Part of the  

Solid State Technology

  and   

The Confab

  Network

About  |  Contact

Headlines

Headlines

Advances in Post-Tape Out Resource Management

By Mark Simmons, CalCM Product Manager, Mentor Graphics, Corp.

Foundries, both pure-play and independent device manufacturers (IDM), rigorously compete for market share. One factor that helps them get and keep business is turnaround time (TAT). As technology nodes advance, achieving targeted production runtimes in the post-tapeout flow gets ever more challenging. TAT is important to the fab’s customers, who are keenly interested in meeting their aggressive time-to-market schedules, and also to the fab because time means resources and money. Although a need for better control of and (in most cases) improved reduction of TAT has always been a challenge, the need for a TAT automated controlling mechanism in the post-tapeout flow has been continually neglected because simply adding more hardware has sufficed up until recently. In addition to adding compute hardware, companies also use these three strategies to achieve faster turnaround time (TAT):

  • Fine tuning OPC recipes
  • Improving input hierarchical handling of incoming designs
  • Upgrading to newer software versions to take advantage of new functionalities and performance benefits

Historically, distributed processing tasks require an upfront allocation of hardware and software resources. These resources must be pre-determined prior to starting the run, and are held for the duration of the run. As the jobs run, you can see periods when not all of the allocated resources are being used. Those resources could be used by other jobs, but there is no way to gain access to them. Clearly, this situation can be improved by better utilizing what you have through dynamic resource allocation. In fact, recently published results using a cluster manager software (Guo et. al. 2013, SPIE 8684) demonstrated just over a 30% aggregate TAT improvement for a large set of jobs in conjunction with a greater than 90% average utilization across all hardware resources.

Here’s how it works. The cluster manager software governs the hardware and software resources for all jobs running on a remote compute cluster (Figure 1). It automatically provides idle resources to jobs that can use them, and revokes resources from jobs when they are not being used. In conjunction with focusing on a single job’s performance, it also improves the efficiency of resources at the cluster level, considering all cluster tasks together as a whole. The software optimizes the distribution of resources to all of the jobs running simultaneously on the cluster. In doing so, the result is both an overall aggregate runtime performance improvement and an effective maximization of utilization across the cluster resources.

Figure 1. CalCM integration with resource management tool. (Click to view full screen)

This type of dynamic resource management does the important work of reducing TAT, which is imperative in the competitive foundry industry. It achieves dynamic resource re-distribution by automatically allocating available CPUs to jobs that can use more capacity and by removing CPUs from jobs that are already at or over capacity. Resource allocation is determined by job need, level of importance, resource partitioning, and available resources. But, this system also gives the foundry monitoring information and system- and job-level controls. These features are vital to making decisions regarding the post-tapeout flow and then providing a means to react to them in ways that make the process more efficient and improve the bottom line.

If you want to read about our experiment with deploying this cluster manager software, you can access the SPIE paper “An Automated Resource Management System to Improve Production Tapeout Turn-Around Time” http://go.mentor.com/3_7w_ (registration required).

Mark Simmons is a product manager responsible for CalCM, the cluster management product, at Mentor Graphics. He works closely with foundries and large IDMs and has gained insight into their needs and challenges. He holds a B.S in physics from the State University of New York at Geneseo, a Masters in Microelectronic Manufacturing Engineering from the Rochester Institute of Technology in Rochester, NY, and an MBA from Portland State University in Portland, Oregon. He can be reached at mark_simmons@mentor.com.



Tags: , , ,

One Response to “Advances in Post-Tape Out Resource Management”

  1. Bill Graupp Says:

    This technology is required in today’s manufacturing environment to maximize the utilization of the capital investment in cpu hardware that is necessary to meet the needs of the market. TAT is a cost which must be minimized. The question will be how this type of technology slides back into the design flow to minimize TAT of product development. I think there is a strong case to use this type of technology in the design side as well.

Leave a Reply