The Chip Shortage Keeps Getting Worse. Why Can’t We Just Make More?


Bloomberg

May 5, 2021

Shortages of semiconductors are battering automakers and tech giants, raising alarm bells from Washington to Brussels to Beijing. The crunch has raised a fundamental question for policymakers, customers and investors: Why can’t we just make more chips?

There is both a simple answer and a complicated one. The simple version is that making chips is incredibly difficult—and getting tougher.

“It’s not rocket science—it’s much more difficult,” goes one of the industry’s inside jokes.

The more complicated answer is that it takes years to build semiconductor fabrication facilities and billions of dollars—and even then the economics are so brutal that you can lose out if your manufacturing expertise is a fraction behind the competition. Former Intel Corp. boss Craig Barrett called his company’s microprocessors the most complicated devices ever made by man.

This is why countries face such difficulty in achieving semiconductor self sufficiency. China has called chip independence a top national priority in its latest five-year plan, while U.S. President Joe Biden has vowed to build a secure American supply chain by reviving domestic manufacturing. Even the European Union is mulling measures to make its own chips. But success is anything but assured.

Manufacturing a chip typically takes more than three months and involves giant factories, dust-free rooms, multi-million-dollar machines, molten tin and lasers. The end goal is to transform wafers of silicon—an element extracted from plain sand—into a network of billions of tiny switches called transistors that form the basis of the circuitry that will eventually give a phone, computer, car, washing machine or satellite crucial capabilities.

More from Bloomberg Big Take: Chip Shortage Forces Carmakers to Strip Out High-Tech Features

So Small Yet So Complex
Most chips are groups of circuits that run software, manipulate data and control the functions of electronic devices. The arrangement of those circuits gives them their specific purpose. Below is Nvidia’s GeForce RTX 3090, currently the best at turning computer code into realistic video game graphics.

NVLink interface

Used for transferring data between central processing units and graphics processing units

and between connected GPUs.

Graphic processing cluster

Clusters of logic circuits that contain most of the GPU’s core graphics functions including portions that calculate the appearance of shadows

and textures in a video frame.

Frame buffer

An area of memory used to store information that will become the picture on a display.

28.3B

Transistors

L2 and memory controller

This is where the chip stores data ready to quickly access and work on.

Total chip area: 6.28 cm²

2.6829 cm

Actual

size

2.342 cm

Input/output, display and video

This part of the chip communicates with other

parts of the computer and the gear attached to it.

Source: Nvidia
Chip companies try to pack more transistors into chips, enhancing performance and making devices more power efficient. Intel’s first microprocessor—the 4004—was released in 1971 and contained only 2,300 transistors with a node size of 10 microns, or 10 millionths of a meter. But Intel’s undisputed leadership of the following decades ended between 2015 and 2020 when rivals Taiwan Semiconductor Manufacturing Co. and Samsung Electronics Co. started building chips with better transistors: ones with dimensions down to 5 nanometers, or 5 billionths of a meter (for comparison, an average human hair is 100,000 nanometers wide.)
Designer and manufacturer:
Intel
Designer:
Apple
Nvidia
Manufacturer:
Samsung
TSMC

Transistorcount1K100K1M10M100M1B10B30BTransistor size10microns1micron100nanometers10nm51971👆Intel 40041993Pentium2001Xeon2006Core2 Duo2013A72017A112009GeForce GTX 2752020GeForce RTX 3090Transistors are getting smallerso chips can contain more

Sources: Company and industry reports, Our World in Data
Cleaner Than a Surgery
Before you put silicon into chipmaking machines, you need a clean room. A very clean room. Individual transistors are many times smaller than a virus. Just one speck of dust can cause havoc and millions of dollars of wasted effort. To mitigate this risk, chipmakers house their machines in rooms that essentially have no dust.

1 cubic

meter of air

10

Particles

10,000

Class 1 chip

manufacturing

clean room

Particles

Hospital operating theater

Each dust particle is counted as anything less than 200 nanometers (billionths of a meter) in size

Source: ASML
To maintain that environment, the air is constantly filtered and very few people are allowed in. If more than one or two workers appear on a chip production line—wrapped head-to-toe in protective equipment—that could be a sign something’s wrong. The real geniuses behind semiconductor design and development work miles away.

An employee wearing protective gear walks past machines in a clean room at the GlobalFoundries semiconductor plant, Malta, New York, U.S.

Photographer: Adam Glanzman/Bloomberg
Even with all those precautions, the wafers of silicon can’t be touched by humans or exposed to the air. They travel between machines in cartridges carried by robots that run on tracks in the ceiling. They only emerge from the safety of those cartridges when they’re inside the machines and it’s time for a key step in the process.
Video: Adam Glanzman/Bloomberg
Atomic-Level Manufacturing
Chips consist of as many as 100 layers of materials. These are deposited, then partially removed, to form complex three-dimensional structures that connect all the tiny transistors. Some of these layers are just one atom thin. Machines made by Applied Materials Inc., Lam Research Corp. and Tokyo Electron Ltd. juggle a host of variables, such as temperature, pressure, and electrical and magnetic fields, to make this happen.

One of the most difficult parts of the process is lithography, which is handled by machines made by ASML Holding NV. The company’s gear uses light to burn patterns into materials deposited on the silicon. These patterns eventually become transistors. This is all happening at such a small scale, the current way to make it work is to use extreme ultraviolet light, which usually only occurs naturally in space. To recreate this in a controlled environment, ASML machines zap molten droplets of tin with a laser pulse. As the metal vaporizes, it emits the required EUV light. But even that is not enough. Mirrors are needed to focus the light into a thinner wavelength.

1

FRONT END

59+

Oxidation and coating

Types of

equipment

Layers of insulating and conducting materials are applied to the surface of the silicon wafer. The wafer is then covered by a uniform coat of photoresist material.

Silicon nitride

Photoresist

Silicon substrate

Silicon dioxide

2

Projected

UV light

Lithography

The integrated circuit patterns

specified in the design are

mapped onto a glass plate

called a photomask.

Ultraviolet (UV) light is shone

through the mask to transfer

the pattern to the photoresist

layer on the silicon disk.

The exposed portion can then

be chemically removed.

Photomask

Projection lens

Patterns are projected

repeatedly onto the wafer

Arrow indicate movement direction

3

Development and bake

Wafers are developed to

remove the non-exposed areas

of photoresist then baked to remove solvent chemicals.

Layers unprotected

by photoresist

4

Etching

Areas of the silicon wafer unprotected by photoresist

are removed and cleaned

by gases or chemicals.

Photoresist layer

5

Doping

The wafer is showered with

ionic gases that modify the

conductive properties of the

new layer by adding impurities,

such as boron and arsenic.

Doped region

6

Metal deposition

and etching

A similar process is used

to lay down the metal links between transistors.

Metal connector

Steps 1-5 are repeated hundreds of times with different chemicals to create

more layers, depending on the desired circuit features.

BACK END

8

Completed wafer

Each completed wafer

contains hundreds of

identical integrated

circuits. The wafers are

sent for assembly, packaging

and testing which includes cutting the wafer into individual chips.

Types of

equipment

Close up of

a silicon wafer

Sources: Boston Consulting Group, Semiconductor Industry Association, Gartner

More from Bloomberg Graphics: How a Chip Shortage Snarled Everything From Phones to Cars

Burdensome Economics

Chip plants run 24 hours a day, seven days a week. They do that for one reason: cost. Building an entry-level factory that produces 50,000 wafers per month costs about $15 billion. Most of this is spent on specialized equipment—a market that exceeded $60 billion in sales for the first time in 2020.

Heavy Duty

Sales of equipment used in chip manufacturing have doubled since 2015

Global wafer fab equipment market

$60B

45

30

15

2010

2015

2020

Source: SEMI

Three companies—Intel, Samsung and TSMC—account for most of this investment. Their factories are more advanced and cost over $20 billion each. This year, TSMC will spend as much as $28 billion on new plants and equipment. Compare that to the U.S. government’s attempt to pass a bill supporting domestic chip production. This legislation would offer just $50 billion over five years.

Once you spend all that money building giant facilities, they become obsolete in five years or less. To avoid losing money, chipmakers must generate $3 billion in profit from each plant. But now only the biggest companies, in particular the top three that combined generated $188 billion in revenue last year, can afford to build multiple plants.

Big-Fish Industry

Intel, Samsung and TSMC generated almost as much revenue in 2020 as the next 12 largest chipmakers combined

Combined total

0

95

189

284

$378B

Intel

Samsung

TSMC

SK Hynix

Qualcomm

Broadcom

Micron

Nvidia

Texas Instruments

$188B

$190B

Mediatek

Infineon

Combined revenue

of the top 3

Combined revenue

of the rest

STMicroelectronics

Kioxia

AMD

Sony

Note: Figures for Samsung and Sony include their chipmaking businesses only.
Sources: Company data compiled by Bloomberg; IDC

The more you do this, the better you get at it. Yield—the percentage of chips that aren’t discarded—is the key measure. Anything less than 90% is a problem. But chipmakers only exceed that level by learning expensive lessons over and over again, and building on that knowledge.

The brutal economics of the industry mean fewer companies can afford to keep up. Most of the roughly 1.4 billion smartphone processors shipped each year are made by TSMC. Intel has 80% of the market for computer processors. Samsung dominates in memory chips. For everyone else, including China, it’s not easy to break in.