NLDM (Non-Linear Delay Model) is the foundational table-based timing model used in Liberty (.lib) files — representing cell delay and output transition time as 2D lookup tables indexed by input slew and output capacitive load, capturing the non-linear relationship between these variables and delay.
Why "Non-Linear"?
- Simple linear delay models (e.g., $d = R \cdot C_{load}$) assume delay is proportional to load — this is only approximately true.
- Real cell delay vs. load relationship is non-linear: at low loads, internal delays dominate; at high loads, the driving resistance matters more.
- Similarly, delay depends non-linearly on input slew — a slow input causes more short-circuit current and affects switching dynamics.
- NLDM captures this non-linearity through table interpolation rather than equations.
NLDM Table Structure
- Two tables per timing arc:
- Cell Delay Table: delay = f(input_slew, output_load)
- Output Transition Table: output_slew = f(input_slew, output_load)
- Each table is typically 5×5 to 7×7 entries:
- Rows (index_1): Input slew values (e.g., 5 ps, 10 ps, 20 ps, 50 ps, 100 ps, 200 ps, 500 ps)
- Columns (index_2): Output load values (e.g., 0.5 fF, 1 fF, 2 fF, 5 fF, 10 fF, 20 fF, 50 fF)
- Entries: Delay or transition time in nanoseconds
- During timing analysis, the tool interpolates (or extrapolates) between table entries to get the delay for the actual slew and load values.
NLDM Delay Calculation Flow
1. The STA tool knows the input slew (from the driving cell's output transition table). 2. The STA tool knows the output load (sum of wire capacitance + downstream pin capacitances). 3. Look up the cell delay table → get propagation delay. 4. Look up the output transition table → get output slew. 5. Pass the output slew to the next cell in the path. 6. Repeat through the entire timing path.
NLDM Limitations
- Output Modeled as Ramp: NLDM represents the output waveform as a simple linear ramp (characterized by a single slew value). Real waveforms are non-linear.
- No Waveform Shape: At advanced nodes, the actual shape of the voltage waveform matters for delay, noise, and SI analysis — NLDM doesn't capture this.
- Load Independence: NLDM assumes the output waveform shape is independent of the downstream network's response — actually, the load network affects the waveform.
- Miller Effect: The non-linear interaction between input and output transitions (Miller capacitance) is not fully captured.
When NLDM Is Sufficient
- At 45 nm and above: NLDM is generally accurate enough for most digital timing.
- At 28 nm and below: CCS or ECSM provides better accuracy, especially for setup/hold analysis and noise.
- Most digital logic: NLDM remains widely used for standard timing analysis even at advanced nodes, with CCS/ECSM used for critical paths.
NLDM is the workhorse timing model of digital design — simple, fast, and accurate enough for the vast majority of timing analysis scenarios.
Explore 500+ Semiconductor & AI Topics
From EUV lithography to CUDA optimization — search the full knowledge base or chat with our AI assistant.