parametric ocv (pocv)

**Parametric OCV (POCV)**, also known as **Statistical OCV (SOCV)**, is the most advanced **on-chip variation modeling methodology** that uses **per-cell statistical delay distributions** rather than fixed derate factors — computing path delay variation as the statistical combination (RSS) of individual cell variations for the most accurate and least pessimistic timing analysis. **How POCV Differs from AOCV** - **Flat OCV**: One derate percentage for all paths — crude, overly pessimistic. - **AOCV**: Depth-dependent derate tables — better but still uses fixed multipliers per depth. - **POCV**: Each cell has its own **mean delay ($\mu$) and standard deviation ($\sigma$)** — path variation is computed statistically: $$\sigma_{path} = \sqrt{\sum_{i=1}^{N} \sigma_i^2 + \left(\sum_{i=1}^{N} \sigma_{sys,i}\right)^2}$$ Where $\sigma_i$ is the random variation of cell $i$ (uncorrelated, RSS) and $\sigma_{sys,i}$ is the systematic variation (correlated, adds linearly). **POCV Data** - Each cell in the library has a **POCV Liberty Variation Format (LVF)** file containing: - Nominal delay (mean) for each timing arc. - Random variation (σ) for each timing arc — uncorrelated between cells. - Systematic variation component — correlated across nearby cells. - Variation as a function of input slew and output load. - POCV data is derived from **extensive silicon characterization** and Monte Carlo SPICE simulation during library development. **How POCV Works in STA** 1. Each cell's delay is modeled as a distribution: $d_i = \mu_i ± k \cdot \sigma_i$ where $k$ is the sigma multiplier (typically 3σ for 99.87% coverage). 2. Random variations of different cells are **uncorrelated** — they combine as root-sum-of-squares (RSS). This is the key advantage: adding more stages reduces the relative variation. 3. Systematic variations are **correlated** — they add linearly (worst case). 4. The tool computes total path delay variation and applies it as a derate. **POCV Benefits** - **Least Pessimistic**: POCV provides the tightest (most realistic) timing bounds — typically **5–10% less pessimistic** than AOCV, and **15–25% less pessimistic** than flat OCV. - **Most Accurate**: Directly models each cell's actual variation characteristics — no approximation by depth or distance. - **Better Path Differentiation**: Two paths with the same depth but different cell compositions get different variation estimates — a path through high-σ cells gets more derate than one through low-σ cells. - **Silicon-Correlated**: POCV data is validated against silicon measurements, ensuring the analysis matches real chip behavior. **POCV Challenges** - **Data Requirements**: Requires per-cell variation data (LVF files) — significant characterization effort. - **Compute Cost**: More complex calculations than flat OCV or AOCV — but modern STA tools handle this efficiently. - **Foundry Support**: Not all foundries provide POCV/LVF data for all process nodes — availability is expanding. POCV represents the **state of the art** in OCV modeling — it provides the most realistic timing analysis by treating each cell as a statistical entity rather than applying blanket derating.