Will large EDA companies be relevant in an ESL era?

, Brian Bailey, Chief Scientist at Poseidon Design Systems

Over the past few years, some would argue that the large EDA companies have done more innovation in their business model than they have in their tools. The transition from perpetual licenses to term licenses, all you can eat tool bundles and strategic relationships or joint research programs with customers are just a few examples. One result of this is that it has made the environment tougher for startup companies, but at the same time the large EDA companies are dependent on those very companies for their new ideas, especially when talking about major changes in technology.

The biggest change on the horizon is the migration to a higher level of abstraction, currently called the Electronic System Level (ESL). Each of the large EDA companies is looking to become the next Synopsys, even Synopsys themselves. They are looking for the one dominant tool that will become the underpinning of the new abstraction, and history shows that if they can be the first to market with a successful tool, they should be able to maintain that leadership position.

Unfortunately for them, they have been trying and failing with many attempts at ESL tools. There are a number of reasons why they have been failing, some of which are discussed below:

1. ESL is not a simple step. The migration from gate to RTL was fairly well constrained. Engineers had been designing state machines and in effect performing an RTL design style even though they did not have the languages to directly support it. This does not imply that it was an obvious or easy transition. Things always look easier when looking back into the past. The migration to ESL is not a well defined step, but rather a combination of many changes in modeling methodology, the integration of multiple disciplines, such as hardware and software, algorithms and implementation, the combination of describing functionality at new levels of abstraction as well as the definition of architectures. ESL is not a single thing and thus it is unlikely that there will be a single dominant tool.

2. ESL is not about design. The advances made in reuse both at the component level and at the platform level have relieved many of the pressures on the design flow, but have increased the pressures on managing the total system complexity and perhaps more importantly on the verification flow. Verification is now stated as consuming about 70% of the design effort and this will continue to increase unless greater verification reuse can be achieved. Amdahl's law clearly shows the futility of concentrating on something that is not a significant portion of the total time.

3. Migration of functionality. Most complex systems today contain at least one processor and there is clear evidence that multi-processors systems will quickly become the norm. The important aspect of this is that a greater percentage of functionality is moving into software. The software industry is very different from the hardware industry, with different skill sets and different business models. Software engineers do not pay for tools in the same way that hardware engineers find acceptable. This requires a completely different way in which tools must be developed and packaged.

4. IP holds the greater value. IP is becoming more valuable than the tools necessary to stitch them together. Many platforms are based on a number of standard interfaces that are used to connect the blocks together. While the implementation of these interfaces is an error prone step and ripe for automation, it is not necessarily of high value when compared to the value of the blocks themselves. Gartner has estimated that the total ESL space may be a $2B market, but I suspect that a lot of this will be given to the providers of the IP blocks or integrated platforms and not the tool vendors. There will still be some blocks that need to be taken from an abstract description into a dedicated hardware solution and thus the need for hardware synthesis will not go away, but its usage will be concentrated on a small portion of the design that differentiates that product from a commodity.

5. The Open Source movement. SystemC was the first new language to gain traction in the ESL space, and is now being used by a large number of companies doing architectural exploration and for distributing models. This is the first time that free open source software has been a success in the EDA industry, and will put pressure on the industry, especially the commodity tools such as simulators. The EDA industry is fighting back with the development and backing of SystemVerilog. While SystemVerilog has more capabilities for verification, it is primarily an extension to the RTL methodology already in place and has few capabilities not found in SystemC. It is not clear if either of these languages are what the industry requires for ESL.

6. Addressing the wrong problem. It is in the best interests of the existing EDA leaders to stop the migration to ESL as this disruption may jeopardize their current revenue streams. As stated in the previous section, this means they have a vested interest in pushing incremental improvements in the RTL tool flow. This will work for a while, but eventually the methodology needs to change.

7. Growing the market. With every change in abstraction in the past, there has been growth in the total market and in the number of users. This has been achieved by giving up something in terms of silicon, in return for greater productivity and a simplification in the use of the tools. There were ten times the number of gate designers compared to polygon pushers, and ten times the number of RTL designers compared to gate level designers. If we are to grow the user based by another factor of ten, is not by attacking the high end problems, it is by finding solutions for the masses.

Large systems companies used to develop their own tools. This became uneconomic and was outsourced. This dissaggregation was the birth of the EDA industry. In return for improved tool value, those customers had to give up some control over the feature sets of the tools, but the existence of multiple EDA companies created sufficient competition that most of their demands were met. During this time, the large EDA companies became marketing powerhouses, snapping up startup companies when their technology had been proven and integrating them into their existing flows. This was a win-win situation for everyone. For the startup it saved them the high costs of creating worldwide marketing and sales organizations. For the EDA companies it created a cost effective research organization that had better success rates than their own internal developments. For the industry it meant that successful tools would be supported and maintained by large stable companies, making their adoption less risky. However, this only works well for incremental types of improvement. When the changes are larger, the EDA marketing and sales teams are suddenly out of their depth, or talking to the wrong customers.

Recently, dissatisfaction has been growing in the users of EDA tools and there are examples of these companies coming together, such as Freescale, Philips, ST, Infineon, to solve joint problems. This is in part because they see the value of their offering being in the IP and silicon, but at the same time the dissemination of this IP is being hampered by the lack of proper tools support. I thus see a new possible business model emerging, in which startup companies will produce customized software for the IP industry enabling them to deliver 'solutions' based on those platforms. The startup companies will not sell their software directly to the end users, and will be able to minimize development costs by sharing development costs between their customers. It is the 'solutions' that will hold the value and allow the market to expand, not the tools or IP alone. This means that startup companies should be looking at creating tight alliances with one or two IP or systems companies, and it also means that young companies must understand the dynamics of the market which is being pulled by both business and technical issues these days.