top of page
Marta Beckwith
Aug 227 min read

Homage to the Semiconductor Chip

A while back (see https://www.sepessentials.com/post/building-a-house), I likened standards to blueprints for a house.  But standards are actually a lot less complete than a typical blueprint.  A blueprint is defined variously as “a design plan or other technical drawing” or “a complete plan that explains how to do or develop something.”[1]  The word blueprint implies that, by following the blueprint, you will end up with a functional house or other product (although some level of detail may be missing and left open for later selection). 


A standard on the other hand is more akin to a set of requirements.  It tells you what functionality an implementing product needs to include but generally does not provide information about how to make that implementing product. Indeed, most chip-based standards are agnostic as to how a chip implements the standard, as long as it provides the specified functionality.  So, if we go back to the house analogy, a standard is less like the blueprint for the house and more like a list of needed components in order to create a house. In other words, it is more like specifying you need one hundred 2x4s, ten windows of particular sizes, a certain amount of framing lumber, drywall, etc.  At the end of the day, you could take those items and put them together in a multitude of different ways to create houses that are quite different from each other.


That is true of standards that are chip-implemented (such as WiFi and cellular standards).  Typically, these standards specify the functionality that must be included but give limited (or often no) details as to how a chip should be designed to implement that standard.  That is where the chipmakers come in. 


When talking about standards and standard implementation, we do not give the chip industry enough credit for the very difficult task they do.  The industry takes a set of technical requirements (i.e. the standard) and makes it into a working semiconductor product.  For purposes of this post, I will call all of these a chip even though they might be a multi-function system-on-chip or system-in-packaging product (see https://www.sepessentials.com/post/the-cellular-multiverse for a description of these different products).


Chips are miraculous objects[2] – they typically use a bit of silicon (obtained from that blindingly white, squeaky sand you sometimes encounter), sometimes compounded with other elements, some metal lines, some electricity sinks and some electricity gates to form and interconnect billions of very small transistors, resisters and capacitors.[3]  Chips are the “building blocks of technology,” “essential building blocks of the technologies that will shape our future”, “Vital Components of the Green Transition” and the development of green technologies, and “The Brain Behind the Machine.”[4]


Chips use nanotechnology.  Some of the elements in a high-end chip are smaller than a human hair by an order of ten thousand (10,000) or more.[5]  As one would expect from a product that contains billions of almost indescribably tiny components, making a chip is incredibly difficult.[6]


Indeed, as the size of the transistors and other chip components have decreased over the years, making a chip has become so complex that no single company does it all anymore.  Instead, chipmaking is divided into three primary sub-pieces, each of which is typically done by different companies:  (i) semiconductor fabrication equipment (done by companies such as Applied Materials (my former employer), ASML and Lam Research); (ii) chip design (done by companies such as Qualcomm and Broadcom); and (iii) chip manufacturing (done by companies such as Taiwan Semiconductor Manufacturing Corporation aka TSMC).[7]


Even the names of the different types of chips are complicated (often reflecting the underlying composition, structure or primary functionality of the chip).  These include: 

  • Analog/digital/mixed signal/quantum computing chips

  • MEMS - microelectromechanical systems

  • Memory chips:

    • RAM - random-access memory

    • DRAM - dynamic random-access memory

    • sDRAM - synchronous dynamic random-access memory

    • DDR sDRAM (which has various versions DDR1 – DDR5) – double data rate synchronous dynamic random-access memory

    • SRAM - static random-access memory

    • EEPROM – electrically erasable programable read-only memory

    • Flash – which comes in two main types: NOR (Not OR) and NAND (Not-AND)

  • Logic chips and microprocessors including CPUs (central processing units) and GPUs (graphical processing units)

  • ASICs (application specific integrated circuits)

  • SoCs (system-on-chips)

  • SiPs (system-in-packaging). 

  • MOS (metal-oxide-semiconductor) ICs (integrated circuits) built using MOSFETs (metal-oxide-silicon field-effect transistors)

  • CMOS (complementary metal-oxide semiconductors) 

  • Photonic chips - these days, as chipmakers reach the end of the physical ability to shrink chip components much past the 2 nm stage and still transmit electricity, some chipmakers are developing photonic chips that use photons (particles of light) instead of electrons (electricity) in their circuits in the once again burgeoning field of optical computing


Regardless of what type of chip is being developed, the first stage of making a new chip is design.  Chipmakers who are implementing a standard have to figure out how to turn a set of functional requirements (the standard) into a series of interconnected on/off switches that code commands capable of being processed to achieve that functionality.  They then have to figure out where to physically put all of those tiny on/off switches, how to provide electrical flow to and from them and how to interconnect them all in 3-dimensions over as many as 100 different interconnected layers.[8]  It is infinitely harder than playing 3D chess (which is, after all, only played on three levels). 


Once a chip has been designed, it must be fabricated.  Fabrication requires a highly specialized, ultra-clean manufacturing facility usually called a “fab.”  Semiconductor fabrication must be done in a clean room that meets ISO 14644-1 Class 5 or lower standards which mandates how clean the air must be.  A clean room is 10,000 times cleaner than a human operating room.  Fabs typically have multiple processing lines, and each line has multiple different types of semiconductor manufacturing equipment.  Building a single fab can cost upwards of $20 billion dollars![9]


The semiconductor manufacturing equipment in these fabs runs the gamut of different types of incredibly complex tools each of which uses cutting edge technology.  There are even specialized wafer transport boxes called FOUPs that contain an internal nitrogen atmosphere to prevent oxidation.  There are photolithographic tools that pattern the wafers using specialized masks[10] and specialized photoresist materials.  There are tools to harden the photoresist.  There are tools to deposit metals or other materials such as chemical vapor deposition (“CVD”) tools and physical vapor deposition (“PVD”) tools.  There are small linear accelerators used for ion implantation.  There are epitaxy machines to grow near perfect crystals on which to build a semiconductor device.  There are machines capable of atomic layer deposition (ALD) which produce ultra-thin, extremely uniform films.  There are tools to etch away unwanted materials some of which use a highly energized gas called a plasma (i.e. in “plasma etching”).


The chip manufacturer (often in conjunction with the chip designer or the equipment manufacturers) determines how to process a wafer to form the transistors and other chip components specified by the chip designer. This includes determining when and how to use sputtering (PVD) and chemical vapor deposition (CVD) processes to lay down tiny tiny metal lines, linear accelerators to impart carefully designed imperfections, and lithography and etching, chemical-mechanical planarization (CMP) and other tools to take away unwanted deposited materials without harming wanted deposited materials. 


Even the materials used in semiconductor manufacturing are specialized.  Most chips are built on single-crystal silicon substrates, but some are built on germanium and others on gallium arsenide (GaAs) or silicon carbide (SiC) wafers.[11]   The materials used in the consumable targets that are sputtered by a PVD tool could be ultra-pure aluminum, copper, tungsten, cobalt, chromium, molybdenum, nickel, niobium, tantalum or an alloy of some of them, or of precious metals like gold, palladium or platinum or more esoteric rare earth materials such as hafnium, iridium or ytterbium.  Plasma etching (and other fabrication) processes often involve the use of very dangerous gases which must be safely manufactured, transported, maintained and managed.  They may also involve the use of ultra-high purity “noble” gases such as Argon.[12]


I have been part of the semiconductor industry for nearly 25 years and chips and chipmaking still amaze me.  The technology is so cool, complex and interesting and the end product is incredibly important to almost every aspect of modern life.  It really frustrates me that the art of chipmaking is given such short shrift in discussions about standards, SEPs and SEP valuations.  The way we currently value standards almost completely discounts the incredible value and technology contributed to implementation by the semiconductor equipment industry, semiconductor chip designers and semiconductor chip manufacturers. 


[1]          The first definition is from putting “definition blueprint” into a Google websearch and the second from https://dictionary.cambridge.org/us/dictionary/english/blueprint#google_vignette

[2]           Here are some choice quotes about chips:

“not to oversimplify: first you have to flatten the rock and put lightning inside it” (https://x.com/daisyowl/status/841806379962646532)

And this story has some very interesting facts about chips: I Saw the Face of God in a TSMC Semiconductor Factory | WIRED

[3]          https://www.semiconductors.org/semiconductors-101/what-is-a-semiconductor/ contains a short overview of the various stages of making a chip from research and design to sand and ingots and final manufacturing.  See also https://www.asml.com/en/technology/all-about-microchips/microchip-basics.

[6]          Even the folks at one of the leading semiconductor research fabs said that the level of precision necessary to make a 7 nm chip (and these days some companies are getting ready to produce 2 nm chips) is “ridiculously narrow.”  https://research.ibm.com/blog/albany-semiconductor-research-ibm.  See also, https://www.wsj.com/story/there-arent-enough-chips-why-are-they-so-hard-to-make-3e29c7e0,

[7]          There are still a few chipmakers who both design and manufacture chips (although none of them make semiconductor fabrication equipment) such as Intel and Samsung.  Integrated chipmakers used to be ubiquitous.  But manufacturing chips has become so complex and expensive that, these days, chipmakers who both design and manufacture chips are a rare breed.

[8]          There are many stages of the chip design process.  See An Outline of the Semiconductor Chip Design Flow (design-reuse.com) for an overview.  Current chips have up to about 100 layers.  How microchips are made | ASML Description.  But Samsung has announced its intent to build a 1000 layer NAND flash memory chip.  Samsung pursues 1,000-layer NAND with hafnia ferroelectrics (chosun.com)

[10]        There is an entire area of intellectual property devoted to protecting the layout designs used for these masks called “mask works.”  See Protecting Semiconductor Chip Design under the Semiconductor Chip Protection Act of 1984 (SCPA) - Part I (Registration and Inspection) - Lexology.

 

bottom of page