The Drivers of Cloud Computing Adoption

The previous post in this series laid out some basic definitions and a premise of where the EDA industry is with respect to the adoption of Cloud Computing and SaaS.

As with any move to a new technology, there are drivers towards adoption and points of inertia slowing down the transition.

Outside of trust, the number one objection that I hear to Cloud Computing is performance. And indeed, this is a key inhibitor to many parts of the EDA flow. But not all.

Cloud Computing Adoption Drivers

Above is the start of a model that could help us understand if there is indeed a path towards Cloud Computing and what the environment would have to look like to get us there.

“Trust” has not been called out separately. In reality, “Trust” is a nebulous combination of all of the above factors and is individual to the specific potential user. The added mystery ingredient being, “who else is using it?”. (Consider the psychology around the use of credit cards on the web in the mid-90’s or the adoption of an IP core.) The other aspect that could be added is “Robustness”. Maybe that driver is significant enough to have its own representation, but for now I’m considering it within “Performance”.

Business Model

This is where it really all starts. If vendors and users were perfectly happy with the current business model within EDA this conversation would never likely arise. However, as has been brought into stark relief recently, the business model could well be the largest challenge that the EDA industry faces today. Above and beyond moving to the next process node.

This can be broken down to two aspects:

1. Revenue generation
2. Cost of sales

Though large EDA vendors have revenue streams to protect, they are constantly pressured to reduce costs. Even with a focus on just the top “10” or “20” semiconductor vendors, there are design centers spread all over the world that need technical support and training. To put an AE on a plane or ship out lap tops for training is expensive. Also, Webex-like tools have limitations that impact the usefulness of online tutorials and training. Especially across time-zones. Users need to have access to the tools themselves - not just be shown them.

The change in business model for a large EDA vendor would be a significant inhibitor to moving to a cloud computing SaaS model. However, as mentioned later, the move beyond EDA to become a “development infrastructure” provider could be an attractive way to look at the problem differently and stimulate growth.

Alternatively, this could be a business model to support infrastructure partners, such as design and verification service companies. By providing their tools as a SaaS (is this “TaaS” – Tools-as-a-Service?) to these partners would encourage proliferation into the service provider’s clients and also lower the overall costs to the partner directly.

Smaller companies have both the cost and the revenue generation challenges. They cannot afford to put in place a large sales team to penetrate a global market. And they’re also open to trying new business models as a means to differentiate and compete.

For medium to small EDA vendors the benefits are clearer. The SaaS model not only enables a smaller company to compete with the large vendors on a new front, it enables them to overcome sales resource limitations. By providing their products via the web instead of directly, a small EDA vendor can now serve the global market without building an expensive sales team.

The bottom line - a requirement to improve ROI, regardless of size.

Cost

I wonder what the percentage utility of the servers in the data centers of large semiconductor vendors is? I would bet that it’s pretty bursty over time and on average fairly low. (Anyone got any data?) If they’ve employed virtualization, utility would have increased significantly, but it’s still a burdensome cost.

Building and maintaining sophisticated server farms is an expensive proposition. Importantly, across the industry it represents gross duplication of investment that does not lend itself to a semiconductor vendors core value proposition. This makes it a prime target for outsourcing to a specialized outfit that can amortize overall costs across multiple customers. Thus reducing the total cost of ownership to individual semiconductor vendors.

An additional benefit to moving EDA flows and their accompanying hardware into the cloud is the conversion of a capital expenditure into an operational one. Acquisition of a service is more streamlined, cheaper and flexible. This would be extremely attractive to start-ups and provide a cost effective method for EDA companies to support them.

Of course, challenges exist for an EDA company to offer their tools via the cloud and to take on additional costs. However, this is an opportunity to differentiate and provide more value to their customers. To move from a tool provider to a “development infrastructure” provider.

One more benefit would be the potential to reduce product development costs. Moving product deployment to a SaaS model would mean that all users are using the latest revision of a tool all at the same time. Upgrade control would be in the hands of the vendor. This would reduce the cost of tool maintenance that would increase the overall efficiency of the industry.

In essence, the movement of costs and the overall opportunity to reduce them provide opportunities and benefits to both EDA and semiconductor vendors alike.

Performance

Performance requirements can drive a tool to or from the cloud depending on the use model and the application itself.

Tool Use Model

Tools that are used continuously with tight iterations may need to stay “local” for a while depending upon their resource requirements. “Bursty” use-models (eg. regression testing) and developments that are stretched geographically would lend themselves to a transition to the cloud.

Some EDA tools are extremely resource intensive. In the past processor, memory and hard-disk requirements would have presented a significant challenge. However, there are infrastructure services available that could accommodate many tools out there.

The challenge that would slow down some applications making the move is bandwidth and latency. For example, interacting with 3D representations of a stacked die would be annoyingly slow when compared to a local application.

However, performing design verification should be straight forward and in our opinion is an application area ripe for transfer to the cloud.

Many engineers are already using VNC to connect to remote servers hosting tools. Many tools are operated in batch mode from scripts and after set up and test are not actively interacted with by the engineer. These types of use-models do lend themselves to transfer to the cloud. Compute intensive applications however, will keep some tools local.

A significant aspect of performance is continuity of service. Service guarantees built into license agreements will be needed to move cloud computing beyond the early adopter phase.

So, depending on the tool, this driver can go either way right now.

Security

This is likely the biggest hurdle to the adoption of cloud computing wholesale. The transition to the cloud will be largely dictated by the degree of actual and perceived exposure and the associated cost-benefit.

Releasing fundamental intellectual property outside of a semiconductor vendor’s “walls” will take the building of tremendous trust and the accompanying legal and technical infrastructure to support it. However, I know of large semiconductor vendors that are already allowing IP onto their supplier’s servers. In particular in the verification domain. It won’t be the whole chip, but it will be individual cores or small “tiles” (sub-systems).

The transition will continue over time and at different rates, but this obstacle is not insurmountable.

So, in summary, there are four main drivers that either accelerate or slow the transition of tools to the cloud. It is because it is not clear-cut and that there are opportunities to compete, reduce costs and create efficiencies that the transition has not only started, but it will continue.

In the next post we will discuss a different way to look at the adoption of cloud computing and SaaS by EDA vendors. A potential cloud computing adoption roadmap for electronic design.

Posted under Features, Xuropa, business, industry, marketing