System-on-Chip Design In The Cloud: One Size Does Not Fit All

At an increasing pace, companies in the semiconductor ecosystem have started seriously considering the cloud for computing and storage. Some have migrated, and others are evaluating the cloud technology choices and are sizing the business impact and benefits to make the leap. Through key adoption reports, the cloud environment is proving to be beneficial for System-on-Chip (SoC) designers by providing access to unlimited compute capacity, flexibility, on-demand tailored virtual machines, advanced computing services, and enabling multi-site collaboration. The various semiconductor companies have different cloud solution requirements. Their business goals and chip design workflows will likely dictate their cloud migration path and their choice of compute environment – whether it is moving to the cloud or choosing an on-premises data center, or a combination with adoption considerations.

Like industries in healthcare and finance, chip design companies have stringent requirements for cloud solutions, including security, just-in-time monitoring, and compliance with regulations. In addition, the type of cloud environment and model will differ by the company’s size and business goals as well as their chip design and verification methodologies. Hence, it is not one size fits all. Therefore, a successful migration and journey to the cloud requires a performance-cost-based assessment and identifying which chip design and verification workflows to move to the cloud. This includes decisions on a single or multi-CSPs (Cloud Services Providers) environment, the type of cloud model (“all in the cloud” or hybrid), and the types of automation and license models.

Fig. 1: The percentage of design project time spent on verification.

Moving design and verification workflows to the cloud – which ones?

Chip design complexity is growing, and the verification workflow continues to be a significant portion of the chip design development cycle. Looking closely at IC design projects, the time spent on the verification workflow, both analog and digital circuit simulations, is a higher percentage (figure 1) of the overall design project cycle. The highly iterative circuit simulation tasks span the entire IP design and verification cycle. Furthermore, every process node migration to the next smaller process geometry results in more simulations due to more packed blocks and more sub-systems to verify to realize the intended performance boost with the next node. These parallelizable circuit simulation tasks make it a suitable workflow to move to the cloud, reducing runtimes from weeks to days. Advanced node library characterization is another highly parallelizable workflow. Characterizing an entire standard-cell library requires hundreds of millions to billions of SPICE simulations, taking days to weeks to complete. With cloud computing, library characterization teams can accelerate their library characterization workflows and complete them within 24 hours.

The availability of compute resources limits the number of simulations each designer can run during the entire design cycle. Typically, design teams run as many simulations as they can with the available compute resources – limiting the extent to which they can perform design space exploration to ensure the most optimal PPA (performance, power, and area) and deliver competitive products on schedule. Furthermore, the compute resource constraint forces design teams to prioritize their various simulation-plan tasks to meet the development deadline, which often leads to cutting corners based on the best historical judgement that may not necessarily translate to thorough verification – hence, putting the design at risk. Cloud computing is a viable option to provide scaling compute resources and meet the demand during peak times of the chip design cycle. The decisions on the type of cloud solutions, migration path to the cloud, and which workflows to employ all must be made thoughtfully with minimal disruption to the existing chip design environment. To that end, learn how Siemens EDA, as a trusted advisor in collaboration with customers, provides cloud-ready AMS verification solutions to help accelerate SoC verification and library characterization on the Amazon Web Services cloud. This enables chip design teams to boost productivity and shorten their time-to-market schedules.

Selecting the cloud migration path and adoption

A customer’s choice of cloud environment and migration depends on several factors, including the company’s size and the type of EDA workflows. There are generally three categories of companies. This includes enterprise (large) companies with an established IT, support team, on-premises compute infrastructure and a central CAD team. They typically have been transitioning to the cloud with a do-it-yourself (DIY) approach. These companies adopt the hybrid cloud model to augment their on-premises compute clusters with cloud computing to meet peak compute capacity demand and customize the cloud application to their needs. Then there are mid-size companies with some compute infrastructure and a CAD team. They typically require automation assistance to implement and support their choice of a cloud model, whether for an “all in the cloud” or a hybrid cloud flow. The third category is small and startup companies with minimal to no computing infrastructure and technical expertise in the cloud. Building a compute infrastructure is costly and has yet to be a feasible option for them. Hence, the trend for these companies has been to outsource their compute capacity requirement to a third party, such as a managed cloud environment service, removing the overhead of internal maintenance.

Cloud-ready and cloud-certified EDA technology tools are essential for companies to successfully migrate their chip design and verification workflows to the cloud. Collaboration within the semiconductor ecosystem, including the end customer and CSPs, is paramount to providing tried and tested optimal cloud reference environment templates, architectures, and best-known methods. Learn how Siemens EDA’s collaboration with CSPs offers customers various cloud solutions to make the cloud journey and adoption easier.

Fig. 2: Example of hybrid cloud flow for an Interactive design analysis and simulation workflow.

Delving more into hybrid cloud

Cloud-ready EDA tools must be optimized and flexible enough to handle a hybrid cloud environment and enable a portion of the chip design and verification workflows to run efficiently on the company’s on-premises compute clusters and on the cloud. One driver for large companies to migrate to a hybrid cloud model is the ability to keep sensitive data in local compute cluster environments, eliminating the potential for security breaches of company IP or licensed third-party technology. Another reason is to capitalize on existing compute infrastructure and the flexibility to support custom flows.

Integration of the workflows for a hybrid environment must support the interaction between the data on-premises and the data in the cloud. For example, for a GUI-driven interactive circuit design and simulation workflow (figure 2), considerations must include key building blocks – such as the simulation job scheduler and job management – to allow the designer to augment their on-premises seamlessly with cloud computing during peak simulation workloads. Moreover, an optimal data transfer latency during downloading and viewing of simulation results is a crucial metric for acceptance of a hybrid cloud workflow. Most companies prefer to build a custom hybrid cloud flow that meets their specific requirements with automation provided by the EDA vendor. Siemens EDA’s cloud solution portfolio provides a Siemens-managed cloud service for a turnkey design environment and scalable computing.

Flexible access to EDA licenses

A given company’s requirement for which EDA license access to use in the cloud will vary based on business and operational goals and the services and products that the company delivers to their end customer. A common theme is a flexible and granular access based on existing EDA license configurations to specifically address their peak demand at various stages of the chip design cycle – with minimal disruptions. Ultimately, the license models should be flexible enough to seamlessly allow chip design teams to use on-premises and peak-demand licenses in congruence.

Conclusion

It is imperative that chip design companies, EDA solution providers, and CSPs collaborate for a successful migration to the cloud. An upfront assessment to determine the cloud environment and model that best fits the given semiconductor companies’ requirements is crucial. As a trusted advisor, Siemens EDA has collaborated with chip design and verification companies and with CSPs to provide cloud-ready products and solutions to help deliver scalable performance for peak capacity needs, ease cloud adoption, and maximize productivity. Learn more at Siemens EDA Cloud solution portfolio.