Table des matières

Wafers

System definition

A silicon wafer is a thin slice of silicon (semiconductor material) that is used in the manufacturing of integrated circuits.

What is it used for?

Wafers are at the basis of many electronic components like RAMs, CPUs, GPUs, SSDs, etc.

Are they different types and technologies?

Yes are many different types of wafers, depending on material, Crystal structure, processing, etc.

Here we focus on the wafers used in electronics : monocrystalline silicon wafer.

Several wafer technologies exist, with the two main types being logic and memory:

Life Cycle

Manufacturing

What is it made of?

A silicon wafer is made from single-crystal silicon, which is derived from high-purity sand.

How is it manufactured?

The process involves purifying the sand, melting it, and re-crystallizing it into a large, pure silicon ingot (Czochralski process). This ingot is then sliced into thin discs, which are polished to a mirror-like finish to form the wafer.

Once the disks are obtained, the main steps are:

The resolution of each photolithography technique directly impacts the achievable node size, with EUV technology leading the way in current semiconductor advancements.

See Gauthier's posters for more details

Who are the main manufacturers?

For logic wafers: Intel, AMD, Qualcomm. They rely on foundires like TSMC and GlobalFoundries. For memory wafers:

Source : https://ieeexplore.ieee.org/document/10413715

The semiconductor industry is comprised of three main types of companies, each playing a distinct role:

Additionally, it is increasingly common for IDMs to outsource part of their production to foundries. Some IDMs, like IBM and Samsung, also offer foundry services to fabless companies.

Where are they located?

The Asia Pacific region is the largest semiconductor industry region, with China as the leading single-country market. Despite a 15% decline since 2021, China's market remains critical. The Americas and Europe show resilience, with Europe's market growing by 4% in 2023. Japan experienced a slight decline, reflecting the competitive nature of the market. These regional dynamics underscore the semiconductor industry's global interconnectedness and sensitivity to geopolitical and economic factors.

Source: https://www.semiconductors.org/wp-content/uploads/2024/05/SIA-2024-Factbook.pdf

The global semiconductor manufacturing industry is highly concentrated in East Asia, the United States, and the European Union with these regions housing over 90% of all facilities. East Asia, with 292 sites, represents around 66% of the total, dominating global chip production. Taiwan alone produces 60% of the world's semiconductors and 90% of the most advanced ones, especially logic chips. Additionally, Korean companies hold a commanding 60% share of the memory chip market. This map visualizes the geographic distribution and processing capabilities of semiconductor manufacturing facilities.

Cutting-edge wafer fabrication, particularly at nodes of 7nm or less, is predominantly found in South Korea and Taiwan. Europe, on the other hand, lacks these advanced facilities and as a consequence lags in adopting more mature technology nodes. A tiny share of wafer capacity for nodes between 10nm to 20nm exists in Europe, mainly due to Intel’s fabs in Ireland and Israel, which are currently not available for contract manufacturing. Furthermore, European fabs like those from STMicroelectronics and GlobalFoundriesoperate nodes from 22nm to 40nm, but the majority of Europe’s capacity (almost 50%) consists of older nodes of 180nm or larger, used extensively for automotive and industrial applications.

Source: https://technologyglobal.substack.com/p/semiconductor-manufacturing-facilities

Environmental footprint

What are the main environmental impacts associated to the materials?

TBC

What are the main environmental impacts associated to the manufacturing processes?

The main impacts are:

What are the potential parameters affecting environmental impact?

Already existing data? (dataset, parametric model, paper, etc.)

Negaoctet's datasets, imec tool, Resilio's parametric model

Next steps

What do we know we don't know?

What are the identified challenges?

TBC

What paths/ideas should be explored?