As much as Intellectual property (IP) is a vague, poorly defined term, it is becoming increasingly important to the semis industry. In this context, IP really refers to pre-made designs for various parts of semis, ranging from Arm’s complex processors to cut-and-past blocs for input/output (I/O) functions common across thousands of designs.
The use of IP saves chip designers from having to re-invent the wheel when they need to add common features to their chips. It is also a handy way to process patent licensing, the other common meaning of the term IP, and is equally important for this function as the design savings. Many in the industry see IP as a necessary evil. Necessary in the sense that it is just that – needed. Evil in a way that makes the most sense to anyone who has ever tried to negotiate one of these contacts. Put simply, there is no easy way to get around using IP.
The trouble is that the IP providers are very concentrated. The top three vendors capture somewhere around 80% of industry revenues. And they have held this position for over 20 years. Any new entrants have to sell against the massive portfolio offerings of the big firms. A start-up may have a highly specialized piece of IP (AI to name one example), but the giants have thousands of libraries, and designers often find it easier to go with the incumbent because they can get everything from a single source, one-stop shopping. There is still value in providing those specialized solutions, but beyond a certain point there is likely only one option available to these start-ups for an exit – sale to the giants.
Another crucial piece of this is that the big IP vendors also sell Electronic Design Automation (EDA) tools – the basic software semis companies use to design and layout their chips. This advantage is two-fold. First, it enhances the bundle – designers have to buy their design tools from the EDA vendors, and while the salesman is presenting those adding the IP blocs is just one more slide in the sales deck. Often the sales process involves effectively buying a bunch of credits from the vendor which start-ups can mix and match (sortof).Probably even more important is that the EDA vendors are tightly coupled into the foundries – which means designers almost always have to go through them to get access to TSMC, Samsung and the rest.
To give a sense of scope we collected the financial information of the vendors below. It is important to remember that this is not exactly a direct comparison. Cadence, Synopsys and Mentor provide a whole range of EDA tools in addition to IP, so Arm is not a perfect comparison. Even among those three there are a whole range of differences, with Mentor providing a very different offering than the other two. Finally, a few notes on these numbers. Revenue figures include sales of both EDA tools and licenses, as we mention above, it is not easy to separate the two. Second, Mentor was acquired by Siemens six years ago, and Siemens only breaks out Mentor’s net income, so the revenue and margin figures here are our best approximation.
One of the fun things about IP is that it is a broad universe, with many more companies encompassing a wide range of offerings. For instance, we did not include Rambus in this list. They specialize in memory, and the memory IP market is very hard to tease out from providers like Samsung and Micron who make most of their revenue from product sales. Nor did we include Ceva in this list. They license IP for connectivity solutions like Bluetooth. They are very small compared to these companies, and their financials have not been so great recently. Our point is that there are dozens more companies that we could include here, but these are the biggest players.
Semis for IP has been a big business for many years. Synposys was founded in 1986 and Cadence in 1988. For many years these companies have languished in the background – important, growing nicely, but fairly hard for investors or outsiders to really track. In recent years, these companies have seen a surge in growth on the back of non-chip companies rolling their own silicon and needing to join the roster of customers. That being said, we think this market is going to become even more important and noteworthy in coming years. The trend towards internally designed silicon remains strong with more companies joining the fold all the time. Moreover, the growing diversity of chips, aka heterogenous compute, is going to make third party IP even more important, as companies seek to focus on areas of differentiation they will rely more on outside IP. Finally, the new workloads like RISC V and AI offer important opportunities for more companies to enter the field. We saw this on full display at the RISC V summit last week, where half the exhibitors were essentially IP companies. The same is true for AI. Synopsys has a library of NPU designs, but the market is moving too fast for them to be able to offer state-of-the-art. For some customers, their solution will be plenty good enough, but many others will need something more specialized, and there are lots of companies springing up to offer this. And finally, no piece on the semis industry in 2023 would be complete without some mention of China. IP providers, especially those in Taiwan, are providing a critical role in helping China’s semis companies climb the value chain, and there are dozens of companies in China opening shop to provide some other form of IP, free from US entanglements.
All of which means we are going to be spending a lot more time looking at IP in the near future.