Library Characterization Tidbits: Recovering from Failures in the Multi-PVT Characterization Flow

Hello readers,

Welcome to the first blog of 2021 in the Library Characterization Tidbits series!

Today, I’ll share with you some tidbits about the multi-PVT characterization flow of the Liberate Trio Characterization Suite. It is a solution for the nanometer designers who find characterization of multiple Process, Voltage, and Temperature (PVT) corners challenging in the times when the library sizes are growing extensively in the advanced nodes.

The Liberate Trio solution that lets us run Liberate characterization for multiple PVT corners in a single run was first introduced in the LIBERATE 18.1 base release. This cloud-ready solution is scalable to thousands of cores for improved throughput. It employs advanced machine-learning techniques to determine critical corners for characterization.

The multi-PVT characterization flow eliminates redundant analysis for each PVT corner by reusing the results found on a single default corner. Thus, it saves our time and efforts.

While working with the customers, we often come across questions like the following:

What should I do if any arc, cell, or PVTs failed in the current run? Do I need to rerun the entire characterization process?

The answer is a simple no. In LIBERATE 19.2, we introduced a new feature called recovery workflow in the multi-PVT characterization flow that not just helps to recover from failed arcs, cells, and PVTs, but also lets you add new cells, PVTs, and so on in a library.

The recovery workflow, which is based on the read_ldb command, helps us to:

1. Find all arcs that had an error when they were run the last time.
2. Find all arcs that did not run the last time because the session was killed abruptly.
3. Recharacterize a few or all cells and PVTs that are specified using the multi_pvt_recovery_rechar parameter.
4. Recharacterize specific or all individual arcs for a cell and PVT combination using the -type option with the set_var command.
5. Rerun the preprocessing job for all cells that have been set using the multi_pvt_rechar_do_preprocessing parameter.

What are the various settings required to enable the recovery workflow in the multi-PVT characterization flow?

To use the recovery workflow in the multi-PVT (define_pvt-based) characterization flow, first set the multi_pvt_recovery_flow parameter. Then, add the read_ldb command to point to the LDB database you want to read in the recovery session. You must note here that in a fresh characterization run, until a write_ldb command is run, the temporary LDB database is saved to the ldb_checkpoint_dir directory. So, if you are trying for recovery before the write_ldb command has been run, you must point to this temporary LDB database directory. However, after running the write_ldb command, the LDB moves from its temporary location to the <LDB>.ldb.gz location that can be used for recovery in the subsequent runs.

The recovery workflow supports the following:

• Changes in preprocessing (vector database) for any cell
• Recharacterization of specific arcs using arc IDs
• Recharacterization of specific cells and PVTs
• Changes in the characterization settings, such as parameters set using the set_var command, for the recovery flow
• Distributed modeling flow to model new libraries or make changes in the setup, such as set_attribute and user_data, that impact modeling only in the distribution mode

To know more about running the recovery workflow, download the Liberate Trio Multi-PVT Recovery Characterization Flow Rapid Adoption Kit (RAK).  

I'll recommend that you try this RAK because it provides a sample test case and an accompanying tutorial to understand the steps involved in the recovery workflow. You can also search for this RAK on the page accessed from the Resources – Rapid Adoption Kits menu of the Cadence Support portal.

Goodbye till I return with another topic!

-Rajni Dhiman

Related Resources

About Library Characterization Tidbits

Library Characterization Tidbits is a blog series aimed at providing insight into the useful software and documentation enhancements in the LIBERATE release. In addition, this series would broadcast the voices of different bloggers and experts, who would share their knowledge and experience about all the tools in Liberate Characterization Portfolio. To receive notifications about the new blogs in this series, click Subscribe and submit your email ID in the Subscriptions box.