Overview of the semiconductor industry

I’m trying to expand my circle of competence to include the semiconductor industry.  Here’s how I see the industry so far.

There are different types of semiconductor companies:

  1. Design companies.
  2. Manufacturers.
  3. Vertically integrated design and manufacturing companies.  e.g. Intel, IBM, Samsung
  4. Companies that make hardware for semiconductor manufacturing (e.g. AMAT) or the software for semiconductor manufacturing (any of the EDA software companies).  I do not understand these companies.

Design companies

As far as stock performance goes, the companies that are #1 in market share or the ones that will become #1 tend to do the best.  Being #1 is a huge profitability advantage as R&D costs are scaled over a greater number of units.

Some types of hardware are incredibly hard for competitors to break into.  Xilinx and Altera have been the top companies in their FPGA markets (usually 90% market share combined) and held off companies such as Intel.  Because it is very expensive to design FPGAs, the quality of the manufacturer’s design software is very important.  In general, good software tends to be difficult to displace.  Similarly, the discrete graphics (a standalone graphics chip that only does graphics) market is difficult to break into as graphics involves very complicated hardware and software drivers.  I think of design companies as enjoying an intellectual property moat.  However, this type of moat isn’t that durable as another company can develop even better intellectual property.  While this occurs rarely (the #1 and #2 companies in tech usually kill off a very large number of competitors to reach their position), it does happen.

Changes in the technology industry tends to happen very fast as technology scales very easily.  There are some industries where change occurs more slowly since customers are entrenched in their current platform and switching costs are high.  Mainframe computers (e.g. IBM) and networking (e.g. Cisco) may be like this.

Manufacturers

Two key factors in semiconductor manufacturing are (1) process size and (2) process technology.  A smaller process size allows more transistors to be crammed onto a piece of silicon.  Suppose that the total number of transistors stay the same.  A smaller process size decreases costs since more chips can be created out of the same silicon wafer.  It also decreases power consumption since all the circuits are packed more densely (the shorter wires require less power).

There can also be improvements in the actual transistors themselves and/or the materials used in manufacturing.  Two major innovations are high-k dielectric materials / metal gates and finFET transistors (Intel calls their variation Tri-Gate).  These improvements lower power consumption and/or improve speed (there are always tradeoffs that can be made between power and speed).  Manufacturers may also have variations on their manufacturing process that can make tradeoffs between power and speed.

The size of the silicon wafers is another factor.  Larger wafers are more challenging from a technology standpoint and require higher capital expenditures.  However, they lower per unit costs.

Historically, semiconductor manufacturing hasn’t always been a great business for all the companies involved.  There is a technology treadmill as companies have to continually invest in the latest technology to bring their costs down.  Without constant reinvestment they will be left behind.  There are economies of scale that entice companies into expanding capacity and give the #1 market share leader an advantage.  Sometimes the semiconductor manufacturing industry will experience intense price competition and times of oversupply.

Vertically integrated companies

The trend of the semiconductor industry has been moving away from vertical integration.  In the past when mainframes dominated the computing industry, all companies designed and made their own silicon.  Nowadays, many semiconductor design companies are “fabless” and outsource their manufacturing to contract fab companies like TSMC.  The capital costs needed to create a new cutting-edge manufacturing plant has continually gone up and is somewhere around $5-$10B currently.  Many vertically integrated companies are selling off their manufacturing units or entering into alliances since they do not have the capital needed to keep up with the technology treadmill.

There may or may not be small advantages or disadvantages to vertical integration.  There are likely small efficiencies from vertical integration.  For example, Intel carefully schedules its design projects and manufacturing investments so that all of its fabs are at high utilization rates.  In practice, Intel currently has some excess capacity at its cutting-edge 22nm fabs (which produce CPUs cheaper and better than its older fabs).  Meanwhile, its competitors’ best fabs are at 28nm (without finFET transistors) and their customers desperately want higher capacity at the 28nm fabs.

In terms of the big picture, I believe that vertical integration may be mostly irrelevant.  It probably isn’t a huge advantage or disadvantage.

Other semiconductor manufacturers

Some companies such as Texas Instruments make analog semiconductors.  This business has different economics than digital semiconductors as it isn’t about constantly doubling the number of transistors on the chip.

Industry trends

Moore’s Law

In the past, performance has roughly doubled every 18-24 months.  The exponential rise in performance is slowing down for many markets as we are running into the laws of physics.  Eventually the dramatic improvements in semiconductors will slow down to a trickle as we butt against fundamental laws of physics.  The heat problem has already forced CPU manufacturers to add more cores instead of continuing to improve single-threaded performance at rates similar to the past.  As transistors get down to the size of atoms there are problems such as quantum tunneling.

I’m not sure what will happen when Moore’s Law slows down to a trickle.  Some companies like Intel enjoy a premium valuation because they are able to drive dramatic increases in performance year after year.  Their margins may get crushed when they can no longer continue to deliver.

System on a chip (SoC)

The industry trend has been towards consolidating multiple chips onto a single system on a chip.  This lowers cost as long as the overall SoC isn’t too large.  This trend can create problems for the design companies whose designs aren’t chosen to be on SoCs.

The investment challenge

Understanding the market and the technology

Sometimes this can be a challenge as journalists are often wrong.  Some of them have a poor technical understanding of their subject (usually because they have worked in journalism but never in the industry they write about).  Sometimes journalists hop onto a bandwagon that turns out to be dubious (e.g. the Y2K scare).  Nobody is going to get a Pulitzer for explaining why Y2K is irrelevant or how other journalists are wrong.  You need to do your own research and filter out the noise.

Talking to industry participants directly may be helpful.  But that is not something that I do.  Sometimes industry participants will have blogs or they will make comments on news websites for their trade(e.g. eetimes.com).

Most semiconductor industries are large enough that you can find a lot of information about that industry on the Internet.  You can find a lot of technical information on how the products work and how they are built.

The correct multiple

Because market share can change very quickly, I believe that technology companies with dominant market share deserve a lower P/E multiple than those in other industries.  If they lose their #1 spot, they will not be able to harvest much profits out of whatever is left in their business.  Of course, not all technology markets are like this.  Mainframe companies are still profitable because it is so costly for their customers to leave them.

Conversely, if technology companies can quickly gain market share, I would think that potential future market share leaders deserve a higher multiple than companies in other industries.  However, this type of investing could conceivably get one into trouble.  Mentally, it is easy to get overoptimistic or greedy or to engage in momentum investing.

Regardless of whether you are investing in the current market leader or a potential future market leader, it is important to be able to predict who will be the market leader several years from now.  This is difficult to do with a margin of safety.

Spotting the winners and the losers- my take

Among semiconductor manufacturers, the company that is #1 in market share will likely make almost all the money.  Unlike other industries, being #2 may not be good enough.  In CPUs, Intel has made enormous profits while AMD still has negative retained earnings.

On the design side, it can be difficult to predict who the market leader will be in the future.  If we go back in time, it is obvious that IBM underestimated the rise of the personal computer.  It invented the IBM-PC yet failed to capitalize on its invention.  In hindsight, it gave away the crown jewels (the operating system and CPU design/manufacturing).  IBM is run by some extremely smart people who failed to predict the future.  I don’t think that investors can expect to do a much better job of predicting the future.  It’s not just IBM.  Back then, Microsoft thought that the smart TV was the future (e.g. Bill Gates wanted to buy John Malone’s Liberty), many people thought that RISC would always rule in high-performance computing, etc. etc.

Brief comments on semiconductor markets

Desktops/laptops

It is likely that Intel will continue to dominate due to its manufacturing advantage over AMD.  The overall market will likely continue to grow slowly as emerging economies increase their adoption of PCs.  The average selling price of CPUs could decline somewhat as CPUs are increasingly becoming good enough and newer processors aren’t really getting faster for desktop usage (desktop performance depends mostly on single-threaded performance, which has stalled).

Discrete graphics

Nvidia and ATI/AMD will continue to see the low-end of the discrete graphics market get eaten away by integrated graphics chips from AMD and Intel.

Some of this loss will be offset by increasing usage of general purpose GPUs, but the GPGPU market may only be a $1B/year market in revenue.  I do not see discrete graphics as being a growth market.

Smartphones

The software side to smartphones is very important as it makes a huge difference between purchasing an Apple phone (Apple designs its own SoCs), an Android phone (currently almost all are powered by ARM SoCs), or another phone.  I am not sure which software ecosystem will win though I would lean towards the company with the greatest market share (Android).  On the hardware side, I think that Intel will eventually dominate though it is not a sure thing.

The smartphone market may see its huge growth continue as there are “killer apps” in maps, GPS, social networking, and web browsing.  I think that eventually almost everybody will own a smartphone much like PCs.

Tablets

Like smartphones, I am not sure which software ecosystem will be the winner (Android, iOS, Windows 8, etc.).  I do not think that tablets will replace/cannibalize the PC as they are mostly used for reading (with light typing) and watching movies.  I don’t see adoption rates for tablets being anywhere as high as smartphones and PCs.

On the hardware side, I would pay close attention to Microsoft’s Surface product as it is somewhere between a tablet and a laptop/ultrabook.  It may be an indicator for how the ARM versus Intel/x86 battles will play out in other markets.

Servers

I think that Intel will continue its dominant market share.  Intel and AMD combined will continue to take share from other vendors.  Intel’s Centerton and AMD’s Seamicro will go head to head in the wimpy core niche.  I’m not sure that ARM chips will gain much traction.

HPC, Supercomputing

This is its own esoteric market (and a fairly small one).  When it comes to high-performance computing, cost goes up exponentially for increased performance since performance is difficult to scale by simply adding more hardware.  The trend away from RISC-based CPUs is likely to continue.

Memory

Historically not a great business.  I wouldn’t want to go long these stocks.  While the flash memory market will probably continue to grow and many consumers will gravitate towards flash/solid state drives in their computers over traditional hard drives, the manufacturers of flash memory may not benefit from this trend.

Contract fab

TSMC is currently the #1 player and will probably make the most money in the contract fabrication market.  I wouldn’t really want to own the smaller contract fab companies.  Fortunately for TSMC, Intel is not entering the contract fab business in a meaningful way.  Unfortunately for TSMC, Intel will compete directly with the smartphone SoC customers that are driving TSMC’s growth.  There is a small possibility that TSMC will one day eclipse Intel as the world’s largest semiconductor manufacturer.

LED lighting

The growth of LED lighting will likely continue for a long time.  However, the LED manufacturers may only make small lumpy profits.  LED lighting is mostly a commodity business for the semiconductor manufacturers.

*Disclosure: Long Intel, I may be a little biased.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s