How NVIDIA is scaling Liberty verification for diverse IP using the Solido Characterization Suite

At our User2User North America event, Eric Hsu, Senior Circuit Design Methodology Staff from NVIDIA’s mixed signal design group, presented a session titled Scalable and Dynamic Liberty Verification for Diverse IP. The presentation outlined the challenges NVIDIA faces in managing Liberty timing models across a complex, multi-platform analog/mixed-signal (AMS) design environment and described how Solido Characterization Suite is being used to address them.

Liberty models provide the timing, power, noise, and functional data that synthesis, static timing analysis (STA), place-and-route (P&R), and signoff tools depend on, making them foundational to any chip implementation and signoff flow. As Hsu’s presentation noted, small library issues can become large silicon problems, with missing arcs, bad constraints, modelling errors, and unit mismatches capable of propagating through the design cycle and resulting in sub-optimal timing, power, and system behaviour. Discovered late, such errors can lead to expensive fixes affecting performance, power, yield, and manufacturability.

Five key Liberty challenges for mixed-signal design teams

Hsu identified five key challenges that make Liberty management particularly demanding for NVIDIA’s mixed-signal teams:

• Diverse IP: AMS designs evolve frequently, and updates to specs, pins, arcs, or corner definitions can require multiple library revisions.
• Consistency: Every delivery must keep pins, arcs, constraints, units, and complex timing models aligned across each update.
• Interoperability: Libraries must work across all AMS teams, projects, and integrated flows, where different assumptions, designs, or modelling methodologies can present unique issues.
• Correlation: AMS behaviour must be corroborated across external vendors, even though each vendor may implement or model interfaces differently.
• Scalability: Verification and QA must be fast, repeatable, and reliable across frequent updates, without slowing delivery.

To address these challenges, NVIDIA has been integrating three technologies from the Solido Characterization Suite into its Liberty workflow: Solido Generator Transform, Solido Analytics Validate, and Solido Analytics Compare.

Automating library modifications with Solido Generator Transform

Prior to adopting Solido, AMS Liberty modification at NVIDIA was described as a time-consuming and tedious process, with quality issues arising from human error. The Generator Transform utility addresses this by providing automated, repeatable, and reportable transformations on .lib files for IO and analog IPs.

Built-in transformation functions include Margin, Copy, Trim, Resize, and Merge operations, with a custom Transform API available for teams that need to apply non-standard modifications. The automated approach ensures that previously generated Liberty data can be reused, reducing the risk of manual errors that characterised earlier in-house processes.

Comprehensive verification with Solido Analytics Validate

NVIDIA’s previous Liberty verification relied on in-house scripts that suffered from long run times, low coverage, and high maintenance overhead. Solido Analytics Validate addresses these limitations, offering more than 160 built-in rule-based checks, covering areas such as:

• Missing or mismatched referenced templates
• Missing groups and timing groups missing OCV
• Propagated noise issues and OCV data issues
• Cell mismatches and DC current waveform outliers
• Non-matching NLDM, CCS, and ECSM models

Beyond static rules, Solido Analytics Validate incorporates AI-powered outlier detection that automatically identifies unexpected values in trends across process, voltage, and temperature (PVT) corners and across measurement tables (slew/load). The AI detection does not require transistor models, SPICE netlists, or characterization setup. The tool runs in both interactive GUI mode and batch mode, with multi-core support and cluster compatibility for fast turnaround.

Results are summarised, ranked, and presented via HTML reports and a GUI that visualizes outliers to aid diagnosis.

Cross-revision and cross-vendor comparison with Solido Analytics Compare

NVIDIA’s previous comparison utilities were described as limited, unable to provide meaningful insight into differences between .lib files. Solido Analytics Compare provides automated comparison of Liberty timing values and attributes, supporting two primary use cases: tracking consistency across different revisions of a library, and validating external vendor IP during integration into NVIDIA’s environment.

Discrepancies are recorded in custom Compare-API reports that note both the location and severity of differences. Results are presented in a summary table showing pass/fail status by failure type across PVT corners, with corners hyperlinked to detailed reports. Data is available in CSV format for human-readable review and JSON for front-end design implementation. Tolerances can be configured per cell type and can be set using absolute or relative methods, or via tolerance equations, allowing teams to tighten or loosen thresholds based on corner variables, pin, arc, or Liberty attributes.

Summary

The presentation from NVIDIA’s Eric Hsu at User2User North America 2026 illustrates how Liberty file management carries direct implications for silicon quality and project schedule. By deploying the Solido Characterization Suite’s Generator Transform, Analytics Validate, and Analytics Compare technologies, NVIDIA’s mixed-signal teams are working toward replacing manually intensive, script-based processes with an automated, scalable, and consistent Liberty flow. As the presentation noted, the next steps for the NVIDIA and Siemens collaboration are to further enhance automation capabilities for the end-to-end Liberty flow and deepen integration with NVIDIA’s internal tools.

Explore the Solido Characterization Suite product page to learn more about its Liberty verification capabilities and stay tuned for more Solido user stories coming soon from the User2User 2026 event.