$12bn won’t even get you in the door, says finance account manager Paul O’Flynn
In contrast to the fashion industry, the semiconductor industry cannot exist without process technology. For example in the fashion industry, it could be argued that the role of process technology is to complement the supply chain in a way that enables faster production and shorter time to market. However process technology is not critical to produce output in the sense that it is still possible to manufacture a sweater based on human precision and the use of manual labour.
By comparison, in the semiconductor industry there is simply no other way to manufacture an Integrated Circuit (IC) without the use of robots, high-precision engineering, and the application of volatile materials science.
To a large degree, process technology sets the pace of the whole semiconductor industry and dictates new product readiness for High Volume Manufacturing (HVM). It does so because of the control it has on the holy trinity of parameters which drives the computer chip industry:
- How far can the performance envelope can be pushed (faster)
- By how much can a single transistor be reduced (smaller)
- Can it be produced in high volume at low-costs (cheaper)
Keith Bontrager is well known in the bicycle manufacturing industry for saying, “strong, light, cheap, pick two”, when he summarized the limiting factors that dictate what a bike manufacturer could produce (SPOKEYDOKEYBLOG, 2014). Unlike the cycling industries, there is no room for compromise on any of the primary parameters in the semiconductor industry.
Stiff competition between industry giants such as Intel, Samsung, TSMC, and Global Foundries leaves no room for compromised results when churning out state-of-the-art technology: it is necessary just to survive as a manufacturer.
The semiconductor industry can be viewed as a ‘node’ based industry (e.g. 14nm, 10nm, 7nm, 5nm, etc.) where each node represents the release of a range of new products with significant improvements on each of the holy trinity (speed, size, and cost). The average cycle time to move from one node to the next is about 2 years and the number of transistors on an IC is considered the most important measure (doubles roughly every two years) (G. Moore, 1964). This is referred to as Moore’s Law and should be considered an ‘economic observation’ rather than an engineering law.
Moore’s Law was defined by Gordon Moore in 1965 and set the pace for the semiconductor industry ever since. Interestingly, in recent years Moore’s Law has become less of a focus in the industry and has since been overshadowed by the limiting factor of material-process technology. Moore’s Law still remains a strong driving force for setting the roadmap for silicon manufacturing but acts more so as a forecast for a theoretical objective. Those wishing to compete in the industry must not focus on debating the relevance of Moore’s Law but instead focus on solving the real problem: process manufacturing.
It is important to conceptualise the idea of a node in order to fully comprehend the true nature of the problem facing the industry. For example on a 14nm device (fourteen nanometer) i7 product from Intel which you can find in an everyday laptop will be about the size of a thumbnail and pack about 1bn transistors. Similarly Apple’s latest 10nm iPad chips produced by TSMC pack a whopping 3.3bn transistors. Transistors are responsible for sending 1’s and 0’s around the computer to make all of the magic happen (the more transistors you have, the better/faster your laptop or phone will operate).
In the most simple of terms, 14nm refers to the smallest necessary component on the chip. Note that this is a somewhat ambiguous definition and is also subject to debate since it is a crucial differentiator for company’s trying to maintain an industry leadership position. It is therefore subject to artistic licence and a company may choose to define it based on their own rationale.
The stage of the manufacturing process which is responsible for producing the 14nm components is referred to as lithography and represents the most challenging, volatile, and costly stage of the chip manufacturing process.
Similar to photography, lithography uses light to create patterns on a printed circuit board but unlike photography, lithography requires ultra-high precision which is achieved by using high powered ultra violet lasers in a vacuum chamber (to ensure absolutely no foreign particles are present, causing contamination).
Today, Lithography drives the highest material-process manufacturing cost due to its ever increasing complexity. The existing process manufacturing tools used in the lithography stage can operate only at 193nm, not 14nm. This means additional passes through the tool (up to 50) and investment in additional ‘reticles’ are needed to get the desired end result.
Simply put, reticles are a ‘stencil’ which guide the laser during the patterning. The average cost of a 28nm reticle is $1.6m and rising (AnySilicon, 2016). Dozens of reticles are needed for a product range and drive millions of dollars in variable spending each quarter.
The problem with lithography is that it continues to get even more difficult to execute at smaller dimensions. As the industry moves to 10nm and beyond, it won’t be long until we get the sub-atomic level (beyond 1nm) where the most common process technologies are rendered useless. Today, Extreme Ultraviolet Lithography (EUVL) is the most likely solution (S. k. Moore, 2018)and is being used by companies such as TMSC. EUVL lasers operate at 13.5nm and achieve the desired effects in one pass. However this does not come without extraordinary action from three of the leading manufacturers.
While the cost of a EUVL tool today is roughly $120m (which in itself is a huge jump in capital expenses and also has a far greater footprint than its predecessor) it doesn’t capture the true price of continuity.
EUVL tools are developed and manufactured primarily by ASML, a Dutch company leading this niche industry. In order to accelerate the development of the mission-critical EUVL tool, ASML partnered in 2012 with Intel, Samsung, and TSMC who each contributed approximately $1.38m to support the research and development efforts (ASML, 2012).
The cost of material-process manufacturing tools such as EUVL is constantly increasing with each new generation (or node) of product being developed. As a result, the overall cost to build or retro-fit a new factory has also continued to rise over the years. A 14nm fabrication plant costs in the region of $7bn (-1 generation) and the latest 10nm fabrication plants costing upward of $12bn (LAPEDUS, 2015b). This continues to rise as the industry moves in line with Moore’s Law, towards 7nm, 5nm, 3nm, 1nm, and wherever follows next (LAPEDUS, 2015a).
Material-processing technology controls the cadence of new product releases, what level of performance can be achieved, how small devices will be, and how much they will cost. Without advancements in process technologies it is arguably impossible for the industry to exist and it continues to be the largest contributor to unit cost. It creates one of the biggest barriers to entry for the semiconductor industry and is central to supply chain success for company’s looking to gain a competitive advantage in the market.
AnySilicon. (2016). Semiconductor Wafer Mask Costs. Retrieved from http://anysilicon.com/semiconductor-wafer-mask-costs/
ASML. (2012). Customer Co-Investment Program. Retrieved from https://www.asml.com/annual-report-2012/management-board-report/customer-coinvestment-program/en/s48039?dfp_fragment=ifrs_cip
LAPEDUS, M. (2015a). 10nm Fab Challenges. Retrieved from http://semiengineering.com/10nm-fab-challenges/
LAPEDUS, M. (2015b). 10nm Fab Watch. Retrieved from https://semiengineering.com/10nm-fab-watch/
Moore, G. (1964). The Future of Integrated Electronics. Fairchild Semiconductor Internal Publication.
Moore, S. k. (2018). EUV Lithography Finally Ready for Chip Manufacturing.
Slack, N., Brandon-Jones, A., & Johnston, R. (2016). Operations Management (book). Operations Management. https://doi.org/9780132342711
SPOKEYDOKEYBLOG. (2014). Strong, Light, Cheap, pick two. Retrieved from https://spokeydokeyblog.com/2014/01/26/strong-light-cheap-pick-two/