As chips continue to become more complex and larger, chip design and verification resources are becoming a bottleneck in the face of increasing time-to-market pressures. At the same time, engineering workloads are growing, with engineers consistently managing more on their plates with fewer resources. Compared to running EDA solutions on an on-premises data center, utilizing the cloud opens up significantly more compute resources to accelerate fundamental chip design and verification processes. There’s also the benefit of elasticity—the ability to quickly scale up or down based on needs.
As an example, let’s consider library characterization, a highly parallelizable task that needs a huge amount of compute resources. Resource planning for library characterization has been notoriously difficult. Pre-cloud computing, for instance, chip design houses invested in their own high-performance data-center capacities for these workloads. But these systems would either be over- or under-utilized based on demand patterns or they’d require sequencing of workloads, causing delays. Cloud computing, on the other hand, can reduce turnaround time (TAT) of tasks such as library characterization from weeks down to days by providing on-demand access to as much compute resources as needed when needed. Amazon Web Services (AWS) customers, for example, have been able to scale their library characterization workloads to more than 120,000 parallel jobs thanks in part to a close collaboration between AWS and Synopsys.
Tasks that are characterized by burst usage periods are ideal for moving to the cloud. Without having to invest in CapEx-heavy infrastructure themselves, designers have the flexibility to tap into compute resources on the fly. If needed, they can also break up their compute-intensive problems into smaller pieces and use cloud-based, massively distributed processing and storage to tackle each piece, provided that the data is partition-friendly. Applications such as timing analysis and physical and functional verification scale very well when distributed. With formal verification, for instance, you can localize the design and perform the verification in independent parts.