based language that defines the operating intentions and external environment of an integrated circuit. Modern EDA engines — including logic synthesis (Synopsys Design Compiler, Cadence Genus, Yosys), static timing signoff (PrimeTime, Tempus, OpenSTA), and physical place-and-route (Innovus, Synopsys ICC2, OpenROAD) — cannot make a single optimization decision without SDC.

Pillar 1: Clock Definitions

Defines master oscillators, clock frequencies, duty cycle waveforms, source jitter, tree insertion delay, and generated derived clocks (dividers, PLL outputs). Clocks establish the fundamental reference grid for all setup and hold checks.

Pillar 2: I/O Boundary Budgets

Constrains primary chip boundary ports. Specifies external driver delay and PCB trace delay (set_input_delay) and downstream receiving device setup requirements (set_output_delay). Closes the timing loop with the external board.

Pillar 3: Timing Exceptions

Informs the STA engine which physical topological paths should not be timed to single-cycle requirements: structurally impossible false paths (set_false_path), multi-cycle paths (set_multicycle_path), and CDC relationships (set_clock_groups).

The Danger of Garbage Constraints: If your SDC is overly tight (e.g. unrealistic zero-delay I/O or excessive jitter), synthesis tools waste die area inserting massive buffers and high-drive cells, skyrocketing dynamic power. Conversely, if constraints are loose or omit false paths/clock groups, real silicon timing violations escape to fabrication, causing catastrophic silicon respins.

Clock Definition — create_clock & Virtual Clocks

Every synchronous digital timing check compares the launch flip-flop clock edge against the capture flip-flop clock edge. The create_clock command defines an ideal periodic timing source at a primary input port or internal pin.

# 1. Standard 500 MHz Primary Clock (2.0 ns period, 50% duty cycle)
create_clock -name SYS_CLK -period 2.0 -waveform {0.0 1.0} [get_ports sys_clk]

# 2. Asymmetric Clock (40% duty cycle, high from 0.0 ns to 1.6 ns on 4.0 ns period)
create_clock -name CLK_ASYM -period 4.0 -waveform {0.0 1.6} [get_ports clk_slow]

# 3. Virtual Clock (No source pin/port!)
# Used as an ideal external reference clock for board-level I/O delay constraints
create_clock -name VCLK_EXT -period 2.0

Understanding Virtual Clocks

A virtual clock is a clock with a defined period and waveform that is not bound to any pin or port in the netlist. It models the external transmitting or receiving device's clock. When your chip receives data from an off-chip FPGA or ADC, referencing set_input_delay to a virtual clock enables the STA tool to account for board clock skew independently.

The -waveform Parameter

The -waveform {rise_time fall_time} switch specifies when the clock edge rises and falls within the period. If omitted, tools default to {0 [expr period/2]} (50% duty cycle). Both edges are tracked for half-cycle paths (e.g., launch on positive edge, capture on negative edge).

Generated Clocks — create_generated_clock

When a clock is derived inside the chip via a frequency divider, phase-locked loop (PLL), or clock gating cell (ICG), you must define it using create_generated_clock. This informs the STA tool of the exact phase, frequency, and latency relationship back to the master clock.

# 1. Frequency Divider by 2 (e.g. from flip-flop Q output)
create_generated_clock -name CLK_DIV2 \
  -source [get_ports sys_clk] \
  -divide_by 2 \
  [get_pins u_clkdiv/q_reg/Q]

# 2. PLL Frequency Multiplier (50 MHz input -> 400 MHz output)
create_generated_clock -name PLL_CLK400 \
  -source [get_ports ref_clk] \
  -multiply_by 8 \
  [get_pins u_pll/clkout]

# 3. Inverted Clock (180 degree phase shift)
create_generated_clock -name CLK_INV \
  -source [get_pins u_clkgen/clk_in] \
  -invert \
  [get_pins u_inv/Y]

# 4. Explicit Edge Specification (Divide by 1.5 or custom waveform)
# Maps master clock edges 1, 2, 4 to generated clock edges 1 (rise), 2 (fall), 3 (rise)
create_generated_clock -name CLK_CUSTOM \
  -source [get_ports sys_clk] \
  -edges {1 2 4} \
  [get_pins u_custom_div/out]
Why NOT create a primary clock on a divider output? If you use create_clock on an internal register pin instead of create_generated_clock, the STA tool treats it as an independent primary source. It completely severs the phase relationship with the master clock, breaks latency calculation through the divider logic, and prevents accurate synchronous timing verification across the divider boundary.

Clock Quality — Uncertainty, Latency & Slew

Ideal clock trees do not exist in silicon. Clock signals suffer from phase jitter, clock-tree skew, dynamic voltage drop variations, and finite rise/fall slew times. SDC provides specialized commands to model these physical effects.

# Clock Uncertainty: Setup jitter deducted from available budget; hold margin added
set_clock_uncertainty -setup 0.08 [get_clocks SYS_CLK]   ;# 80 ps setup jitter
set_clock_uncertainty -hold  0.03 [get_clocks SYS_CLK]   ;# 30 ps hold uncertainty

# Inter-Clock Uncertainty: Between distinct clock domains
set_clock_uncertainty -from [get_clocks CLK_A] -to [get_clocks CLK_B] 0.12

# Source Latency (Off-chip delay: crystal oscillator -> chip pin)
set_clock_latency -source 0.35 [get_clocks SYS_CLK]

# Network Latency (Estimated on-chip clock tree insertion delay before CTS)
set_clock_latency 0.75 [get_clocks SYS_CLK]

# Post-CTS Sign-off: Tell the tool to propagate real extracted clock tree RC delays
set_propagated_clocks [get_clocks SYS_CLK]

# Clock Transition: Set maximum slew on clock arrival
set_clock_transition 0.05 [get_clocks SYS_CLK]            ;# 50 ps clock slew
SDC Command Physical Meaning Pre-CTS vs Post-CTS Impact on Slack
set_clock_uncertainty -setup PLL phase jitter + dynamic voltage drop noise + skew guardband Higher margin pre-CTS (100–150 ps); reduced post-CTS (jitter only, 40–80 ps) Reduces Setup Slack
set_clock_uncertainty -hold Clock skew margin + fast-path dynamic uncertainty Essential pre-CTS (50–100 ps); post-CTS accounts for OCV/AOCV hold margin Reduces Hold Slack
set_clock_latency Clock network insertion delay from clock port to sink register CK pin Used pre-CTS. Must be removed post-CTS with set_propagated_clock Shifts Arrival
set_clock_transition Rise and fall slew time at the clock root pin Affects standard cell switching delays in the initial clock buffer chain Cell Delays

I/O Timing — Input and Output Delay Derivation

A common beginner mistake is thinking set_input_delay specifies the arrival time at the chip's internal flip-flop. It does not. set_input_delay specifies how much of the clock period is already consumed outside the chip before the signal reaches the primary input pad.

Input Delay (Max - Setup) = T_cq_ext(max) + T_pcb_data(max) − T_pcb_clk(min)
Input Delay (Min - Hold) = T_cq_ext(min) + T_pcb_data(min) − T_pcb_clk(max)
Output Delay (Max - Setup) = T_pcb_data(max) + T_setup_ext − T_pcb_clk(min)
Output Delay (Min - Hold) = T_pcb_data(min) − T_hold_ext − T_pcb_clk(max)
# Input delay: External sensor drives chip inputs.
# External device takes up to 0.7 ns (Tcq + PCB). Internal logic has: Period - 0.7 ns
set_input_delay -max 0.7 -clock SYS_CLK [get_ports {data_in[*] valid_in}]
set_input_delay -min 0.15 -clock SYS_CLK [get_ports {data_in[*] valid_in}]

# Output delay: Downstream SRAM requires 0.5 ns setup time + trace delay
set_output_delay -max 0.5 -clock SYS_CLK [get_ports {addr_out[*] data_out[*]}]
set_output_delay -min -0.05 -clock SYS_CLK [get_ports {addr_out[*] data_out[*]}]

# Referencing Virtual Clocks for Board Timing
set_input_delay -max 0.65 -clock VCLK_EXT [get_ports ext_bus_in]
set_output_delay -max 0.45 -clock VCLK_EXT [get_ports ext_bus_out]

Why Specify -min Delay?

If you only provide -max, the tool assumes the minimum delay equals the maximum delay. This falsely optimistic hold analysis ignores the case where an external transmitter sends data ultra-fast (low $T_{cq\_min}$), which causes hold violations at your input registers!

Negative Output Delay

A negative set_output_delay -min (e.g. -0.05 ns) is completely normal! It occurs when the downstream receiving device has a positive hold time requirement ($T_{hold}$), meaning data must remain stable for a short duration after the clock edge.

Advertisement

Clock Domain Crossing — set_clock_groups vs set_false_path

Modern SoCs contain dozens of independent clock domains (PCIe, DDR, USB, CPU core, UART). When data travels between two unrelated clock domains, Static Timing Analysis cannot calculate setup or hold slack because the phase relationship between the oscillators is indefinite. You must declare these domains asynchronous.

# RECOMMENDED: Define Asynchronous Clock Groups atomically
set_clock_groups -asynchronous \
  -group {SYS_CLK CLK_DIV2} \
  -group {UART_CLK} \
  -group {PCIE_REFCLK PCIE_TXCLK}

# Logically Exclusive: Clocks that pass through a multiplexer (only one runs at a time)
set_clock_groups -logically_exclusive \
  -group {TEST_CLK} \
  -group {FUNC_CLK}

# Physically Exclusive: Clocks that share the same physical pad/trace
set_clock_groups -physically_exclusive \
  -group {XTAL_CLK} \
  -group {EXT_OSC}
Feature set_clock_groups -asynchronous set_false_path -from [get_clocks A] -to [get_clocks B]
Directionality Bidirectional — covers A→B and B→A automatically Unidirectional — requires two separate commands (A→B and B→A)
Tool Performance Prunes timing graph early during synthesis & STA (up to 40% faster signoff) Path exceptions evaluated during detailed graph traversal
Priority & Overrides Highest precedence — clean and robust against netlist hierarchy changes Can mask intended point-to-point multicycle exceptions
Industry Recommendation Mandatory Standard in modern ASIC signoff Legacy approach — prone to omission of return path

Multicycle Paths — set_multicycle_path

By default, STA assumes data launched on edge $T=0$ must arrive and stabilize at the destination register before edge $T=1 \times T_{period}$. However, complex arithmetic blocks (multipliers, floating-point dividers, multi-stage pipelines with clock enable) intentionally take multiple clock cycles to complete.

Setup Check Window: Launch Edge (0) → Capture Edge (N × T_period)
Default Hold Check (WITHOUT -hold): Launch Edge (0) → Capture Edge ((N−1) × T_period) ← Extreme Hold Violation!
Corrected Hold Check (WITH -hold N-1): Launch Edge (0) → Capture Edge (0 ns) ← Physically Correct
# Example: 16-bit Hardware Multiplier taking 2 Clock Cycles
# STEP 1: Relax Setup to 2 clock cycles
set_multicycle_path -setup 2 \
  -from [get_cells u_core/mult_a_reg*] \
  -to   [get_cells u_core/mult_res_reg*]

# STEP 2: MANDATORY! Move hold check back by 1 cycle (N - 1 = 1)
set_multicycle_path -hold 1 \
  -from [get_cells u_core/mult_a_reg*] \
  -to   [get_cells u_core/mult_res_reg*]

# 3-Cycle Path (e.g. Memory Controller Read Data Access)
set_multicycle_path -setup 3 -from [get_cells u_sram/*] -to [get_cells u_dsp/*]
set_multicycle_path -hold  2 -from [get_cells u_sram/*] -to [get_cells u_dsp/*]
The Hold Rule of Thumb: Whenever you write set_multicycle_path -setup N between registers on the same clock frequency, you MUST immediately write set_multicycle_path -hold (N-1). If you forget the hold constraint, synthesis tools will attempt to insert hundreds of delay buffers to delay the data by $(N-1)$ clock periods, destroying area and power!

False Paths & Case Analysis — Pruning Non-Functional Logic

A false path is a timing path that physically exists in the gate netlist but cannot be sensitized or exercised during real operational modes. set_case_analysis forces static logic values (0 or 1) onto pins or ports to configure the circuit for specific operational scenarios.

# 1. Asynchronous Reset: Deasserts once during power-on reset sequence
# (Recovery/Removal checks are performed; setup/hold on reset pin is false)
set_false_path -from [get_ports rst_n]

# 2. Static Configuration Registers (written once at boot, never toggle during runtime)
set_false_path -from [get_cells u_config/device_id_reg*]

# 3. Mode Select Mux Pin (Exclude paths through scan test mux)
set_false_path -through [get_pins u_scan_mux/S]

# 4. Point-to-Point False Path
set_false_path -from [get_cells u_src_reg] -to [get_cells u_dst_reg]

# Case Analysis: Force Scan Enable to 0 for functional timing signoff
set_case_analysis 0 [get_ports scan_en]
set_case_analysis 0 [get_ports test_mode]

# Case Analysis: Force Clock Mux Select to analyze High-Speed PLL mode
set_case_analysis 1 [get_pins u_clkmux/sel]
Do NOT use false paths to silence real timing violations! If your ALU or state machine fails setup timing, using set_false_path to "make the report green" is an engineering offense. In silicon, that path will be exercised, cause metastability or corrupted data, and brick your chip. Only declare a path false when it is mathematically and protocol-wise impossible to sensitize.
Interactive SDC Studio & Timing Lab
Configure clock parameters, external I/O delays, CDC groups, and timing exceptions with live waveform preview.
Live Engine
Load Preset:


Dynamic Timing Diagram Setup: 1 Cycle
Period Budget: 2.00 ns Available Core: 0.84 ns (42%)
In: 30%
Core: 42%
Out: 25%
Unc
design.sdc (Tcl / SDC 2.1)
Generating SDC constraints...

SDC Sign-off Checklist: 10 Critical Audit Rules

Before tape-out or final CTS, physical design and STA teams perform strict constraint validation (e.g. using Synopsys check_timing or OpenSTA report_checks -unconstrained). Violating any of these 10 rules jeopardizes silicon success:

1. Zero Unconstrained Endpoints

Run check_timing -include {no_clock unconstrained_endpoints}. Every sequential register D pin and every chip output pad must have a valid clock reference and timing path.

2. Realistic Clock Uncertainty

Never sign off with zero clock uncertainty! Even post-CTS, you must maintain 30–60 ps of setup uncertainty for PLL phase noise, jitter, and on-chip voltage variations.

3. Both -max and -min I/O Delays

Always declare both set_input_delay -max AND -min, and set_output_delay -max AND -min. Omitting minimum delays hides board-level hold violations.

4. Paired MCP -hold Constraints

Every set_multicycle_path -setup N must have a corresponding set_multicycle_path -hold (N-1) constraint. Unpaired MCPs result in massive false hold buffer insertion.

5. Bidirectional CDC Groups

Use set_clock_groups -asynchronous for all cross-domain clocks. Do not rely on loose, one-way set_false_path rules that miss reverse return handshakes.

6. All Derived Clocks Constrained

Every frequency divider flip-flop, PLL output, and ripple counter output must be constrained via create_generated_clock referencing its true master source pin.

7. Transition & Slew Limits

Define set_max_transition according to standard-cell characterization tables (typically 100–200 ps). High slew slows cell switching and increases dynamic power.

8. Drive Strength & Pin Loads

Primary input ports must have an explicit set_driving_cell (or set_drive), and output ports must have an explicit set_load modeling the external capacitive load.

9. Remove Latency Post-CTS

Before CTS, set_clock_latency estimates clock tree insertion. After CTS, remove it and apply set_propagated_clocks to trace real extracted routing delays.

10. Document Every False Path

Every set_false_path must include an engineering comment explaining why it is topologically or functionally impossible. Never mask real paths to pass timing.

Production SDC Template — Multi-Clock SoC Block

Below is a complete, production-grade SDC template incorporating primary oscillators, generated clocks, asynchronous CDC groups, I/O delay budgets, and exceptions for a modern RISC-V SoC with UART and SPI peripherals:

################################################################################
# Design: soc_core_top | Technology: TSMC 28nm / SKY130 | SDC Version: 2.1
################################################################################

# ------------------------------------------------------------------------------
# 1. PRIMARY CLOCKS
# ------------------------------------------------------------------------------
# Main Core Clock: 500 MHz (2.0 ns period, 50% duty cycle)
create_clock -name SYS_CLK -period 2.000 -waveform {0.000 1.000} [get_ports sys_clk]

# Peripheral Bus Clock: 100 MHz (10.0 ns period)
create_clock -name APB_CLK -period 10.000 -waveform {0.000 5.000} [get_ports apb_clk]

# External Virtual Reference Clock for Board-Level Timing
create_clock -name VCLK_EXT -period 2.000

# ------------------------------------------------------------------------------
# 2. GENERATED CLOCKS
# ------------------------------------------------------------------------------
# Divided Clock (SYS_CLK / 4 = 125 MHz) for DSP Engine
create_generated_clock -name CLK_DIV4 \
  -source [get_ports sys_clk] \
  -divide_by 4 \
  [get_pins u_clkgen/u_div4/q_reg/Q]

# ------------------------------------------------------------------------------
# 3. CLOCK UNCERTAINTY, LATENCY & SLEW
# ------------------------------------------------------------------------------
set_clock_uncertainty -setup 0.060 [get_clocks SYS_CLK]
set_clock_uncertainty -hold  0.025 [get_clocks SYS_CLK]
set_clock_uncertainty -setup 0.100 [get_clocks APB_CLK]
set_clock_uncertainty -hold  0.040 [get_clocks APB_CLK]

set_clock_transition 0.050 [get_clocks {SYS_CLK APB_CLK}]
set_clock_latency -source 0.250 [get_clocks SYS_CLK]

# ------------------------------------------------------------------------------
# 4. CLOCK DOMAIN CROSSING (CDC) GROUPS
# ------------------------------------------------------------------------------
# SYS_CLK and APB_CLK are asynchronous; exclude cross-domain checks
set_clock_groups -asynchronous \
  -group {SYS_CLK CLK_DIV4} \
  -group {APB_CLK}

# ------------------------------------------------------------------------------
# 5. INPUT / OUTPUT DELAYS
# ------------------------------------------------------------------------------
# Core data inputs (referenced to SYS_CLK)
set_input_delay  -max 0.600 -clock SYS_CLK [get_ports {data_in[*] valid_in}]
set_input_delay  -min 0.150 -clock SYS_CLK [get_ports {data_in[*] valid_in}]

# Core data outputs
set_output_delay -max 0.500 -clock SYS_CLK [get_ports {data_out[*] ready_out}]
set_output_delay -min 0.000 -clock SYS_CLK [get_ports {data_out[*] ready_out}]

# APB peripheral inputs (referenced to APB_CLK)
set_input_delay  -max 3.000 -clock APB_CLK [get_ports {paddr[*] pwdata[*] pwrite}]
set_input_delay  -min 0.800 -clock APB_CLK [get_ports {paddr[*] pwdata[*] pwrite}]

# ------------------------------------------------------------------------------
# 6. TIMING EXCEPTIONS (False Paths & Multicycle Paths)
# ------------------------------------------------------------------------------
# Asynchronous Reset (asserted/deasserted externally)
set_false_path -from [get_ports rst_n]

# 2-Cycle Multicycle Path on 32-bit Integer Multiplier
set_multicycle_path -setup 2 -from [get_cells u_core/u_mul/a_reg*] -to [get_cells u_core/u_mul/res_reg*]
set_multicycle_path -hold  1 -from [get_cells u_core/u_mul/a_reg*] -to [get_cells u_core/u_mul/res_reg*]

# Case Analysis: Force scan test enable to 0 during mission-mode STA
set_case_analysis 0 [get_ports scan_en]
set_case_analysis 0 [get_ports bist_mode]

# ------------------------------------------------------------------------------
# 7. DESIGN RULE CONSTRAINTS (DRC) & OPERATING ENVIRONMENT
# ------------------------------------------------------------------------------
set_max_transition 0.150 [current_design]
set_max_fanout 20 [current_design]
set_load 0.040 [all_outputs]
set_driving_cell -cell BUF_X4 [all_inputs -no_clocks]
set_operating_conditions -max WORST -min BEST
################################################################################

Frequently Asked Questions

Without an SDC file, the STA tool has no clock reference, frequency target, or I/O arrival constraints. The tool will report all timing paths as unconstrained (giving a falsely clean green report). In synthesis (e.g., Yosys or Design Compiler), the engine optimizes strictly for minimum silicon area, producing slow combinational logic that fails in real silicon. Missing SDC is one of the leading causes of first-silicon respins.
A false path (set_false_path) is a path that is physically or logically impossible to sensitize during functional operation (e.g. static configuration bits, scan test pins). STA disables all setup and hold checks on it. A multicycle path (set_multicycle_path) is a functionally valid, real data path where hardware protocol guarantees the capture flip-flop samples data $N$ cycles later. STA still validates timing, but with relaxed cycle budgets.
When you specify -setup N, the STA engine moves the setup capture edge forward by $N$ clock periods. By default, the tool performs the hold check on the clock edge immediately preceding the setup capture edge (i.e. cycle $N-1$). Without adjusting hold, the engine demands that data remains stable across $(N-1)$ clock cycles, causing severe false hold violations. Adding -hold (N-1) moves the hold check reference back to the launch cycle ($T=0$), which is the physically correct requirement.
For setup analysis: Input Delay (max) = External Driver Tcq(max) + PCB Data Trace Delay(max) - PCB Clock Trace Delay(min). For hold analysis: Input Delay (min) = External Driver Tcq(min) + PCB Data Trace Delay(min) - PCB Clock Trace Delay(max). This accurately calculates the portion of the clock period consumed before the signal reaches your chip's input pad.
set_clock_groups -asynchronous applies atomically and bidirectionally between all specified clock groups (both Group 1 → Group 2 and Group 2 → Group 1). It has higher precedence in STA tools and prunes the timing graph early, reducing tool runtime by up to 40%. In contrast, set_false_path requires manual -from and -to commands in both directions and can introduce unintended priority masking bugs.
A virtual clock is a clock defined with create_clock -name VCLK -period ... without associating it to any input port or cell pin. It is used to specify external I/O delays when the transmitting or receiving external device is clocked by an independent off-chip source or when modeling board-level clock phase shifts.
set_case_analysis forces a pin, port, or net to a static logic value (0 or 1) during timing analysis. The STA engine propagates this constant state through combinational logic and prunes inactive paths. It allows engineers to analyze different operational modes (functional mode, scan test shift, BIST, PLL bypass) independently without modifying the RTL netlist.
Source latency (set_clock_latency -source) represents the off-chip delay from the actual oscillator source to the chip's clock pin. Network latency (set_clock_latency) estimates the on-chip delay through the clock distribution tree before Clock Tree Synthesis (CTS). Post-CTS, set_propagated_clocks replaces estimated network latency with actual extracted wire delays.

Explore Next in Static Timing Analysis & RTL

← Clock Tree Synthesis (CTS)
SDC copied to clipboard!
Advertisement