-
- News
- Books
Featured Books
- design007 Magazine
Latest Issues
Current IssueLevel Up Your Design Skills
This month, our contributors discuss the PCB design classes available at IPC APEX EXPO 2024. As they explain, these courses cover everything from the basics of design through avoiding over-constraining high-speed boards, and so much more!
Opportunities and Challenges
In this issue, our expert contributors discuss the many opportunities and challenges in the PCB design community, and what can be done to grow the numbers of PCB designers—and design instructors.
Embedded Design Techniques
Our expert contributors provide the knowledge this month that designers need to be aware of to make intelligent, educated decisions about embedded design. Many design and manufacturing hurdles can trip up designers who are new to this technology.
- Articles
- Columns
Search Console
- Links
- Events
||| MENU - design007 Magazine
Is the Cloud a New Paradigm for Electronic Design?
April 10, 2013 |Estimated reading time: 9 minutes
This article originally appeared in the February 2013 issue of The PCB Design Magazine.
A cloud is typically defined as a set of virtualized resources, most commonly software, platforms, and infrastructure that can be accessed through the Internet. Cloud computing can be private, meaning that the resources are in-house, or they can be public, where a company offers cloud infrastructure as a service. To be useful, clouds tend to be large, providing the illusion of unlimited resources and are often associated with service (as opposed to licenses or capital goods) as a business model. One example is software as a service (SaaS).
A commonly held view is that the cloud is shifting the computing paradigm: Simply put, large-scale commodity computing (millions of servers) virtualized and delivered through the Internet is better than smaller computer centers, regarding cost of ownership, scalability, performance, and utilization. As a result, the cloud is being adopted widely across consumer and enterprise applications. We use cloud-based applications every day with e-mail, search, social media, and gaming. Every time you pick up your smartphone, the apps rely on sourcing data from another location over the Internet, which is cloud-based. Indeed, the cloud plays a role in our daily lives, but how can it be beneficial to electronic design?
At first glance the advantages of the cloud for designing integrated circuits, packages, and boards may seem obvious:
- Numerous parallel jobs running simultaneously.
- On-demand computing resources.
- No need to purchase computers or tools upfront.
- Eliminates the need for dedicated IT staff.
But despite these benefits, electronic design has been a laggard in adoption of the cloud. As the first EDA company to have the cloud as a core component of its DNA, Nimbic often hears the concerns of organizations and their reasons for not adopting the cloud, so let’s take a look at some of these and put them to the test.
1. Security
This is often the initial concern with using a public cloud. However, every day, millions of people entrust the Internet with some of their most important banking and financial data with little concern, so it seems reasonable that electronic design data can be handled in the same manner. Transmitting data through the Internet can generally be as secure as your own network through use of security protocols, encryption, and authentication. However, there are other components to security: auditability and accountability. Once the data has reached a public cloud, there is no real way of auditing or verifying the accountability of the service provider to secure your data because processing data in a “secure” manner is harder than transmitting data securely. We address this issue with dedicated personal nodes (p-nodes) for each user and personalized encryption keys for data transmission, ensuring that you have the most secure environment possible.
2. Large Data Files
A design flow consists of many tools, even several dozen different tools for a complex application, and these tools often read and write large data files. Unless all the tools in the design flow are available in the same cloud, these files need to be transferred through the Internet. Transferring a terabyte through a (fast) 100 megabit/second Internet connection takes about one day and costs approximately $100 (the typical 2012 cloud price). Integrated circuit design generates terabytes of data at some stages, so this starts to get expensive in both time and hardware costs. However, package and PCB designs generally are smaller amounts of data, making it practical to transmit these through the Internet. So while a complete design flow from IC to board may not be practical, our solution focuses on fragmenting what is done in the cloud.
3. Interactive Graphics
Many design applications involve interactive graphics and network latency, making it challenging to run such applications remotely. To ensure instantaneous response of an interactive application, the round-trip time between the client and the cloud providing the computing must be in the tens of milliseconds at most. The theoretical limit for fiber is about 16 ms for 2,000 miles round-trip; in practice, latency is much higher, depending on the number of hops and the distance to an Internet backbone, among other things. Interactive graphics in the cloud is only possible using modern technologies such as WebGL, which load a model of the graphics being displayed locally and execute graphic operations locally. The alternative is to have local installations of the graphical interface and all heavy processing happens in the cloud to utilize the performance and demand elasticity offered in the cloud.
4. Existing In-house IT Resources
Companies involved in electronic design typically have extensive in-house IT resources, including an IT department and policies. They deploy design software, which they either buy or lease long-term, typically 1-3 years. Overcoming this way of doing things is perhaps the largest logistical obstacle to adopting the cloud. It also blurs the real cost of deploying design software in-house, since these resources already exist and the incremental cost of deploying one more application is low and hard to quantify. In cases where no in-house IT resource exists and IT is handled through contracting, the cost/benefit analysis is easier to understand and quantify.
5. Cost
One of the key promises of the cloud is cost reduction. There is little doubt that eventually the cost of using computing power generated by large utility-type data centers with millions of servers will be lower than the cost of deploying your own small- to medium-size computer center. However, at present this is still not the case. For example, consider infrastructure as a service. Today, a typical, fast 8-core machine with 64 Gb of memory costs on the order of $1 per hour in the cloud, which comes to $8,760 per year if used 24 hours a day, 7 days a week. Such a server would cost on the order of $3,000, or $1,000 per year if depreciated in three years. Hosting it (rack, power, AC, Internet) costs about another $1,000 per year, making the hardware-only cost $2,000 per year (purposefully neglecting other costs such as maintenance, personnel to run a data center, etc.).
So, from a hardware point of view, if the machines are used more than 2000/8760 ˜ 20% of the time, buying is cheaper than using the cloud. There are, of course, different calculations, but the point is that cloud computing is not clearly cheaper, unless the average utilization of the resources is low. The utilization of design software licenses that cost hundreds of thousands of dollars is likely to be high, so running it locally is often still more cost-efficient than running it in the cloud.
6. Business Model
The argument for SaaS – or pay-per-use – is similar to the cost argument above. If the utilization of the software is low enough, SaaS is more cost-effective. However, most companies are used to and prefer a more predictable licensing or a 1-3 year subscription model. As already pointed out, expensive software tends to have a high utilization. Users also may receive considerable discounts when they purchase design software in large volume, which offsets the attraction of the SaaS model for the electronic design space.
So, what does work? Which electronic design applications are expected first migrate to the cloud first? For an application to work well in the cloud, it must not exhibit the 6 issues listed above, and must meet the following characteristics:
- Security requirements are not extreme; a PCB or a package is a better design candidate for the cloud than the latest and greatest microprocessor or an ITAR (International Traffic in Arms Regulations) design.
- Transmit small amounts of data compared to the computation time; the data transmission time needs to be almost negligible in comparison to the time the application will run. The data associated with early design stages (e.g., netlists as opposed to layouts) and with packages and PCBs is typically small enough to be transmitted quickly through the Internet.
- No interactive graphic user interface, or the graphics should be separated from the computation and run locally. For example, a schematic editor can be run locally and the data for the simulation of the resulting circuit can be sent to the cloud, which may require extensive compute resources not available locally.
- The organization would likely have limited in-house IT resources or have an IT department that is actively moving applications to the cloud.
- The cost of the computing resources should be small compared to the cost of the software, which is the case for most integrated circuit design solutions. Or the utilization of the application should be low enough on average to benefit from temporarily allocated cloud resources. A third case arises if a design application can leverage parallelism and temporarily run multiple copies of the software across many machines to accelerate a task: 1 computer for n hours costs the same as n computers for 1 hour. Exploiting this elasticity is one of the core value propositions of the cloud. In the case of electronic design, this elasticity can be extremely beneficial in reaching a short time to results for compute tasks that may normally take hours or days, such as electromagnetic simulation. In a compressed engineering schedule, this can mean the difference between making or missing your market window.
- Providers of design software in the cloud need to offer predictable subscription business models to satisfy average loads in addition to pay-per-use for peak loads.
Many applications in the design of electronic systems are starting to move or are ready to move to the cloud. To name a few:
- Electromagnetic simulation.
- Design and manufacturability rule check for PCB and packages (in most cases the Internet is still too slow to transmit chip layouts in a reasonable time).
- Library characterization.
- Variability analysis for cells, analog circuits, packages and boards.
- Functional, logic and circuit simulation.
As the cloud matures, all but a few applications (e.g., applications with extreme security requirements), will move to the cloud. It will naturally be adopted first by the users who benefit most right away, such as small companies with limited or no IT resources. It is also more likely that designers of printed PCBs and packages will be earlier adopters than IC designers due to the design size. Price and convenience are the ultimate differentiators, and large scale “utility” public clouds will at some point be cheaper than any alternatives, offering virtually unlimited resources on demand. Migration of electronic design to the cloud will happen, but the speed of migration will depend on the successes, failures, and costs along the way.
Dr. Raul Camposano has over 25 years of experience in electronics and design technology with careers in industry and academia. He is currently the CEO of Nimbic, a startup developing design technology for electromagnetic simulation in the cloud. Formerly the CTO of Synopsys, Raul was elected Fellow of the IEEE in 1999.
Steve McKinney is an application engineer at Nimbic, where he helps package and PCB design engineers address their high-speed modeling challenges with Nimbic's electromagnetic tools. He has over 10 years of experience working in high-speed signal/power integrity design and simulation tools. Steve has a BSEE and MSEE from North Carolina State University.