Because embodied carbon is a poor architectural design metric and operational carbon remains a major issue

Because embodied carbon is a poor architectural design metric and | itkovian

I. Carbon incorporated

Recently, embodied carbon, defined as the Scope 3 greenhouse gas emissions that result from manufacturing processes leading to computer electronics, has become popular as an architectural metric for sustainability. If you are considering, or using it, as a technology metric for computer architecture projects, it is imperative that you understand its limitations as a technology metric. Embedded carbon has several problems:

Figure 1. Ratios of embodied carbon for microprocessors and H2 gas.

    1. Embedded carbon is highly sensitive to supplier selection. For example, Intel recently announced that its Sapphire Rapids server processor is made with more than 90% renewable energy. [Intel23a,Intel23b]. Processors or GPUs from other silicon manufacturers are estimated to be 12% renewable energy, according to recent revelations [TSMC22]. Simple arithmetic suggests (88%/10%) at least an 8x difference in embodied carbon based on supplier selection. See Figure 1. The difference for microprocessors is due to major efforts by the US semiconductor manufacturing industry to « clean up » since the 1970s, regional differences in electrical grid design, and various corporate priorities in decarbonizing their power supply.

A simplistic view of this situation would be that to get the same embodied carbon, Supplier A’s designs would need to be 1/8th the size, to get the same embodied carbon. This is a difficult conversation to have in a semi-fabulous design company (e.g. Apple, NVIDIA, AMD, …), where design optimization usually already means « offering maximum capacity » for a given area of ​​silicon. Of course, there are other elements of embodied carbon beyond the carbon content of electricity, but a number of studies suggest that it is the main and most interesting element, manufacturing process experts have suggested that many of the other greenhouse gases emitted can be destroyed with clean combustion processes. [IMEC21].

Is it surprising that there is such a big difference? From a sustainability point of view, really not. Consider the situation with hydrogen gas (H2) and the differences in incorporated carbon due to the manufacturing process. The « brown » gaseous hydrogen, the most widespread, has 21 kgCO2/kgH2 because it is generally made up of natural gas. In contrast, « green » hydrogen, produced using hydrolysers and renewable energy, can contain embodied carbon as low as 1.17 kgCO2/kgH2 [Gupta22]. This difference is 18 times for the simplest of chemicals.

Hence, we should expect large and persistent differences in embodied carbon, based on supplier, process and supply chain. If these differences are large (10x), it makes it difficult to use embodied carbon as an architectural metric.

2. The embodied carbon numbers are disproportionately large. The purpose of Scope 3 GHG is precisely to ensure broader liability, including both upstream (production) and downstream (use), so Scope 3 carbon is typically the largest share (75% vs 25%) over Scopes 1 and 2. This economic approach is similar to that for toxic chemicals, child labour, etc. to align economic forces to reduce carbon emissions (and prevent « outsourcing »). These forces work in two ways.

First, the economic pressure from « customers » on their « suppliers » to reduce their embodied carbon (e.g. Apple, others). These forces are similar to recent campaigns to reduce mercury content, forced labor conditions, and even prison labor in Xinjiang. Such an approach will indeed make headway over time, with supplier decarbonisation goals by 2030 being an example [Apple22].

The second approach is for customers to request/pay their suppliers for decarbonisation. Since supply chains typically feed into an escalating value chain, downstream profits are often much greater. Should TSMC customers just pay for TSMC’s rapid decarbonization? By our estimates it would require less than 1% of their annual profits and any of them could easily foot the entire bill. They just need an economic incentive to pay for it! (The nascent SEC climate risk reporting requirements for Scope 3 GHGs – includes embodied carbon – could be such an incentive. Consider lobbying FOR these reporting requirements!) And yes, it’s true that many of TSMC’s clients are lobbying against these reporting requirements. To eliminate the carbon footprint for its estimated sustained energy consumption of 2.3 GW, we use an estimate of $1.4 billion.

Company (TSMC 2022 Revenue Sharing) Recent Fiscal Year Earnings Estimated Profit Share – Eliminate your share of TSMC’s carbon footprint in 1 year Profit Fraction to Eliminate 100% of TSMC’s Carbon Footprint in 5 Years

Apple (30%) $99.8 billion (2022) 0.5% (1.4%/5) = 0.28%
NVIDIA (2.8%) $4.4 billion (2023) 0.9% (32%/5) = 6.4%
AMD (4.9%) $3.0 billion (2021) 2% (46%/5) = 9.2%
Qualcomm (3.9%) $12.9 billion (2022) 0.4% (11%/5) = 2.2%

The table omits other significant customers such as MediaTek, Broadcom and Sony. But the point is, if motivated, TSMC’s customers could fund the decarbonization of TSMC’s power supply in a short period of time, eliminating the associated embodied carbon footprint.

3. Researchers have highlighted problems with embodied carbon as a metric, involving double, triple, quadruple, etc. counting. [Bash23]. While these shortcomings do not make embodied carbon directionally incorrect (reducing it is still good!), they do make methods that use sums of embodied carbon and operating carbon as a metric problematic because they overweight embodied carbon; most companies with complex supply chains will be 75-90% embodied carbon. So, you should read any statement that reads « embodied carbon is the majority of the carbon footprint » with caution.

II. Operating carbon

The operational carbon of computing is still a major issue and will remain so for decades. Large-scale computing and edge facilities (such as data centers) continue to consume an increasing number of terawatt-hours of energy, much of it generated from fossil fuels. There are three important things to understand:

    1. Despite claims about « carbon neutral, » « 24×7, » and « zero carbon, » the reality is that none of these offsetting measures eliminate the carbon impact of computer energy consumption. When data centers consume energy from a mixed generation grid (e.g. fossil fuels), they are responsible for those carbon emissions.
    2. Data center load is growing rapidly, perhaps as much as 25% a year [Chien23], and is at levels threatening to grid reliability (20%, 14% of total grid capacity) in a growing number of power grids. This phenomenon should be expected in an increasing number of networks in the coming years [DomVPE23]. (see figure 2)
      1690205682 239 Because embodied carbon is a poor architectural design metric and | itkovian

      Figure 2. The rapid growth of data centers in Northern Virginia has required utilities to raise estimates of future energy needs. For Dominion Energy, the 2030 estimate increased by 12 GW in the short period from 2019 to 2023.

    3. The decarbonisation of the network is slow. The US electric grid is projected to be 44% renewable and 56% carbon-free by 2050 with full decarbonization well beyond 2070. [EIA22] The situation in the EU is better with a binding target of 42.5% renewables by 2030 [EC22]
    4. Data center load delays grid decarbonization as it generally fails to respond to the availability of renewable energy, a growing problem as grids strive for ever higher levels of renewable generation [LC21]. The problem is near-constant loads that fail to align the variation in renewable generation; the key to cost-effective grid decarbonisation. While promising, battery energy storage remains too expensive to balance supply on a large scale for hours, days or weeks. The energy storage used in many grids is mostly used for short-term purposes, such as increasing the ramp rate.

How to reduce the operational carbon of IT? The key is to align the computing load with the availability of renewable energy. This means the displacement of the load in time and space. And that requires new innovation in computer architecture, chips, systems, and data center scale.

The architectural challenge is to provide a fixed amount of compute, at a time-varying rate, that allows for alignment with renewable generation, all at the same (or similar) TCO. It is difficult to deliver processing wide dynamic range at low capital cost, but our growing dark silicon challenge means we can achieve it with solutions that increase the margin in power/heating budgets. And, operating at higher levels of PUE! [Chien22]

About the author: Andrew A. Chien is the William Eckhardt Distinguished Service Professor at the University of Chicago. His research interests include parallel computer architecture, cloud software, datacenters, programming systems. He is the leader of the Zero-Carbon Cloud project and leader for cyber sustainability. From 2017-22, Dr. Chien served as Communications Editor-in-Chief at the ACM and from 2005-2010 as Vice President of Research at Intel. He has held professorships at UCSD and Illinois (UIUC) and is a member of the ACM, IEEE and AAAS. Dr. Chien received his bachelor’s, master’s and doctoral degrees from the Massachusetts Institute of Technology.

Disclaimer: These posts are written by individual contributors to share their thoughts on the Computer Architecture Today blog for the benefit of the community. Any views or opinions represented in this blog are personal, belong solely to the blog author, and do not represent those of ACM SIGARCH or its parent organization, ACM.

Hi, I’m Samuel