Special report: EDA startups in the “silicon renaissance” era

Despite the booming semiconductor industry, the number of new EDA startups is smaller than one would expect. Is the EDA industry’s status quo hindering innovation? Is venture capital overlooking EDA? We asked these and other questions to industry veterans (Lucio Lanza, Wally Rhines, Alberto Sangiovanni-Vincentelli) and startuppers (Chouki Aktouf, Keshav Amla)

“EDA, where electronics begins”: this was the title of a video produced in 2002 by the EDA Consortium (now Electronic System Design Alliance). The motto literally holds true today, in the so-called “semiconductor renaissance” era – an expression alluding not only to record high chip revenues and fab investments, but also to the dozens of startups developing new processing architectures for deep learning acceleration. In contrast, the number of new EDA startups seems to be smaller than one would expect. Undoubtedly, the current boom of the semiconductor industry is also benefitting the big EDA vendors and the EDA industry as a whole; just see the latest ESDA figures. Still, the relatively small number of EDA startups may raise some concerns, as these nascent companies are usually considered a key indicator of the liveliness and innovation capabilities of hi-tech industries. To shed some light on this matter, EDACafe talked to industry veterans and startuppers, getting interesting answers.

“Fewer obvious opportunities” in EDA

Let’s start by addressing a specific case: Wally Rhines, an EDA veteran who has been leading Mentor for many years, is now the CEO of Cornami (Campbell, CA), a startup engaged in developing a new computing architecture. Should his move be considered as a clue suggesting that today EDA is not attractive enough for an industry veteran?

“Not at all,” Rhines answered in an email message. “There are always opportunities in an area like EDA, where the technology changes regularly. It’s just not as easy as it once was, when major EDA companies achieved a substantial share of their growth from acquisitions. Potential customers now have very stable design infrastructures so you have to target new capabilities that are not already served.  I’m on advisory boards for two EDA startups, Suitera and Caspia, both of which are developing tools to solve design problems for which there are no commercial solutions yet.” According to Rhines, in other words, low hanging fruits in EDA have already been harvested: “There are not as many unserved niches in the design flow so there are fewer obvious opportunities.”

Alberto Sangiovanni-Vincentelli – professor at Berkeley University and co-founder of Cadence and Synopsys – expressed a similar opinion on this point: “How much can you innovate in EDA? The basic research results were in the late 70s and in the 80s, and then everything that you are seeing in EDA has been improvement on these basic ideas, that were the pillars for Cadence and Synopsys and to some extent Mentor,” he said.

Lucio Lanza – an EDA veteran and CEO of the early stage investment firm Lanza TechVentures (Palo Alto, CA) – highlighted the customer’s demand side: “Historically, the EDA industry has always been required to keep pace with Moore’s Law. That was a major push toward the creation of new EDA startups. Today, that kind of pressure is lower, as the capability of the semiconductor industry is no longer growing exponentially, and the possibility of applying the technology to new unexpected fields has also reduced. Therefore, the existing EDA companies are capable of responding to the demand much better, and there are less opportunities for outsiders to come in with EDA revolutionary innovation.”

Higher entry barriers

Continuing the historical perspective brought to the discussion by industry veterans, it should be reminded that another key evolution process took place in the EDA industry over the past few decades: concentration. The market is currently dominated by a handful of incumbent vendors, each offering complete solutions. Along with the scarcity of obvious innovation opportunities, concentration contributes to a scenario that translates into high entry barriers for startups.

“There are always technological problems to solve, but the point is that you have to provide an entire solution for the companies that do design of semiconductors or systems,” pointed out Sangiovanni-Vincentelli.

Keshav Amla, CEO of San Jose-based EDA startup Avishtech – that develops design and simulation software for PCBs – is well aware of the challenges posed by high entry barriers.

“Even if we make a small little improvement over what the incumbents are doing, the moment they hear about it there wouldn’t be anything to stop them from trying to catch up. It’s just a race where we’re the small guy and they’re the big guy. So, the only way to break through in an industry like this is to do something fundamentally different,” he said.

And that can only be achieved through a significant effort: “Innovation in this field requires a deep understanding of multiple disciplines beyond the software development side,” Amla added. “Now, that’s critical as well, of course, but in order to really innovate in the sector you need to have domain-specific physics knowledge for whatever particular domain you’re working on. It’s really hard to build an environment where people have the right backgrounds for that.”

Selling against incumbent EDA vendors is obviously an additional challenge for a startup: “The differentiation needs to be fairly great in order for companies to look to make the switch,” Amla said.

A similar concept was expressed by Chouki Aktouf, CEO of Defacto (Grenoble, France), a startup providing RTL platforms to enhance integration, verification and signoff of IP cores and SoCs. He also recently co-founded another startup, Innova Advanced Technologies. According to Aktouf, the biggest challenge for an EDA startup is to “convince a first ‘champion’ to evaluate/adopt the new solution and subsequently reach a first success.”

Is EDA not attractive enough for venture capital?

Even though a complete picture should include Chinese investments (see below), that are difficult to evaluate, it looks like EDA startups are getting a much smaller amount of money from venture capital firms in comparison to deep learning acceleration startups. A significant difference, considering the current abundance of global VC investments in many industries (see this CB Insights report). Is EDA not attractive enough for investors?

“Venture funding is not as easy to obtain for EDA as it was in the past,” observed Wally Rhines. “VC firms aren’t paying as much attention as they did in the past, partly due to the attraction of other software opportunities like AI and partly due to the difficulty of growing companies that will command high valuations.”

Sangiovanni-Vincentelli elaborated on the ‘competition’ from neural network startups: “The EDA market is considered to be not as wide as what you could get with neural network processing.” Additionally, “there is nobody who has a dominant position in the market of neural network acceleration. That is obviously a good situation for a new company to be born and also for investors to be attracted. In EDA we do have dominant players, the three major companies: Synopsys, Cadence, and Mentor.”

Being also involved in venture capital initiatives himself, Sangiovanni-Vincentelli then added another element: startups that – due to the high entry barriers – are only able to develop ‘point tools’ are particularly unappealing for venture capital firms.

“The direction that EDA has taken is you tend to buy a solution more than buying tools,” he said. “A solution implies a very large investment, to come up with a toolset and a proven design flow, and a number of people that are needed to support this. So the only entry point for startups is ‘point tools’. And so the only reasonable exit strategy is to sell to Cadence, or Synopsys, or Mentor. Going public is almost impossible, in my view, because to quote a company you have to have a sizeable revenue, tens of millions, and you have to have the presence in the market. And this is very, very difficult to do. I would not put a company with a point tool in IPO myself.”

According to Aktouf, however, there are more reasons why several venture capital firms are overlooking EDA investment opportunities: “Few venture capitalists understand EDA in general and the potential of EDA as a continuous and healthy business. Recently, a discussion around deep learning and the related breakthrough technologies went in favor of other areas (medical, agriculture, etc.) but not EDA,” he maintained.

The role of corporate venture capital

Our brief research did not provide many useful findings about the role of corporate venture capital in supporting the creation of EDA startups. According to Wally Rhines, however, the major EDA players don’t have good reasons for getting involved in such an activity: “Big EDA companies can’t be expected to fund their future competition,” he pointed out.

As for Siemens, that owns former Mentor, a company’s spokesperson stated in an email that the German company’s venture capital firm – called Next47 – “is not actively involved in the EDA space.”

Apart from the major EDA players, other tech companies might be interested in investing on EDA startups. Intel Capital, for example, has recently invested on Motivo. The two companies could not be reached for comments.

Privately funded startups

But doing without venture capital investment can also be a choice, for startups that can count on enough resources to go privately funded. Both Avishtech and Defacto belong to this group, and both mentioned freedom as the biggest benefit of this choice.

“Avishtech is privately funded,” explained Amla, “and the decision largely came down to our desire to remain independent and not be beholden to other financial interests. We knew that the venture side was an option, and it was something that we considered, and we’ve been approached by many VCs along the way, but the basic decision for us really came down to the fact that what we’re looking to do is something fairly complex. That would likely be compromised by having additional financial interest to which every decision has to be explained. One example is that we’ve built out a significant lab here in San Jose. We spent nearly a million dollars on this lab, and the obvious question that would come to most people’s minds – including any potential venture capital investors – would have been: what’s a software company doing with a lab?” As Amla explained, Avishtech needs a lab to test PCB materials, to validate its modeling and simulation solutions.

Aktouf has a similar opinion: “Freedom has no price,” he maintained, explaining that being privately funded is “much more convenient for strategic decisions.” He also pointed out that European startups can count on stronger government support than their U.S. counterparts: “European governments put much more resources to help startups; however, investment opportunities from venture capital are much lower.” Aktouf had positive experiences, as an EDA startupper, in dealing with EU officers or French government officers: “Europe and France are more and more sensitive and aware of the importance of EDA in the semiconductor value chain in general,” he added.

As compelling as these explanations are, they are also a bit unexpected. After all, we are used to consider venture capital firms as a positive driver for the tech world, an important part of the innovation ecosystem. Instead, should they be considered as a second choice that entrepreneurs would gladly avoid?

As a venture capitalist, Lucio Lanza has of course a different point of view: “Let me put it this way,” he said. “If I were an entrepreneur trying to start a company in the advanced electronics world, I would not go to somebody whose only contribution is giving money. Absolutely not. I will go to somebody whose contribution is knowledge of the industry. Somebody who can bring in consultants for you when you need it to help you create this connection until you can afford to make this connection part of your company, which normally doesn’t happen the first year or two. You need to find somebody to not only give you the financing but give you the access to resources that you need, but you cannot afford full time.”

Endorsement from a respected venture capital firm can also help startups to convince customers to adopt a new EDA solution, as Aktouf conceded: “Yes, this is usually very helpful to answer even partially the question ‘if I engage with a startup, what is the level of risk that the company will stay alive in a few years from now? what is the impact to my production design flow?”

Sources of EDA innovation

Higher entry barriers, less VC support: is this scenario hindering EDA innovation? The fact that 50+ billion transistor chips keep hitting the market is the evidence that EDA innovation doesn’t stop – but the contribution provided by different sources has changed over time. With large EDA players now performing more R&D activity on their own, the support for academic research and the R&D role of startups has reduced.

“Big EDA companies now have the R&D budgets that allow them to pursue a wide range of new product and technology opportunities internally,” Wally Rhines pointed out.  “There was a time when startups were almost an R&D arm of these companies, providing the efficiency of testing many approaches in the market and then allowing the big EDA companies to acquire the most successful ones for a premium.  That’s not as necessary as it was in the past.  But it’s still a viable approach and there will still be successful EDA startups.  My concern is that so many of them will start outside the U.S.”

As a professor, Sangiovanni-Vincentelli provided an academic perspective on this point: “It’s certainly true that the number of people that are doing EDA research has decreased over the years, especially in the large universities. Reason being that the number of problems to be fundamentally solved is decreased, and so there is less payoff for a large or powerful research institution to invest in hiring new people in EDA,” he said.

According to Aktouf, even in the current scenario startups can play a key role in EDA innovation: “I think with deep submicron technologies there are too many specific problems to solve. Look at verification or system level design, I think it is not realistic to expect them to be managed by big EDA companies only. A startup here in Europe has a strong link with research labs where technical problems are handled first. The path is shorter from a lab to a startup!”, he maintained.

Criticisms to machine learning in EDA

In terms of EDA innovation, two new factors have come into play over the past few years: machine learning being added to many commercial EDA tools, and hyperscalers like Google contributing to EDA research – a well known example being the paper on deep learning-based chip block placement. On these two points, Sangiovanni-Vincentelli does not share the general excitement.

As for machine learning in general, Sangiovanni-Vincentelli attracted the attention on the cases where this approach implies giving up the possibility of modeling a phenomenon and build a ‘conventional’ algorithm. “If you look at machine learning, you are substituting potentially some algorithmic solution with something which is based on statistical inference. So, you look at what people do or what they have done in the past, and you try to use that information to project onto a new design what you learned from the past. This is not in principle robust as an algorithm. If you know something about the physics of the problem, you use that. You shouldn’t use machine learning, because machine learning does not know what the structure of the problem is. I am really angry at people that are over-playing the machine learning technology. The extreme damage that machine learning has brought to the education domain is that people forget how to think and how to try to structure the problem in such a way that it’s easier to understand what is going on,” he said.

As for the above-mentioned Google research on machine learning-based chip floorplanning, the point raised by Sangiovanni-Vincentelli is the scarcity of publicly available data. “If you look at the experiments that were made, a small number of experiments, people could not really look at what they did because it was an internal Google design. I’m not convinced that these results are in Google better than any other existing EDA companies. The test set that they run their algorithm on is not open. So how do we know what they did and how they did it and all of that?”

EDA startups in China

When trying to assess the total number or birth rate of new EDA startups, one should obviously include the ones that are being created in China. But how significant are the Chinese EDA startups in terms of growth potential and technological innovation?

Wally Rhines highlighted the size of the Chinese effort in EDA and its consequences for the U.S. industry: “Unfortunately, China has picked up the slack and is pouring enormous funding into EDA startups and attracting some of the resources that would otherwise create companies in the U.S.  Valuations of EDA companies are about ten times as high in China as they are in the U.S.,” he pointed out.

According to Sangiovanni-Vincentelli, China is investing on EDA for necessity, and no major technological breakthrough should be expected from its startups: “It’s a matter of life or death for China, because the U.S. government is preventing sales to China of the basic technology. So they got to develop them themselves,” he said. “It would take years for them to develop the technology, but they will copy the technology that has already been used in the United States for sure.” Nonetheless, “We are going to find competitors in China, because once you have the tools you are likely to try to spread your market elsewhere,” he added.

Lucio Lanza did not rule out the possibility that Chinese EDA startups could come up with some major technological breakthroughs: “In China they are moving very, very aggressively in doing their own EDA tools,” he said. “Not necessarily very sophisticated tools today, but the growth might be fast enough and sophisticated enough to include AI, not to be limited to just trying to copy what the EDA leaders are doing but trying to include some new areas.”

Beyond EDA

The new areas Lanza alluded to are, in his view, some solutions that today are not considered part of EDA. “We need to redefine EDA,” he said. “We need to open our minds and not think of EDA as an evolution just of the existing EDA companies. Today the EDA concept can include many new solutions aimed at accelerating and helping the design of new applications targeting different segments of the society and making them more effective. Examples include solutions to ensure memory consistency across a group of processors (CPUs, GPUs, DLAs etc.); solutions to ensure data security in applications where you have a cloud-to-edge communications, such as remote healthcare; solutions to make electronic design more affordable and ‘democratized’, such as open-source design tools or the approach being pursued by many IP vendors in India.”

While chip design continues to be something sharply defined and clearly separated from other applications, it’s interesting to note that also Sangiovanni-Vincentelli mentioned an area different from ‘pure’ chip design as an example of a promising and attractive technology: “If you start thinking about engineering design automation, then you have several companies that have been born, and several researchers are working on it. So for example, CPS, Cyber-Physical Systems, is definitely the evolution of electronic design automation. But it’s about heterogeneous system design more than just chip design,” he said.

These ‘beyond EDA’ perspectives suggest that one of the reasons for the small number of EDA startups could be a different role of chip design within the new, wider and complex landscape of today’s ICT world – ranging from chiplets to datacenters.

Conclusion

The so-called ‘silicon renaissance’ has not fundamentally changed the unique nature of EDA: a vitally important technology for the semiconductor industry, constrained by the relatively small size of its own market. While the EDA industry’s status quo is undoubtedly matching the needs of even gigantic chips, pain points still exist: among them, the amount of resources required by design verification. EDA innovation, therefore, would probably benefit from new ways to lower the entry barriers for talented startuppers. It will be interesting to see if the “chips acts” now being shaped in the U.S. and in Europe will also address these aspects.

This entry was posted on Thursday, April 21st, 2022 at 2:38 pm. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.