An recent article in USA Today, Why semiconductor ecosystem is more fragile than ever, triggered a lot of thinking inside fortylines. After all if our customers have trouble to manufacture their integrated circuits (ICs) we have a bigger problem on our hands than lowering the cost of innovation through cloud computing.

The big game

Leading-edge digital processing technology is not a game for the faint heart. The technology prowess necessary to work at the limits of physics and deliver seamlessly bleeding edge processors in high volume is very much cost intensive.

Intel decided to spend $13 billion in 2013 to develop and build future manufacturing technology. That is what is required to be the sole player having FinFET in production Today.

Each process node requires deeper and deeper integration between EDA tools and manufacturing facilities. Samsung is still definitely in the game, manufacturing is a critical piece of their strategy. TSMC will boost Capital Expenditures in 2013 to Record $9 Billion in order to stay in the game.

With total revenue for the entire EDA industry around $5 billion in 2011, you have to wonder why TSMC has not bought Synopsis, Cadence or both yet. Maybe Samsung will. It has a stronger history of managing software, a more international and diverse work-force, and will directly benefit from owning EDA expertise.

Lenovo is going into the chip business to secure access to parts it needs, either directly or through leverage in negotiation with other semiconductors companies. Apple might not have other choice than going into manufacturing itself.

Fabless

So where does that leave our market of vibrant innovative fabless companies?

Fabless companies that would be somewhat immune from the silicon manufacturing chock are either far enough remote from it or have found alternatives.

Intellectual Property suppliers are candidates. Traditionally those companies provide hardware blocks that end-up into SoCs. Recently new breeds of Intellectual Property suppliers have also emerged. These are really software application providers with an accelerator component in silicon. Even though many of those IP companies consider themselves in the hardware business, hardly any of their engineers have seen the inside of a factory or worry about yield and supply chain management. As the software world moves towards multi-threaded distributed design technologies, differentiation between Hardware and Software IP suppliers is becoming arbitrary.

For companies making their revenue actually selling physical ICs, control on the supply chain is too important to the bottom line to be left unattended.

Improving latency is hard, dammed hard. Fortunately a lot of emerging opportunities around Big Data processing require throughput more than latency. This could translate to architecture that leverage older process nodes. Analysis and correlations on large volumes of data also means that interconnect technologies might be more important than shrinking silicon technology for many new designs.

Smaller vendors might benefit from out-of-phase sales cycle. For example, if you are selling processors that go primarily into fridges, your biggest volumes might be expected in spring when factory utilization is low, giving you a chance to get your batch through on time.

There hasn't been a revolution like 3D printing in the IC world yet but reading through Producing Semiconductors from Graphene and A Computer Inside a Cell, Manufacturing of ICs could economically scale to very low volume in the future.

In conclusion, you can still build a vibrant IC business even if you are not Intel or Samsung but you need to deal with the manufacturing constraints head-on first.