As chip complexity has increased, design has climbed in abstraction, and it has created a Byzantine network of business models and opportunities. But are too many entities trying to take profit across the design process killing the goose that lays the golden IC?
Way back in the early 1980s it was quite usual for a semiconductor company to design everything itself using internally developed EDA tools and to manufacture the ICs in its own wafer fabrication facilities. The business model was what we would now call "vertically integrated." Over the years, we have seen many aspects of this break down.
As complexity increased, those chip pioneers started needing specialized computers with graphics to run their software. This was outsourced and came back as workstations, and soon it did not make economic sense for many of these chip companies to develop and maintain their own software tools, even though they could make them tailored for their own particular types of chips or processes.
Little did they know at that time how complex the EDA tools would become, so this turned out to be a smart move for most. And then as the cost of chip manufacturing escalated it prompted the concentration of resources in foundries, while others cut back capital expenditure and became "fab-lite" or exited manufacturing completely. Meanwhile fabless startups were being created that outsourced their manufacturing needs to the foundries from day one.
Even the last few integrated device manufacturing companies remaining, such as Intel, have started to make their fabs available to others. And the rate of introduction of manufacturing process technologies -- although not necessarily of geometry reduction -- has been increasing over the past few years, and this requires both additional investment and a shorter period of time to make a return on that investment.
And outsourcing has also happened within the chip design. Companies realized it made no sense to reinvent a particular sub-circuit. Indeed large areas of any chip design provide no differentiation and thus provide little incentive for internal design. By licensing functions and circuit designs from external sources the costs associated with design, verification, and the ability to harden the core are spread across multiple users; and it enables a semiconductor company to focus its engineering resources on the things it does best and where it creates value and differentiation from its competition.
But does that provide enough return on its own? In many cases it is beginning to look like the answer to this is no.
Consider the recent move by Nvidia. (See: Nvidia to license graphics IP to other chip vendors.) It now intends to license its graphics cores as soft IP. In a way, you could say that Qualcomm licenses out its technology as well, although that is more in the way of patents than implementations. Many other companies are attempting to monetize their patents and we have seen not only patent portfolios fetching amazing prices, but increasing numbers of licensing deals. We may hate non-practicing entities (a.k.a. patent trolls), but practicing companies are using these organizations as agents to monetize their IP.
So, what does this complicated web of technical and business connections hold for the future?
I think we will see fewer and fewer integrators, and it is possible that they may not do any unique circuit-level design themselves. The barriers to entry will be dealing with the complexity -- understanding the market requirements and the huge costs and risks associated with developing, making, and shipping a new product. Everyone else will be an IP provider.
Of course, it will not be a straight line from here to there. Consider Apple, which at the moment is attempting to reverse that trend by designing its own slightly custom processor. That company has managed to make huge advances by using the IP of others connected together in creative ways and by careful attention to the specifications of the pieces so that size, weight, battery life, and other aspects of the design become important. And by keeping a large part of the design in-house, from IC design up to selling computers to users, but not making chips, it keeps more of the profit margin.
How do you see semiconductor business models evolving? It sure speaks for a need for more standards and tools related to IP and IP integration.