Astera Labs Uses AWS to Accelerate Chip Development

QuadX

Meeting Tape-Out Deadlines

In semiconductor development, the process of designing integrated chips culminates in the tape-out phase, at which point chip designs are released for wafer fabrication. For semiconductor companies like Astera Labs, getting to the tape-out phase requires a massive amount of compute capacity to quickly perform complex simulations and to place and route millions of transistors that physically represent the chip. Astera, founded in 2017 in California’s Silicon Valley, is a fabless semiconductor company that develops purpose-built connectivity solutions for data-centric systems. “We have to be able to scale our workloads during development to run many simulations in parallel, so we can verify that our chip will work as intended and deliver the high quality our customers expect,” says Sanjay Gajendra, chief business officer for Astera Labs.

If Astera is unable to run enough simulations during the chip design and verification process, the company risks missing potential design flaws. “We get one chance, and, if we make a mistake, it’s a six- to nine-month process to find the mistake and fix the problem,” says Gajendra. “That process is very costly and time-consuming, which we can’t afford as a startup.”

“We completed our chip design in less than a year on AWS, compared to the years it would have taken using an on-premises HPC environment.”

Sanjay Gajendra, Chief Business Officer, Astera Labs

  • About Astera Labs

  • Astera Labs, Inc., a fabless semiconductor company headquartered in California’s Silicon Valley, develops purpose-built connectivity solutions for data-centric systems. The company’s product portfolio includes system-aware semiconductor integrated circuits, boards, and services to enable robust PCIe connectivity.

  • About Intel

  • Intel is the world’s leading designer and manufacturer of high-performance processors for servers, PCs, IoT devices, and mobile devices. AWS and Intel engineers have worked together for more than 10 years, building custom hardware to ensure AWS services run on a platform optimized for customer workloads for the best value. Intel Xeon Scalable processors power Amazon EC2 instances to help enterprises drive performance for their compute-intensive workloads.

  • Benefits
    • Runs more than 1,000 chip design simulations overnight on AWS
    • Completes chip design in months instead of years
    • Meets deadlines for tape-out process
    • Avoids spending many months and millions of dollars fixing chip design errors
  • AWS Services Used

Running HPC Workloads on AWS

To address potential challenges in meeting tape-out deadlines, Astera has run its HPC workloads in the Amazon Web Services (AWS) Cloud since the company’s founding. “We thought the public cloud would give us more scalability and be less costly than buying and maintaining a server farm,” says Gajendra. “We selected AWS because we considered it the leader in the cloud industry and had confidence AWS services would help us minimize our risk.”

Astera worked with Six Nines, an AWS Partner Network (APN) Premier Consulting Partner, to create an HPC environment to support chip design and verification simulations. The platform runs on thousands of Amazon Elastic Compute Cloud (Amazon EC2) R5, C5, and z1d instances, powered by Intel Xeon Scalable processors. The company uses the R5 and C5 instances for running physical design and simulation jobs and the z1d instances for long-running simulations that require the fastest-possible virtual machines. Astera also uses Amazon Simple Storage Service (Amazon S3) buckets to store more than 100 TB of data.

Six Nines and AWS Professional Services helped Astera implement the HPC platform on AWS in just two weeks, instead of the three months it would have taken Astera to set up the platform in a data center, with AWS Professional Services designing and configuring the environment.

Running Hundreds of Chip Design Simulations in Parallel

Astera can now run hundreds of chip design simulations in parallel and speed design verification time by using AWS. “We can implement a very complex chip design on AWS,” says Gajendra. “AWS improved our chip design and verification environment, so we have the compute resources we need to run more than 1,000 simulations overnight. That would be very difficult to do in an on-premises data center, given the capital requirement.” With its new capabilities, Astera was able to complete the tape-out of a new chip that implements PCIe 5.0 specification, which achieves 2Tbits/sec of bandwidth over an x16 interface. The chip meets industry requirements for high bandwidth and low-latency interconnects.

Relying on AWS, Astera can also quickly identify and fix bugs in a virtual environment before a chip design is sent for tape-out. As a result, the company finished the chip design and verification process for its new chip despite an aggressive timeline. “We completed our chip design in less than a year on AWS, compared to the years it would have taken using an on-premises HPC environment,” says Gajendra. “Using AWS to enable cloud scalability and provide the latest instance types to eliminate bottlenecks for our designers, we have reduced our cycle time for design verification runs and physical implementation by a factor of two or more.” Additionally, Astera can produce more chip designs, and higher-quality products, than its competitors as a result of the compute scalability enabled by AWS.

Avoiding Millions of Dollars in Costs

Taking advantage of its more efficient chip development process, Astera avoids spending several months correcting mistakes and investing millions of dollars to overhaul a new chip design following a tape-out. “We don’t want to have our chip fail in production or in our customers’ systems and spend months fixing the problem,” says Gajendra. “We can avoid this issue by ensuring a high-quality tape-out using AWS.”

The company also benefits from using multiple AWS Availability Zones to maintain high availability for its HPC environment during design and development. Gajendra says, “Overall, using AWS, we can meet strict industry deadlines, fulfill device performance expectations, and continue focusing on creating the best chips we can build.”

Learn More

To learn more, visit aws.amazon.com/hpc.