One of the key reasons to migrate EDA workloads to the cloud is to take advantage of the ability to scale compute infrastructure up and down as needed. Some compute-intensive tasks—such as power estimation and noise analysis, as well as formal verification—are better suited than others to be broken up into smaller parts across massively distributed compute and storage resources, which is a native cloud approach.
It’s also important to evaluate the compute resources available: How would you access them? What kinds of storage technologies are used? What kinds of cloud instances are available? What kinds of file system options are provided? Where are the data centers located and what is the latency?
Migrating EDA workloads to the cloud takes advantage of the ability to scale compute infrastructure up and down as needed.
To effectively manage distributed workloads, EDA flows are ideally supported by robust job scheduling with streamlined storage. Design and verification solutions that accommodate the addition of cores on-the-fly provide the elasticity for such burst tasks.
An example of a highly scalable, cloud-based solution for signoff physical verification is Synopsys IC Validator. While physical verification has long relied on a single-CPU approach, in the last decade, we’ve seen a need emerge to utilize a distributed architecture for this task. Our IC Validator engineering team has been innovating, improving its architecture to scale to thousands of cores available on the cloud and implementing a distributed scheduler to make the most efficient use of these compute resources. With its elastic CPU management technology, you can add or remove resources on-the-fly. IC Validator can complete the largest advanced-node physical verification jobs within hours versus what could have taken weeks earlier.
The availability of hybrid scaling is another cloud advantage, providing the flexibility to run certain workloads on premises or on the cloud based on the needs of the task at hand. For example, consider library characterization, a highly parallelizable task that requires a substantial amount of compute resources. Designers often struggle with resource planning for library characterization because of how unpredictable the periods of demand for compute resources are. Utilizing the cloud and a continuous deployment pipeline for library characterization can reduce turnaround time from weeks to days via on-demand access to as much compute resources as needed when needed. Synopsys PrimeLib provides a library characterization solution that scales beyond 10,000 parallel jobs on the cloud, delivering 10x faster turnaround time.
Re-architecting EDA solutions to leverage cloud architectures natively is very important. This is no different than how EDA products embraced multi-threaded and multi-processing opportunities in the past. Embracing new technologies like distributed computing, distributed storage, and micro-service architectures, while keeping in mind the cost of results by building technologies like checkpointing to leverage spot-pricing etc., will go a long way from optimized to native cloud journey.