Home
  • Products
  • Solutions
  • Support
  • Company

This search text may be transcribed, used, stored, or accessed by our third-party service providers per our Cookie Policy and Privacy Policy.

This search text may be transcribed, used, stored, or accessed by our third-party service providers per our Cookie Policy and Privacy Policy.

  • Products
  • Solutions
  • Support
  • Company
Community Blogs Breakfast Bytes > AWS: Amazon's Own Experience with EDA in the Cloud
Paul McLellan
Paul McLellan

Community Member

Blog Activity
Options
  • Subscribe by email
  • More
  • Cancel
aws
cadence cloud
Amazon

AWS: Amazon's Own Experience with EDA in the Cloud

1 Oct 2021 • 3 minute read

cadenceLIVE At CadenceLIVE Americas there was a cloud track. Opening the day was David Pellerin of Amazon/AWS who presented Accelerating EDA Productivity, The Why, What and How of the Journey to the Cloud. He emphasized that he would be including several examples of AWS's own experience doing design in the cloud.

Amazon designs a fair bit of its own silicon and sources it from the global supply chain...oh yes, and designs it all in the AWS cloud:

  • datacenter infrastructure (such as Graviton and the Nitro Project)
  • consumer devices (Alexa)
  • robotics and AI (for warehouses?)
  • and more

David said that there were a number of reasons to go to the cloud:

  • innovate faster
  • collaborate better
  • reduce risk
  • reduce total cost

I'm going to assume that you already know that AWS has datacenters across the world in 25 regions and 80 availability zones, and networks connecting everything. One area I've written about extensively is Annapurna, which is a company in Israel that AWS acquired that designs the Graviton Arm-based server processors and a lot more. Its EDA flow is entirely cloud-based. See my post Climbing Annapurna to the Clouds for some more detail and also Nitro, a whole silicon infrastructure that allows AWS to deliver near-bare-metal performance with a hardware-based hypervisor. For more details on that, see my post HOT CHIPS: The AWS Nitro Project, 

 The reason AWS designs their own chips is for specialization, speed (to market), security, and innovation. Like so many system companies that are designing their own silicon, they cannot get differentiation and performance with off-the-shelf parts.

AWS's Own Journey into the Cloud

David focused on Annapurna but also reminded us all that there are other groups in Amazon doing custom silicon. When he says "custom silicon" he doesn't mean as opposed to digital, how we normally use it in Cadence, but as opposed to merchant silicon. Maybe we should use "bespoke" as upmarket tailors do.

Most of Annapurna Labs' developers are in Israel, but some are in Texas and California, so multisite support is very important. They started migration via a hybrid model, with on-prem data centers and cloud. Gradually they reduced the amount of on-prem data centers until today they only have emulation in their own data centers and everything else is in the AWS cloud. Even the emulator builds are done in the cloud.

Semiconductor design flows are complex, so to move it all into the cloud there is a lot to move. There are also other complications like partners, 3rd party tools, contract designers, global distribution, massive compute requirements, cooperation with foundries and OSATs, and so on You can see the details in the above picture. David also pointed out that they see most traction with the stuff around tapeout like static timing analysis. Even though these are harder to move into the cloud than some of the earlier stuff, it is the part of the design cycle than get benefit most from being able to get hundreds or thousands of machines. Lots of companies, including Annapurna Labs, are running full flow, but the place where people build a beachhead is design closure and signoff.

Supply Chain

David went on to discuss some aspects of the entire semiconductor supply chain and how it can be cloud-enabled. The semiconductor supply chain involves moving large amounts of data, intense security, and lots of cooperation between design groups, EDA companies like Cadence, cloud providers, design services providers, foundries and more. Sometimes a lot of these groups together, such as debugging an EDA tool with PDK information from the foundry, package information from the OSAT. The best approach is to use cloud-based collaboration chambers.

And if you pull everything together across the whole supply chain you get to his last intimidating slide (I'm sure the kitchen sink is in there somewhere!):

Learn More

If you were registered for CadenceLIVE Americas then you can watch a replay of David's presentation.

Or see the Cadence Cloud product page.

 

Sign up for Sunday Brunch, the weekly Breakfast Bytes email