I have put out some posts about generic business models and startups. However, if you want some advice that is more EDA-specific, then at DAC a couple of years ago (before I was back at Cadence) Jim Hogan and I did a presentation and Q&A on EDA startups that we called "Show Me the Money." EDA Startups EDA startups are not the same as a typical internet startup. Usually they require funding, either by a rich founder who has money from a previous exit, or by an angel investor (like...Jim Hogan). An internet startup can take kids straight out of college who know how to program. An EDA startup requires more seasoned engineers who have built up expertise in some technical domain. Occasionally a company essentially productizes someone's PhD thesis, but that's about as early as it gets. It is still a lot easier than it used to be. Workstations are not that expensive (hint: go to the gaming shops and buy overclocked gaming machines) and Amazon can provide unlimited compute power for regressions, test cases, and so forth. In fact, these days, any EDA solution probably should at least be cloud-ready. Nobody quite knows if EDA will truly move into the cloud, but it would be foolish to bet against it. There are a number of reasons why EDA startups are less common than they used to be: one on the investment side, and one on the technology side. On the investment side, people are just not investing in EDA and IP startups, at least in the US. They are in China, but not here. One reason for that is that the exits are not there. When I joined Cadence in 1999 on the acquisition of Ambit, it was a $300M deal. The exact amount is vague since it involved Cadence providing tools to LSI Logic since they had a veto over the deal (normal VC terms are that each series has to approve an acquisition, to stop the later investors selling out the earlier or vice versa, but that makes less sense with a strategic investment). Deals like that don't happen any more, and so the attractiveness of an EDA startup is diminished. At the same time, every VC wants to invest in the next Instagram or WhatsApp, sold for billions very early. Jim and Lucio Lanza are among the only people looking at EDA, IP, and fabless semiconductor investments. The technical reason is that startups are very good at creating point solutions, but increasingly the challenges require a full tool flow as a prerequisite. You can't be the point tool for multiple-patterning, or FinFET, or gate-all-around FETs. These affect the whole flow, at a minimum: DRC, custom layout, extraction, P&R. There is still some scope for point solutions in some areas, but they tend to be off the mainstream: automating variability characterization of memories, or specialized tools for flat-panel display design. Expectation of Each Phase Jim put together a couple of tables showing the expectations of each phase of investment. The phases are: Seed: System architecture, marketing specification, proof of concept Round A (validation): Build system infrastructure (database, UI), validation with alpha customers or technology partners Round B ( production): Stability and productization, beta customer program, focus on performance Round C (scalability): Focus on supporting early customers and proliferation in at least some accounts IPO (this never happens these days): Growth >20%, EPS 20%, follow-on products, global expansion The table below shows a rough timeline for a successful company, including size in terms of revenue, number of people in R&D, number of people in sales and marketing, and key hires at that stage. Return Expectations One thing that it is always good to do with any startup is to look at it from the investors point of view. There are some vanity projects where investors might provide money for something other than just the return (restaurants, Broadway shows, for example). But generally people invest in your company to make a return on their investment, the greater the risk the greater the return needs to be to justify the investment. There are three types of exit: Asset sale, also called a technology sale, or acquire Company sale IPO After Sarbanes-Oxley, the scale of company required to be able to afford the accounting necessary to go public went up to $50M or more. It is very hard for an EDA company to get there, since it almost certainly requires multiple products, significant penetration into some major semiconductor companies and so on. Almost certainly a company like that would get acquired by a company like Cadence unless the company can be successful enough fast enough that the price rises out of reach very fast. The recent history with companies like Apache were that they would file an S-1 (to go public) mainly as a way of putting the company in play, and then getting acquired. The table below shows the expected ROI at exit for the investors. In all cases, the asset sale is breakeven to negative. If the company is sold as a going concern, then the returns get higher, but no EDA company is likely to get Instagram type returns. Successful Attributes of a Software/IP Startup Luck is always a major part of the success of any company. I mentioned Ambit earlier but I fully believe that had we come out with a product earlier then it would have failed, since people weren't looking for an alternative. We had a much better solution than Cadence's internal synthesis tool, Synergy, which they shut down. I focused engineering on a major evaluation with Philips Semiconductors (now NXP and soon to be Qualcomm). We won, and they standardized on Ambit. At the time, Philips was Cadence's largest account. Unlike almost everyone else in the industry, they standardized on Ambit as their synthesis tool. After that, the only question was what price Cadence would pay to acquire us. In the current environment, the first challenge is to raise the minimum amount of cash you can get away with...and, as with all startups, never run out of cash. Just as everyone dies from lack of oxygen, whatever the root cause, startups die because they run out of cash, whether it is due to not delivering the product, sales too low, sales too late, a VC recession, or whatever. The founding team needs to be adaptable. It is always better if they have had prior success, so they know what it takes. It is like running your second marathon or learning your third language, it helps to have gone through it all before. It needs staying power since it will take at least four and perhaps six or seven years. Another aspect is having an adaptable business plan, since it always changes. It is important not to use up much cash to get to technology and market validation, in case a pivot is required. At some level, beggars can't be choosers, and any investment is better than no investment. But ideally you want an investment group that has marketing and operational expertise, not just money. My experience has been that the people who have run EDA companies make much better board members than people who have only been involved in very different industries. Obviously, the more companies that the investor can bring to the table as customers (perhaps because they invested in them, too), the better. EDA is not a single round and out, so investors need to have staying power. Segments Any startup needs to have a discrete and defendable niche. In EDA, that means some technology that is hard for the large EDA companies to reproduce internally fast enough. It also needs to be a fast growing segment or the business will grow to slowly, and obviously it needs to be poorly served since customers would rather buy from established players if they can. Timing is tough. More companies fail by being too early to market than too late. Old technology/methodology often lasts one more process generation than you expect and that is two or three more years of cash. There are three kinds of attributes of technology in EDA: Optimization: This has the highest value since it gets baked into the product. If you produce lower power designs that shows in battery life, for example. Productivity: This is tough in some areas, but others like verification are more measurable. The ROI then is easy to justify. Cheaper: Not good in general. Each SoC is too important to not get the best tools, just as "cut-rate heart surgeon" isn't likely to have a line out of the door. Where are attractive areas? Analysis of something that was always second or third order and will become important. Technology always first shows up in tools to find the problems, at least one generation before tools learn how not to create that problem in the first place (signal integrity, DFM hot spots, etc.). IP, since chips are not so much designed as assembled. But standards-based IP can be hard (any IP that meets the standard is almost as good as any other IP). Non-standards based IP requires deep expertise (say in automotive vision or RF). You will not be the next Arm. The ends of the EDA flow, going up to tie IC design into the software load, or down to tie IC design into manufacturing (yield improvement for example). Cash Did I say it already? Don't run out of cash.Image may be NSFW.
Clik here to view.![]()
Clik here to view.