Today’s market forces have changed the paradigm for semiconductor manufacturers from “innovate to succeed” to “innovate or perish.” High tech customers want microprocessors tailored to their specific needs. Device design has become impossible without automated semiconductor design software that requires massive compute and data storage capacity. Hence, computing models for running automated semiconductor design workloads on premises need to evolve and chip manufacturers must explore alternative deployments for their design workloads.
A good fit for new companies, this approach brings scalability, elasticity and high performance as well as better performance, collaboration, etc.
Hybrid public/private cloud
This approach allows established companies to leverage their existing infrastructure while taking advantage of the benefits of the public cloud.
Flexible on-premises private cloud
The server OEMs install servers at the facility of the semiconductor client, that can be accessed for a fee during high demand periods.
Different deployment models offer varying degrees of scalability and cost benefits. Considering organizational priorities, companies can match-up with the best model that will help them succeed in the modern environment.
Electronic design automation (EDA) software is an essential part of chip development, but it is highly resource intensive and complex. Analog design requires different tools and methodologies than digital design, each having their own challenges to cloud execution and compute needs. For optimal performance, the hardware configuration needs to be tailored to the application.
A semiconductor high-performance compute (HPC) environment involves enormous capital investment as well as EDA software licensing fees, which scale with infrastructure.
Restricting semiconductor design workloads to private resources have drawbacks like security, value of the intellectual property (IP), ransomware attacks, the evolving threat environment, mismatch between resource need & availability and the engineering teams competing with one another for capacity.
A hybrid cloud provides a scalable, elastic HPC environment with unlimited compute cycles and storage. It can use the public cloud for burst workloads, returning capacity to the cloud service provider (CSP) rather than maintaining idle servers, thus saving cost. The engineering teams get immediate access to compute, storage and analytics during needful times.
For companies committed to maintaining their EDA workloads on premises, several server OEMs are offering flexible private compute models that provide cloud-level scalability while reducing total cost of ownership. During high demand periods, the company could immediately access these “cold” servers, paying a usage fee for the surge period only.
State of EDA on Cloud today: While EDA applications today run mostly on premises, in the long term, the greater value for both EDA vendors and clients would be in the software becoming cloud native.
Scalability and flexibility
Immediate access to compute and storage resources by CSPs. The latency issues are addressed by on-premises private clouds.
CSPs offer sophisticated tools and services for innovation. Design teams can leverage software tools from the EDA ecosystem.
Instant access to compute and storage resources enhances productivity, enabling simultaneous testing while helping boost efficiency.
Predictability and visibility
Presenting a comprehensive view of the various functions by timelines, time for closure and any potential roadblocks.
Hyperscalers offer a comprehensive suite of automated tools, state-of-the-art best practices like encryption of data in flight and at rest.
Moving to action
The choice of deployments for semiconductor design workloads depends on a number of factors, including where the enterprise is in its corporate journey. Unlocking value requires a strategic approach.
Determine which workloads belong in the cloud
EDA workloads require specific hardware configurations. It’s essential to match the workload to the appropriate instance to get optimal performance.
Be mindful of data cost
Data management should be a key consideration in the decision of which workloads to deploy to the public cloud and which to leave on premises.
Focus on value
As soon as the infrastructure-that is the tool for doing business-is provided by a CSP, companies should move capacity to a public infrastructure.
Organizations should run a pilot project to learn the process, uncover weaknesses and make any failures on a small scale.
Don’t go it alone
Semiconductor companies should work with software vendors, CSPs, system integrators and/or server OEMs for guidance for a successful pivot.
For semiconductor companies, infrastructure is just a means for achieving their ultimate goal of designing and building chips. Moving EDA workloads to a flexible cloud infrastructure is the logical endpoint for the semiconductor industry. Hence, the chip companies should be focused on identifying a migration path to a flexible cloud infrastructure for their EDA workloads.