History Repeats Itself
The ITRS has been tracking the semiconductor industry for decades now. It is very balanced and adds a global view that represents the five leading chip manufacturing regions through its sponsoring organizations: the European Semiconductor Industry Association (ESIA), the Japan Electronics and Information Technology Industries Association (JEITA), the Korean Semiconductor Industry Association (KSIA), the Taiwan Semiconductor Industry Association (TSIA), and the United States Semiconductor Industry Association (SIA).
Going back to the 2003 ITRS report, the elements that have kept design costs in line in the past become clear. Much like today, it was predicted that without productivity improvements driven by design tools and methodologies, semiconductor design cost would grow to unmanageable levels, effectively ending the progression of the semiconductor roadmap. In reviewing the different possible productivity improvements, it turns out that besides what ITRS calls smart engineers, who became about 50% more productive, IP reuse and tool-supported design methodologies became the industry saviors keeping design costs at bay.
IP reuse as a combination of small block reuse, large block reuse, and very large block reuse resulted in a 580% productivity improvement. Improvements in tool-supported design methodologies, like place and route, IC implementation, functional verification, transaction-level modeling, homogeneous parallel processing, and intelligent test benches, caused an additional 336% productivity improvement.
Productivity Improvements Ahead
As productivity improvements have continued, we have seen roughly an eightfold overall increase in productivity in the industry over the last decade to counteract the effects of rising design cost. So what will save us from here?
I was as shocked as most of the rest of the audience when Gary Smith, founder and chief analyst at Gary Smith EDA, put up a slide during last year’s Design Automation Conference in Anaheim outlining how semiconductor design costs would reach half a billion dollars by the middle of the next decade. Based on ITRS cost data, the chart referred to the 2007 update.
Again, ITRS had suggestions for mitigating the cost explosion. The important news now is that slightly more than half of the productivity improvements have to happen in designing software, not hardware. To be precise, this is the software required to make semiconductor chips work, i.e., drivers, firmware, operating systems, middleware, and application software.
The key approaches leading to about a collective combined 597% hardware development productivity improvement according to the ITRS are homogeneous parallel processing, intelligent test benches, heterogeneous massive parallel processing, usage of transactional memory, system-level design automation, and executable specifications. Yet all of these approaches now concurrently also help improve software development productivity.
Adding to that is the concurrent software compiler, which is a pure software-related approach to automate the parallelization of software across multiple processors. When all of these approaches related to software are combined, a 638% productivity improvement for software development will be achieved. Bottom line: significant innovations have to be made in design automation to address the cost issues. And the big news is that the cost of developing the software on top of the hardware has significantly grown in importance.
Software Development Productivity
Some of the innovation in this domain started about a decade ago and is ready to reach the mainstream now. Ten years ago, several companies were formed to productize the virtualization of embedded hardware for pre-silicon software development. This approach has always had two main objectives: to increase productivity at the hardware/software interface and to parallelize the development of hardware and software by advancing the starting point at which software can be developed to the earliest possible point in time during a project.
In the absence of a pre-existing modeling standard, all early providers of virtual platform technology developed proprietary modeling solutions. In 2006, proprietary offerings for virtual platforms had saturated the market of early adopters, who were willing to sacrifice flexibility, model interoperability, and standards compliance in lieu of getting a working solution.
In early 2007, Synopsys donated key technologies to the OSCI TLM working group. Since then, the application programming interfaces (APIs) for TLM-2.0 have been standardized and ratified. The new TLM-2.0 API standard finally provided the necessary techniques to enable interoperability between models and simulation solutions, making the virtualization of embedded systems hardware for early software development finally ready for mainstream adoption.
The returns on investments for virtual platforms become much more quantifiable, allowing corporations to make decisions to impact the cost of semiconductor development. Given that the results of the early software development still can impact architectural decisions on the hardware side, the likeliness to meet hardware deadlines increases dramatically. As market research firm IBS has analyzed, with 200 engineers developing software, a three-month delay in hardware design has a direct cost impact of $12.5 million and a loss of revenue that can total hundreds of millions of dollars.
Another aspect lies in the ability to replace traditional development kits with virtual equivalents. If 5000 prototypes are needed for a given hardware platform to equip hardware, software, and test engineers, at a direct bill of materials of $1500 per board, the cost can be prohibitive. Switching to virtual platforms for at least part of the intended audience can save up to $5 million per platform.
Innovation in design automation will be key element to overcoming the current crisis. The important news is that for semiconductor design, not only hardware does development productivity need to be improved, so does software development productivity. Some innovations, like virtual platforms for software development, have been created and standardized to improve productivity. Yet more innovations like this will have to be developed to mitigate the daunting semiconductor design cost issues over the next decade.