Introduction: After doing a couple of IP deals (ARC and getting ARM Ltd going), in most cases, I'm not sure why anyone would ever do one again! The idea is sound, where a bunch of people with deep domain expertise set out to create something like software, but in much more of a hard form - namely RTL for chips. The idea is to take very complex behavior and reduce it to a HDL description, test it and sell it to a number of customers for a bunch of money. The problem is that only a few companies have been successful doing it!
Industry Problems: There are a number of problems common to this industry, that are shown below:
The first problem is that most of the functions are (almost 100%) digital only and standards-driven. That creates the issue that any great team of coders in another country can create designs that are just as good. This also tends to reduce the value of the IP. Lots of supply, some demand.
The second problem is that most customers haven't figured out yet that paying for IP, either up-front, or in a combination of up-front and with royalties, is in their interest. They tend to treat IP vendors as one-shot purchasing arrangements - not one where they contemplate a long term relationship with future designs. This tends to give the buyers much more power in the selling equation.
The third problem - is that most Mom-and-Pop semiconductor IP vendors never bother to actually put their designs into silicon. They might go through the entire design flow one time, so they can prove they can hit certain timing targets, but there is something reassuring about knowing that a design has been used successfully in silicon a number of times before your team actually starts to try to use it themselves. This tends to reduce the value of the IP.
The last problem is that creating IP for a narrow range of applications is difficult, but creating IP so that it is configurable to be used in a very wide range of applications is extremely (>>2x) difficult. The other issue that goes with this is that most interesting IP is a combination of software and gates. The software must be automatically changed to support the changes in the gates. This overwhelms most smaller IP vendors (eg, at ARC the compiler would automatically be changed to support changes to the instruction set architecture).
Industry Characteristics: There are a number of characteristics in common for IP companies. A short summary of them follows:
While it's not necessarily a problem, an important characteristic of this industry is that it has exhibited a Winner-take-All characteristic where the number one company in a particular market segment gets the vast majority of revenues and the number two and three players are left far behind picking up the scraps of business that the number one couldn't service for some reason. A particular example comes to mind with ARM Ltd vs MIPS and ARC Cores. ARM is clearly in the number one market position with MIPS and ARC trailing way behind. Tensilica is also in this race, but with revenues slightly less than ARC's.
Another important characteristic of this business is that before there is a major customer with a highly visible design-win, the value of a license is very difficult to defend, usually boiling down to an argument over how much effort it takes the customer to make vs. buy the same technology and the risks associated with the make side of the equation. In the early days of IP at VLSI Technology (we had PC building block chips called MegaCells), some IP blocks were more difficult to use due to design styles, documentation or other implementation issues than they were worth. In multiple cases, we brought in IP from some other group only to throw it out and start from scratch.
Another issue facing IP vendors is the Horizontal vs. Vertical decision. On one hand, they can create a very large market by making something that is useful in a very broad range of markets. This takes lots of work and rarely meets all of the specific needs of customers. The other approach is to make something very Vertical - or focused on a single market application. This is usually done with lots of domain-specific knowledge in the design team. This creates the dilemma that to add a new vertical segment requires adding people from that area, some of which are very difficult to find or recruit.
Opportunities: With these difficulties - there are two bright shining spots in the IP business: multi-domain patent protected IP (e.g., Qualcomm & Rambus) and physical IP (e.g., Rambus).
The first category consists of problems that require deep insight in two or more domains, such as in the case of Qualcomm, spread-spectrum RF, cell phone radio, modem and baseband knowledge. Each of these areas requires specialized understanding of an area that is not commonly known.
The latter category of physical IP is more common, but still is a lot of work. For each new process variation from TSMC or others, the IP company has to alter their design to still meet all of the critical specs. It really is a treadmill - but one where customers are willing to pay for the work.
Conclusions: With all of this said - the IP business is inevitable, but the industry has a lot of growing up to do before it is a (relatively) low-risk investment category.