Behind the Scenes: How GPUs Are Made and Why It Takes So Long

We may earn a commission when you use one of our links to make a purchase.

The High-Level Design: Where It All Begins

The journey of a GPU starts long before it hits the shelves. It begins with the high-level design phase, where engineers draft the Product Requirement Specification (PRS). This document outlines every feature the new chip must have, acting as a checklist throughout the design process. Given the complexity and cost involved, this phase alone can take up to six months and involves thousands of engineers, including architects, hardware designers, and software engineers[8]. The result is a consumer product like the RTX 4090, enabling AI products and advanced gaming experiences.


Floorplanning and Circuit Verification

Once the PRS is finalized, the next step is floorplanning and creating a netlist, which maps out where each component will go on the chip. This is followed by circuit verification and emulation, a processor-intensive operation that ensures the design is error-free. Given the complexity of modern GPUs, this phase requires supercomputing resources to simulate and test the design, often involving thousands of tests per block[8].

Making the Masks: The Photolithography Process
After verification, the design transitions to the fabrication stage, starting with the creation of photomasks. These masks are used in the photolithography process to etch the intricate patterns of the circuit onto silicon wafers. Each layer of the chip requires a separate mask, and creating these masks is a highly precise and time-consuming process. The masks are made from quartz glass coated with a patterned layer of metallic chromium, which acts as a stencil for the photolithography process[8].

The Role of TSMC and Samsung
Most GPU manufacturers, like AMD and Nvidia, are fabless, meaning they design the chips but outsource the manufacturing to foundries like TSMC and Samsung. These foundries use advanced processes, such as the 5nm technology node, to pack more transistors into the same space, increasing power and efficiency. However, the complexity of these processes means that even minor issues can cause significant delays[11].

Assembly and Testing: Bringing the GPU to Life
Once the wafers are ready, they are cut into individual dies and assembled onto printed circuit boards (PCBs). This involves attaching various components, including the GPU itself, memory chips, and power connectors. The assembly process uses both automated machines and manual labor to ensure precision. After assembly, each GPU undergoes rigorous testing to ensure it meets performance and reliability standards. This includes stress tests that push the GPU to its limits to identify any potential issues[12].

Packaging Bottlenecks: The CoWoS Challenge
One of the most significant bottlenecks in GPU production is the packaging process, particularly for high-performance GPUs. Nvidia's H-class GPUs, for example, use TSMC's 2.5D Chip-on-Wafer-on-Substrate (CoWoS) packaging technology. This multi-step, high-precision process is crucial for performance but slows down production. TSMC has acknowledged that it will take at least 1.5 years to bring the packaging process backlog back in line, highlighting the challenges involved[17].

Environmental and Ethical Considerations
The production of GPUs also has significant environmental and ethical implications. The raw materials required, such as tungsten, copper, and gold, are often sourced through environmentally detrimental methods. The manufacturing process itself generates emissions and waste products, adding to the environmental impact. Companies like Nvidia are increasingly aware of these issues and are taking steps to mitigate their environmental footprint[14].

The Future of GPU Manufacturing
The GPU manufacturing process is a marvel of modern engineering, involving multiple stages requiring precision and expertise. From the initial design to the final assembly and testing, every step is crucial and time-consuming. As demand for GPUs continues to rise, driven by advancements in AI and machine learning, manufacturers are investing in new technologies and processes to meet this demand. However, the complexity of GPU production means that delays and bottlenecks are likely to remain a challenge for the foreseeable future.

By understanding the intricacies of GPU manufacturing, we can better appreciate the technology that powers our digital world and the efforts required to bring these powerful chips to market.