The Monster Footprint of Digital Technology
The embodied energy of a memory chip can exceed the energy consumption of a laptop
Tuco’s Child Preface
Dear Readers, when I was a semiconductor process engineer at Applied Materials and other companies in the Valley, I was always astounded by the massive power consumption and inefficiencies in the processes used to make computer chips. The power consumption, material consumables and process tool costs for chip manufacturing are truly astounding.
This is because computer chips are the most complex low defect and low entropy (low disorder) devices ever created by human-kind. For all the order created in a chip, more disorder is created elsewhere. The following article discusses this in some detail:
Every 1 and 0 costs, even as I type these characters that register on some Substack remote server, generating waste heat.
But before one can get to a working microprocessor or a solar panel, there is even more energy consumed and carbon burned, because silica specifically needs carbon to make pure silicon, as covered here:
Given the above, I share the following article, written by Kris De Decker of Low-tech Magazine. This article is an oldie but a goodie, and is an easy to understand overview about how costly 1s and 0s really are.
The Monster Footprint of Digital Technology
The power consumption of our high-tech machines and devices is hugely underestimated.
When we talk about energy consumption, all attention goes to the electricity use of a device or a machine while in operation. A 30 watt laptop is considered more energy efficient than a 300 watt refrigerator. This may sound logical, but this kind of comparisons does not make much sense if you don’t also consider the energy that was required to manufacture the devices you compare. This is especially true for high-tech products, which are produced by means of extremely material- and energy-intensive manufacturing processes. How much energy do our high-tech gadgets really consume?
The energy consumption of electronic devices is skyrocketing, as was recently reported by the International Energy Association ("Gadgets and gigawatts"). According to the research paper, the electricity consumption of computers, cell phones, flat screen TV’s, iPods and other gadgets will double by 2022 and triple by 2030. This comes down to the need for an additional 280 gigawatts of power generation capacity. An earlier report from the British Energy Saving Trust (The ampere strikes back- pdf) came to similar conclusions.
There are multiple reasons for the growing energy consumption of electronic equipment; more and more people can buy gadgets, more and more gadgets appear, and existing gadgets use more and more energy (in spite of more energy efficient technology - the energy efficiency paradox described here before).
The 180 watt laptop
While these reports are in themselves reason for concern, they hugely underestimate the energy use of electronic equipment. To start with, electricity consumption does not equal energy consumption. In the US, utility stations have an average efficiency of about 35 percent. If a laptop is said to consume 60 watt-hours of electricity, it consumes almost three times as much energy (around 180 watt-hour, or 648 kilojoules).
So, let’s start by multiplying all figures by 3 and we get a more realistic image of the energy consumption of our electronic equipment. Another thing that is too easily forgotten, is the energy use of the infrastructure that supports many technologies; most notably the mobile phone network and the internet (which consists of server farms, routers, switches, optical equipment and the like).
Embodied energy
Most important, however, is the energy required to manufacture all this electronic equipment (both network and, especially, consumer appliances). The energy used to produce electronic gadgets is considerably higher than the energy used during their operation. For most of the 20th century, this was different; manufacturing methods were not so energy-intensive.
An old-fashioned car uses many times more energy during its lifetime (burning gasoline) than during its manufacture. The same goes for a refrigerator or the typical incandescent light bulb: the energy required to manufacture the product pales into insignificance when compared to the energy used during its operation.
Advanced digital technology has turned this relationship upside down. A handful of microchips can have as much embodied energy as a car. And since digital technology has brought about a plethora of new products, and has also infiltrated almost all existing products, this change has vast consequences. Present-day cars and since long existing analogue devices are now full of microprocessors. Semiconductors (which form the energy-intensive basis of microchips) have also found their applications in ecotech products like solar panels and LEDs.
Where are the figures?
While it is fairly easy to obtain figures regarding the energy consumption of electronic devices during the use phase (you can even measure it yourself using a power meter), it is surprisingly hard to obtain reliable and up-to-date figures on the energy consumed during the production phase. Especially when it concerns fast-evolving technologies. A life cycle analysis of high-tech products is extremely complex and can take many years, due to the large amount of parts, materials and processing techniques involved. In the meantime, products and processing technologies keep evolving, with the result that most life cycle analyses are simply outdated when they are published.
The embodied energy of the memory chip alone already exceeds the energy consumption of a laptop during its life expectancy of 3 years.
For more recent and emerging technologies, life cycle analyses simply do not exist. Try looking for a research paper that calculates the embodied energy of a Light Emitting Diode (LED), a lithium-ion battery or any device full of electronics meant to save energy: you won’t find it (and if you do, please let me know).
Embodied energy of a computer
The most up-to-date life cycle analysis of a computer dates from 2004 and concerns a machine from 1990. It concluded that while the ratio of fossil fuel use to product weight is 2 to 1 for most manufactured products (you need 2 kilograms of fuel for 1 kilogram of product), the ratio is 12 to 1 for a computer (you need 12 kilograms of fuel for 1 kilogram of computer). Considering an average life expectancy of 3 years, this means that the total energy use of a computer is dominated by production (83% or 7,329 megajoule) as opposed to operation (17%). Similar figures were obtained for mobile phones.
While the 1990 computer was a desktop machine with a CRT-monitor, many of today’s computers are laptops with an LCD-screen. At first sight, this seems to indicate that the embodied energy of today’s machines is lower than that of the 1990 machine, because much less material (plastics, metals, glass) is needed. But it is not the plastic, the metal and the glass that makes computers so energy-instensive to produce. It’s the tiny microchips, and present-day computers have more of them, not less.
100 years of manufacturing
The energy needed to manufacture microchips is disproportional to their size. MIT-researcher Timothy Gutowski compared the material and energy intensity of conventional manufacturing techniques with those used in semiconductor and in nanomaterial production (a technology that is being developed for use in all kinds of products including electronics, solar panels, batteries and LEDs).
Digital technology is a product of cheap energy
As an example of more conventional manufacturing methods, Gutowski calculated the energy requirements of machining, injection molding and casting. All these techniques are still used intensively today, but they were developed almost 100 years ago. Injection molding is used for the manufacture of plastic components, casting is used for the manufacture of metal components, and machining is a material removing process that involves the cutting of metals (used both for creating and finishing products).
6 orders of magnitude
While there are significant differences between configurations, all these manufacturing methods require between 1 and 10 megajoule of electricity per kilogram of material. This corresponds to 278 to 2,780 watt-hour of electricity per kilogram of material. Manufacturing a one kilogram plastic or metal part thus requires as much electricity as operating a flat screen television for 1 to 10 hours (if we assume that the part only undergoes one manufacturing operation).
The energy requirements of semiconductor and nanomaterial manufacturing techniques are much higher than that: up to 6 orders of magnitude (that’s 10 raised to the 6th power) above those of conventional manufacturing processes (see figure below, source, supporting information). This comes down to between 1,000 and 100,000 megajoules per kilogram of material, compared to 1 to 10 megajoules for conventional manufacturing techniques.
Manufacturing one kilogram of electronics or nanomaterials thus requires between 280 kilowatt-hours and 28 megawatt-hours of electricity; enough to power a flat screen television continuously for 41 days to 114 years. These data do not include facility air handling and environmental conditioning, which for semiconductors can be substantial.
Embodied energy of a microchip
The energy consumption of semiconductor manufacturing techniques corresponds with a life cycle analysis of a “typical” 2 gram microchip performed in 2002. Again, this concerns a 32 MB RAM memory chip - not really cutting edge technology today. But the results are nevertheless significant: to produce the 2 gram microchip, 1.6 kilograms of fuel were needed. That means you need 800 kilograms of fuel to produce one kilogram of microchips, compared to 12 kilograms of fuel to produce one kilogram of computer.
If we take the energy density of crude oil (45 MJ/kg), this comes down to 72 megajoules (or 20,000 watt-hour) to produce a 2 gram microchip. Converted to a one kilogram microchip this comes down to 3.3 megawatt-hours of electricity (or 36,000 MJ), well within the range of the 280 kilowatt-hours (1,000MJ) and 28 megawatt-hours (100,000 MJ) calculated above.
Also, the International Technology Roadmap for Semiconductors 2007 edition gives a figure of 1.9 kilowatt-hours per square centimetre of microchip, so 20 kilowatt-hours per 2 gram, square centimetre computerchip seems to be a reasonable estimate.
How many microchips in a computer?
A gadget or a computer does not contain one kilogram of semiconductors - far from that. But, we don’t need a kilogram of microchips to ensure that the manufacturing phase will largely outweigh the usage phase. The embodied energy of the memory chip alone already exceeds the energy consumption of a laptop during its life expectancy of 3 years.
Today’s personal computers have a RAM-memory of 0.5 to 2 gigabyte modules that typically consist of 18 to 36 two-gram-microchips (as the ones described above). This equates to 1,296 to 2,595 megajoules of embodied energy for the computer memory alone, or 360,000 to 720,000 watt-hour. Enough to power a 30 watt laptop non-stop for 500 to 1,000 days.
Microprocessors (the “brains” of all digital devices) are more advanced than memory chips and thus contain at least as much embodied energy. Unfortunately, no life cycle analysis of a microprocessor has been published. Certain is that modern computers contain ever more of them.
One trend in recent years is the introduction of “multicore processors” and “multi-CPU systems”. Personal computers can now contain 2, 3 or 4 microprocessors. Servers, game consoles and embedded systems can have many more. Each of these “cores” is capable of handling its own task independently of the others. This makes it possible to run several CPU-intensive processes (like running a virus scan, searching folders or burning a DVD) all at the same time, without a hitch. But with every extra chip (or chip surface) comes more embodied energy.
The energy savings realized by digital technology will merely absorb its own growing footprint.
Another trend is the rise of the “Graphics Processing Unit” or GPU. This is a specialized processor that offloads 3D graphics rendering from the microprocessor. The GPU is indispensable to play modern videogames, but it is also needed because of the ever higher graphical requirements of operating systems. GPU’s do not only raise the energy consumption of a computer while in use (GPU’s can consume more energy than current CPU’s), but they also stand for more embodied energy. A GPU is very memory-intensive and thus also increases the need for more RAM-chips.
Nanomaterials
Why are microchips so energy-intensive to manufacture? One of the reasons becomes clear when you literally zoom in on the technology. A microchip is small, but the amount of detail is fabulous. A microprocessor the size of a fingernail can now contain up to two billion transistors - each transistor less than 0.00007 millimeters wide. Magnify this circuit and it becomes a structure as complex as a sprawling metropolitan city.
The amount of materials embedded in the product might be small, but it takes a lot of processing (and thus machine energy use) to lay down a complex and detailed circuit like that. While the electricity requirements of machines used for semiconductor manufacturing are similar to those used for older processes like injection molding, the difference lies in the process rate: an injection molding machine can process up to 100 kilograms of material per hour, while semiconductor manufacturing machines only process materials in the order of grams or milligrams.
Another reason why digital technology is so energy-intensive to manufacture is the need for extremely effective air filters and air circulation systems (which is not included in the figures above). When you build infinitesimal structures like that, a speck of dust would destroy the circuit. For the same reason, the manufacture of microchips requires the purest silicon (Electronic Grade Silicon or EGS, provided by the energy-intensive CVD-process).
The manufacture of nanotubes is as energy-intensive as the manufacture of microchips.
Every 18 months the amount of transistors on a microchip doubles (Moore’s law). On one hand, this means that less silicon is needed for a certain amount of processing power or memory. On the other hand, when transistors become smaller, you need even more effective air filtration and purer silicon. Since the structure also becomes more complex, you need more processing steps.
Nanotechnology operates on an even smaller scale than micro-electronics, but its energy requirements are comparable. Carbon nanofiber production, which is based on many of the same techniques used by semiconductor manufacturing, requires 760 to 3,000 MJ of electricity per kilogram of material, while carbon nanotubes and single-walled nanotubes (SWNTs) manufacturing requires a hefty 20,000 to 50,000 MJ per kilogram. The manufacture of nanotubes is thus as energy-intensive as the manufacture of microchips (36,000 MJ). Many of the large-scale applications proposed for nanotubes will simply not be possible because of energy requirements.
Recycling is no solution
Encouraging recycling is often proposed as a way to lower the embodied energy of products. Unfortunately, this does not work for micro-electronics (or nanomaterials). In the case of conventional manufacturing methods, the energy requirements of the manufacturing process (1 to 10 MJ per kilogram) are small compared to the energy required to produce the materials themselves.
For instance, producing 1 kilogram of plastic out of crude oil requires 62 to 108 MJ of energy, while a typical mix of virgin and recycled aluminum requires 219 MJ. To make a fair comparison, you have to multiply the energy requirement of the manufacturing process by three (1 megajoule of electricity requires 3 megajoules of energy) but even then (with 3 to 30 MJ/kg) conventional manufacturing processes appear to be quite benign compared to materials extraction and primary processing (in the order of 100 MJ/kg - see table).
Recycling is not a solution if all your energy use is concentrated in the manufacturing process itself.
In the case of semiconductor manufacturing, this relation is reversed. While it takes 230 to 235 MJ of energy to produce 1 kilogram of silicon (already quite high compared to many other materials), chemical vapour deposition (an important step in the semiconductor manufacturing process) requires about 1,000 MJ of electricity and thus 3,000 MJ of energy per kilogram.
That is 10 times more than the energy consumption of material extraction and primary processing. In the case of conventional manufacturing techniques, the use of recycled material is an effective way to lower overall energy use during manufacture. In the case of semiconductors, it is not. Recycling is not a solution for energy consumption if all your energy use is concentrated in the process itself.
This does not mean that the manufacture of microchips does not require materials. In fact, producing microchips and nanomaterials is also more intensive than the manufacture of conventional products, by the same orders of magnitude. However, this concerns auxiliary materials which are not incorporated into the product.
For example, the embodied energy of the input cleaning gases in the CVD process (not included in the figures above) is more than 4 orders of magnitude greater than that of the product output. Furthermore, these gases have to be treated to reduce their reactivity and possible attendant pollution. Gutowski writes: “If this is done using point of use combustion with methane, the embodied energy of the methane alone can exceed the electricity input.”
The benefits of digital technology
Microchips also have positive effects on the environment, by making other activities and processes more efficient. This is the subject of a publication by the Climate Group, an initiative of more than 50 of the world’s largest companies. The report ("Smart 2020 - enabling the low carbon economy in the information age") confirms the findings of other studies regarding the electricity use of electronic equipment, but also calculates the benefits.
According to Smart 2020, the emissions from Information and Communications Technology (including the energy use of data centers, which the IEA report does not include) will rise from 0.5 Gt CO2-equivalents in 2002 to 1.4 GtCO2-equivalents in 2020, assuming that the sector will continue to make the “impressive advances in energy efficiency that it has done previously”. By enabling energy efficiencies in other sectors, however, ICT could deliver carbon savings 5 times larger: 7.8 Gt CO2-equivalents in 2020.
Addressing technological obsolescence would be the most powerful approach to lower the ecological footprint of digital technology
These benefits are smart grids (2.03 Gt), smart buildings (1.86 Gt), smart motor systems (970 Mt), dematerialization and substitution (by replacing high carbon physical products and activities such as books and meetings - with virtual low carbon equivalents such as electronic commerce, electronic government, videoconferencing, 500 Mt) and smart logistics (225 Mt). One of the first tasks of ICT will be to monitor energy consumption and emissions across the economy in real time, providing the data needed to optimize for energy efficiency.
The report concludes: “The scale of emission reductions that could be enabled by the smart integration of ICT into new ways of operating living, working, learning and travelling, makes the sector a key player in the fight against climate change, despite its own growing footprint.”
But even if we assume that all these savings will materialize (the report acknowledges that this will not be an easy task), this conclusion does not take into account the energy needed to manufacture all this equipment. If we assume the share of manufacture to be 80 percent of total energy consumption by ICT (following the only life cycle analysis of a computer we have), then the 1.4 Gt in 2020 in reality should be 7 Gt - almost as much as the 7.8 Gt that will be saved by ICT. No environmental benefit would appear and the energy savings realized by digital technology would merely absorb its own growing footprint.
Digital technology is a product of cheap energy
The research of Timothy Gutowski shows that the historical trend is toward more and more energy intensive processes. At the same time, energy resources are declining.
Gutowski writes:
“This phenomenon has been enabled by stable and declining material and energy prices over this period. The seemingly extravagant use of materials and energy resources by many newer manufacturing processes is alarming and needs to be addressed alongside claims of improved sustainability from products manufactured by these means.”
Production techniques for semiconductors and nanomaterials can and will become more efficient, by lowering the energy requirements of the equipment or by raising the operating process rate. For instance, the “International Technology Roadmap for Semiconductors” (ITRS), an initiative of the largest chip manufacturers worldwide, aims to lower energy consumption (pdf) per square centimetre of microchip from 1.9 kWh today to 1.6 kWh in 2012, 1.35 kWh in 2015, 1.20 kWh in 2018 and 1.10 kWh in 2022.
But as these figures show, improving efficiency has its limits. The gains will become smaller over time, and improving efficiency alone will never bridge the gap with conventional manufacturing techniques. Power-hungry production methods are inherent to digital technology as we know it.
The ITRS-report warns that:
“Limitations on sources of energy could potentially limit the industry’s ability to expand existing facilities or build new ones”.
Gutowski writes:
“It should be pointed out that there is also a need for completely rethinking each of these processes and exploring alternative, and probably non-vapour-phase processes”.
Technological obsolescence
The ecological footprint of digital technology described above is far from complete. This article focuses exclusively on energy use and does not take into account the toxicity of manufacturing processes and the use of water resources, both of which are also several orders of magnitude higher in the case of both semiconductors and nanomaterials. To give an idea: most water used in semiconductor manufacturing is ultrapure water (UPW), which requires large additional quantities of chemicals. For many of these issues, the industry recognizes that there are no solutions (see the same ITRS-report, pdf). There are also the problems of waste & war.
Last, but not least: the energy-intensive nature of digital technology is not due only to energy-intensive manufacturing processes. Equally as important is the extremely short lifecycle of most gadgets. If digital products would last a lifetime (or at least a decade), embodied energy would not be such an issue. Most computers and other electronic devices are replaced only after a couple of years, while they are still perfectly workable devices. Addressing technological obsolescence would be the most powerful approach to lower the ecological footprint of digital technology.
Artwork by Grace Grothaus. More information on manufacturing methods.
What this assists in demonstrating is that a modern and advanced society depends on cheap and abundant energy. Also, it’s also clear that next to energy its clean water that is required to support most industrial applications. Even in the industrial revolution factories were built close to coal and water.. and nothing much has changed. So, unless you are a crazy degrowther then we need to go back to these basics if we want the things a modern society provides and its spin off….. prosperity!
This was so interesting and enlightening even a bird dog trainer could understand it. Gave me a true appreciation for the amount of energy required to manufacture semiconductors. TSMC, Intel, ASML, AMAT, LRCH etc have always intrigued me. Incredible how dependent our world is on energy abundance.