Is security the last problem to be solved before EDA moves into the cloud? There are growing pressures to make this happen.
In the first part of this article, EDA’s Clouded Future, the types of application suitable for cloud-based solutions were examined and the cost benefits that could arise for both EDA suppliers and consumers. Security has stood in the way of widespread adoption, but it is a little more complex than just being concerned about a theft of sensitive design data.
Security involves data protect from loss or tampering, licensing, geography, liability, legal and several other concerns. In addition, there are limitations imposed by machine architectures, geography and dataset sizes. Many of the compute architectures are also different from those typically used by EDA software.
Even with all of these concerns, there is a growing chorus from small semiconductor companies that this may be the only way to go in the future, both to control startup costs and to enable increasing amounts of innovation.
What is the cloud?
Most of the time, when we hear “The Cloud” mentioned in a conversation, it brings to mind Amazon Web Services or similar offerings from other companies. But that is just one way in which it can be envisioned. “We think IC design companies will actually look for ways to bring more cloud-based infrastructure into their design environments,” says James (Jim) Ready, chief technology advisor to Cadence. “The ability to access more compute capacity for shorter durations than they would otherwise be able to provide via in-house IT is something that we think they will value.”
And that could involve a variety of compute sources. “Cloud to us means any data center — on premise, or off premise, private or public,” explains Bruce Feeney, vice president of business development and sales at Zentera Systems.
Many in the industry see local and private clouds as the first steps in the journey. Sharing servers with others and moving data in and out of the cloud are non-trivial but solvable speed bumps to adoption, but it will take some time.
“Using a public cloud for 100% of a design project isn’t practical for large semiconductor companies,” explains Ready. “They have massive investments in their compute infrastructure and unique methodologies for how they launch and use EDA technology.”
How is it different?
In order to make compute on demand work, the architecture of the machines is different than the current bare-metal setups used for EDA today. “You have a bunch of preconfigured machines and a bunch of tasks,” says Mojy Chian, CEO of Silicon Cloud. “Load balancing is how best to maximize the distribution of the tasks onto the bare metal machines. Virtual machines are different in that there are no pre-configured machines. The user, based on the application, creates the machine”
A semiconductor company today may have hundreds or thousands of pre-configured machines allocated to different queues. One queue may contain large RAM machines set up for LVS or DRC types of tasks. These may be 128GB or 256GB machines. There are computers in the middle and small machines that may have only 4GB of RAM used for writing RTL. Another stack of machines could be allocated for regression. A company may have 10 or 20 of these queues and depending on the job, it gets sent to one of these queues.
“Virtual machines are built from racks of CPUs and RAM,” Chian continues. “Depending on the task, the user creates a machine. The first question is what kind of machine do you want? I want a machine with 128GB of RAM, 8 cores and 2TB of disk and the system gets built for that task. When done, the machine is closed and the resources go back into the resource pool.”
The security and protection associated with virtual compute resources is different than bare metal and this can, in some cases, help to protect data.
The providers of cloud services and resources believe their systems are more secure than the average enterprise customer. As part of their business model, they undergo independent security audits. Cloud providers know they will be out of business quickly if the customers’ data is not secure. But reality and perception can often be disjointed.
Les Spruiell, applications engineering and security manager Zentera, says you should “look at a semiconductor company. They have their outer firewalls that harden them against the outside world. If you look at the public cloud providers, they have capabilities that are beyond that, that are more intense and have been shown to match government security issues, regulations and standards for security.”
But not everyone is ready to accept that. “The cloud has a long way to go before it is considered to be fundamentally safe,” believes Shiv Sikand, vice president of engineering for IC Manage. “The issue is the relationship between security and sharing. They are orthogonal.”
“Security has prevented the proliferation of cloud computing into semiconductor design,” adds Chian. “It has two components, data protection and tamper protection. Data protection is about making sure that nobody has unauthorized access, that it cannot be downloaded etc. Tampering is to ensure that someone else does not modify my design.”
Some believe security always will be an issue while these systems are built from outdated technology and standards. “It would be nice if security stopped being an issue,” says Sikand, “but unfortunately the current architectures that are in place for sharing files between users were all developed in the 80s. Until we get widespread adoption of content distribution and sharing technologies it is not going to happen.”
OneSpin Solutions has spent a considerable amount of effort ensuring that IP is protected, on top of all of the machine security layers. “So why, you might ask, is there still a question mark about IP security?” asks David Kelf, vice president of marketing for OneSpin. “Well, the engineers have to convince non-technical people that IP can’t leak through this mechanism, and the thought of going to senior management, the legal department, etc., and persuading them that this is OK is just too significant. So the barrier to entry right now is its perception to non-technical people. When this is overcome, as it has in other industries, The Cloud could take off because the benefits are so obvious.
Approaches to security
Security is being tackled in a number of different ways and is specific to virtual machines. A virtual private network (VPN) creates a secure and exclusive tunnel between a user and a server. “These have been around for decades,” says Chian. “A virtual private machine (VPM) means that when a user goes to the cloud, a machine is assigned for that user’s exclusive usage and only that user has access to the machine during the operation.”
Silicon Cloud issues a controlled client that has to be used to access the cloud. “Within it are four levels of security and authentication, and this controls downloading and uploading,” explains Chian. “This means that by definition the user cannot upload or download anything. These are controlled tasks and managed through Silicon Cloud and have to be approved.”
But should we take their word for security? What if it’s a Trojan in disguise? “We have been through numerous penetration testing projects both with customers and external white-hat hacking teams,” explains Zentera’s Spruiell. “We spend a good portion of our time working with customers and they have snapped on things like monitors to watch what we do. They can see that the only communications are between their machines and nothing else.”
Even if the machine is secure, there are additional ways to protect the data. “OneSpin has developed a mechanism that does not transfer design IP off site,” says Kelf. “This is using properties of the way that formal verification operates. Instead, mathematical proof problems are sent to the cloud.” He explains that the data is encrypted, as are the results, and that at no time does the complete IP ever leave the local machine. OneSpin believes that it is impossible to ever reconstitute the design.
Some countries have restrictions on technology exports, which may be inadvertently breached by using the Cloud. That means users have to be very aware of the geographic distribution of machines.
In addition to the security aspect, geography can play a role in performance. “One of the biggest problems we face today is that we have extremely large dataset sizes,” says IC Manage’s Sikand. “You have to be able to make this data globally available. Design groups are smaller and distributed around the globe, wherever you can find the talent and the lowest costs. But they work with these massive datasets and keeping all the team members in sync is a real issue.”
Spruiell adds that “you cannot solve the diameter of the earth. If you want to connect to 12 hours away, you are going to have a slower response time. Most of the time it is about the compute and how much faster that will be performed.”
For the EDA companies, another problem is being able to track usage of their tools and to be able to ensure that full value is paid. “The key thing that attracts people to the Cloud is the pay-per-use business model,” points out Kelf. “The customer pays for the software and the infrastructure (including support of the machines, among other things) to run it as one complete package with total flexibility. It gives them the sense that they can fire up a license, use it, and then give it up without any infrastructure hassle, sales negotiation, etc. This is the essence of the value of the model.”
So how are such services priced? “For EDA, we have pre-negotiated models with the providers,” says Chian. “We have different types of agreements with different EDA providers. Some of them have agreed with university deployment only, and some of them are allowing commercial deployment.”
IBM recently announced a secure cloud service, which provides on-demand access to electronic design tools. The service will be a pay-as-you-go model and opens up internally developed tools used by IBM Microelectronics to other electronic and semiconductor companies. Given that IBM has not been selling tools in the past, it does not have to compete with licensing revenues, but this will be much more difficult for more established EDA companies.
“We priced this model by taking the cost of our traditional license and factoring in the average usage to which one of those licenses was subjected,” explains Kelf. “This enabled us to create a corresponding equitable price. Now we could have imposed an uplift on the price to allow for the flexibility the cloud pay-per-use offers, but we didn’t, mostly as an incentive and to make the solution more competitive.”
According to Synopsys’ Coverity group, “an increasing number of organizations are leveraging big data to realize efficiencies in their business processes and are using analytics to track customer behavior and campaign efficiency. With the volume of data being handled by organizations growing exponentially, big data technology is critical for analysis of data that is too diverse, fast-changing, or voluminous to address with conventional technology.”
“Within EDA, characterization and verification in the cloud would bring significant value to many companies who do not have access to big computing facilities,” says Zhihong Liu, chairman and chief executive officer for ProPlus Design Solutions.
EDA companies are beginning to look at ways in which big-data analytics can be used to improve tools and flows. But so far, few actual design and verification tools are being made available. It is a Catch 22 – until users believe that Cloud-based computing is secure they will not use it, and that means that EDA companies see little incentive to upgrade their tools to be able to offer them in an efficient and effective manner.