—Joni Mitchell Unless you really haven’t been paying attention, everyone is talking about the cloud, and the EDA world is no exception. At DAC last month, the two main themes were machine learning and EDA in the cloud. (Paul’s recent blog posts about DAC, including DAC Monday , Tuesday , and Wednesday , and the "Straight Talk" discussion with Lip-Bu Tan will give you more of an idea.) I’ve written extensively about ML/DL, but haven’t touched much on the cloud yet. So let’s talk about it. How We Work: On-Premise or in the Cloud Here’s a question. What do you do when you’re at work, working? (Sure, it depends on what your job is, but I’m talking more in the abstract). You get to work, and after getting a coffee and checking your email, and from your desktop (or My Computer or somewhere native to the box that you call your computer), you open the application (say, Microsoft Word or Excel or OrCAD or Photoshop or whatever) and navigate to the file that you’re working on. You have a license from the company that made the application to have that instance of the application on your computer, and all of the processing required by that file is done locally. After you do whatever amazing thing it is that you do, you save the file, then you send it on to whomever is in the next stage in whatever process you’re working on. But if you do your work in the cloud, the process is slightly different. Instead of opening the application for which you have a license from your computer, you open a browser and log in to the website of the application. From that company’s website, you have access to your application that contains the file(s) that you’re working on, where you can perform amazing feats of whatever it is that you’re doing, and the processing of that file is done using the processing power of the application’s company or a third-party provider. (Incidentally, you don’t need as much processing power on your computer as you did when it was done locally). When you’re done, you then send the file—that is, a link—to the next person who has to do the next thing in your process. This can also be known as SaaS: software as a service. How Chip Verification Engineers Work Engineers are working on designing a chip, and after—well, during—the chip is (being) designed, verification engineers make sure that each little piece of the design works as it’s supposed to. We’re talking about millions and millions of variables. According to this white paper , verification can take, on average, over half of the development time of a chip. And this is not a steady need—there are peaks and valleys of need over the course of chip development. Believe it or not, verification engineers work the same way that the rest of us do. But here, the point about processing power is much more relevant. If you are processing umpty-squillion bits of data, you can either do it with your company’s own stack of servers (or even your company’s own datacenter) (the processing power of a single computer is a drop in the bucket of what you need)—or you can do it in the cloud. Servers and datacenters are expensive, unwieldy, and require a lot of person-resources to get them up and running, not to mention maintaining them after they’re in place. If your company has a processing power to spare, great—you are probably part of a big company. Most of us don’t have access to an entire fleet of servers that we can use as soon as we need them. Even if you are part of a big company that has its own datacenters, there’s no guarantee that you can access the computing power that you need, or that the servers there are the most efficient for what you need. You have to get IT onboard to work with you to provision for your verification needs, as shown below. Capacity vs. Usage, IT Approach So come verification time? It may be prudent to look to the cloud. When using the cloud for verification, peak needs are addressed by periodically “bursting” onto the cloud when compute requirements and schedule demands require. Using this approach, workloads associated with peak needs are moved to the cloud at a job- or complete project-level, allowing flexible and efficient scaleability to meet development schedules. This can also be known as IaaS: infrastructure as a service. The Cadence Cloud This is precisely what the Cadence Cloud is all about. With three different models in the Cadence Cloud portfolio, chip designers and engineers can access all of the processing power they need—especially for verification, simulation, and emulation—without needing to wait for their own infrastructure to be put into place. Customers can also decide whether they want to manage the cloud themselves, whether they just want emulation without installing specialized hardware to manage the workload. Cloud Passport Model Customer-Managed A model that provides easy access to cloud-ready Cadence software and a cloud-based license server with high reliability for customers establishing and maintaining their own cloud environments. Cloud-Hosted Design Solution Cadence-Managed A managed EDA-optimized cloud environment that supports customers’ peak or entire design environment needs. Can be flexibly deployed as a single tool, flow, or complete environment on Amazon Web Services (AWS) or Microsoft Azure. Cadence Palladium Cloud Solution Cadence-Managed A managed emulation cloud solution that frees customers from installation and operational responsibilities. Can be deployed in combination with other Cadence Cloud offerings. There is lots more information about our Cloud offerings here , including a nice introductory animation about it. Check it out! So maybe I know something about the cloud, after all. www.youtube.com/watch —Meera![]()
↧