Two big international semiconductor companies recently announced that they have succeeded in building plants capable of producing highest-performance chips in volume. TSMC in Taiwan and Samsung in South Korea have now joined Intel, the long-term world leader in manufacturing technology, in a very exclusive group capable of producing such sophisticated products.
The emergence of TSMC is particularly important because it is the only purely contract manufacturer serving many fabless and system customers worldwide. While once only Intel could manufacture highest-performance products economically, now the most sophisticated production capabilities are available to any company prepared to design products aimed at these high-end markets. This has profound implications, because major system companies such Apple and Cisco can design proprietary chips bypassing traditional semiconductor product vendors and relying on TSMC for production.
Why so few high-end production facilities in a US$470 billion semiconductor industry? It is the result of increasingly costly production technology and corresponding huge capital investments that can only be justified by a very few very big companies.
Integrated-circuit chips for computing applications consist of a large number of interconnected switching devices (transistors). Since the 1960s the computing performance of such chips has increased along with unit cost reductions following what became known at Moore’s law. Just for reference: In 1960 a single silicon transistor cost $25. Today, this price buys a chip with more than a billion transistors on a product the size of a thumbnail.
Basically, the size of the transistors has continued to shrink such that the computing capacity in a given chip area doubled about every two years. Since the computing power increases with the device density, chips have become ever more powerful and cheaper for a given computing capacity while the power dissipated in the computing process decreased because of the shrinking transistor size.
This progress made it possible to provide within a smartphone the computing capacity of what used to be mainframe computers and to provide extraordinary capabilities to systems ranging from clocks to complex communications networks and systems using artificial-intelligence software. Computing chips of various levels of complexity (and cost) are practically everywhere.
This unprecedented technology improvement has been the major driver of the digital age. But this progress has not come cheaply
This unprecedented technology improvement has been the major driver of the digital age. But this progress has not come cheaply. Many billions of dollars of investment have gone into research and development and equipment design, mostly in the US where the major inventions were made.to sustain increasingly complex processes. Semiconductor plants became increasingly automated and costly to build and maintain. In fact, the degree of robotics is such that few human operators are involved directly in the production process. That is the job of robots driven by programmed computers.
It is not surprising that the cost of building such state-of-the-art plants with commercial-scale capacity has also increased dramatically. For example, a single optical pattern generator using UV lasers now costs $100 million, versus $10 million for a much simpler machine 20 years ago when a state-of-the-art plant would cost about $400 million. Today a similar-capacity plant costs $10 billion to $15 billion and will be depreciated over a period of five years because of the pace of new equipment innovation.
What this means is that over time with better technology the production economics became increasingly scale-dependent for the leading-edge technology plants. With highly automated and costly equipment, small plants were not economical. State-of-the-art production facilities could only be justified by production volumes that few semiconductor companies needed.
The answer was contract manufacturing. Companies emerged that only operated as contractors for others that only designed chips. This was the opportunity that was exploited so well by TSMC, which was established in 1987 strictly as a production contractor offering state-of-the-art capacity to all comers. It is noteworthy that Intel did not offer such contract production facilities, leaving the field to TSMC and a few others that offered similar services.
This is a very difficult business model, and of these contractor production companies, TSMC stands out as the most successful by far. Its revenues grew to $33 billion, always profitably, as it gradually became the service provider of choice by maintaining leading-edge production facilities. More recently, joining TSMS as a high-end contract manufacturer is Samsung. The company offers contract production services in addition to selling its own products – sometimes in competition with its contract production customers.
What are the industrial implications of this production consolidation? Many fabless companies in the US and overseas designing and marketing high-end chips depend on TSMC for production. Hence the competitive marketplace for chip vendors is based on design skills, not manufacturing prowess as was once the case. The second major impact on the electronics industry is that big electronic-systems vendors bypass the traditional chip companies and design their own proprietary chips using TSMC or Samsung as production partners. And this trend is increasing, as it allows cost reductions by the big-system companies not available to their smaller-system competitors lacking such design skills and having to pay the margins of traditional chip vendors..
Finally, there is an impact on chips designed for high-end defense products. Such chips frequently need secure production facilities to protect the design and have to be located in the US. Such facilities in the US will have to emerge to maintain defense-product leadership.
Henry Kressel is a technologist, inventor and long-term Warburg Pincus private equity investor.