In today’s blog post, we will explore the potential of additive manufacturing (AM) in improving the process of fabricating chips. As many of you may know, Moore’s Law, formulated in 1965, states that the number of transistors on an integrated circuit (IC), commonly known as a microchip, doubles every two years. This principle has held true for more than five decades and has been a driving force behind the rapid advancement of technology. However, in recent years, there has been a slowdown in the progression of Moore’s Law, with some even suggesting its demise.
One of the main reasons for this stagnation is that transistors have reached their atomic limit, making it increasingly challenging to shrink them further. Additionally, the cost of chip fabrication plants has been increasing significantly. According to Moore’s Second Law, also known as Rock’s Law, the cost of a chip fabrication plant doubles every four years. Modern chip plants can now cost up to 10-20 billion dollars, accounting for a substantial portion of the funding allocated by President Biden for chip manufacturing in the US.
To address the need for increased computing power and reduced costs, researchers and manufacturers have turned their attention to chiplets. Unlike traditional monolithic chips, chiplets consist of multiple small ICs specialized for distinct functions that come together to form a larger circuit. This approach, known as heterogeneous integration technology, allows chiplets to be mixed and matched for various applications, like graphics and AI computation. Advances in packaging technology have made it possible to organize chiplets in both two dimensions and three dimensions, stacking them on top of each other for optimal positioning. High-speed electrical connections between chiplets are crucial for their efficient operation.
Chiplets have already gained traction and have been adopted by major players in the industry, such as Amazon, AMD, Apple, IBM, Intel, and Tesla. According to the Yole Group, a market research firm, “as much as 80% of microprocessors will use chiplet-style designs by 2027.” Intel’s recent introduction of chiplets can be seen in their new processor, “Ponte Vecchio,” which boasts over 100 billion transistors. This powerful GPU will provide Argonne National Laboratory’s supercomputer, Aurora, with over two exaflops of computing power. To put this into perspective, the computer can perform more floating-point operations in a single second than there are total seconds in the lifespan of every American.
Not to be outdone, AMD, Intel’s main competitor, has also embraced chiplet technology. Their accelerated processing unit (APU) called the MI300 surpasses Intel in transistor count, with 146 billion on the mega-chip. This APU will power a supercomputer at the Lawrence Livermore National Laboratory.
However, despite the potential of chiplets, there is a significant challenge for the US. Much of the labor-intensive packaging process has been outsourced to Asia, leaving the US at a disadvantage. With packaging becoming increasingly critical for the incorporation of chiplets, establishing domestic facilities for this process has become a top priority.
This is where additive manufacturing, or 3D printing, can play a crucial role. Chip fabrication parts are incredibly intricate and require multiple parts to be brazed together. The conventional manufacturing process for these parts can take months, and any changes in their design can significantly impact project lead-time. On the other hand, additive manufacturing can create monolithic parts in a faster and more accurate manner. This not only reduces manufacturing defects but also increases return on investment (ROI).
Moreover, companies utilizing additive manufacturing for chip fabrication can benefit from the Research and Development (R&D) Tax Credit. This tax credit, which is now permanently available, allows companies to receive tax benefits for developing new or improved products, processes, and software. The wages for technical employees involved in creating, testing, and revising 3D printed prototypes can be included as a percentage of eligible time spent for the R&D Tax Credit. Additionally, the time spent integrating 3D printing hardware and software qualifies as an eligible activity when used to improve a process. Finally, the costs of filaments consumed during the development process can be recovered when used for modeling and preproduction.
In conclusion, chiplets offer a faster and more cost-effective alternative to traditional monolithic chip manufacturing. With major players in the global chip market shifting their focus towards chiplets, it is evident that they will shape the future of computing. However, to fully harness the potential of chiplets, the US needs to invest in domestic packaging facilities. Additive manufacturing, particularly 3D printing, can aid in this endeavor by enabling faster and more accurate fabrication of chip parts. Companies utilizing this technology should also consider taking advantage of the R&D Tax Credit to further enhance their competitiveness. The future of chip fabrication is exciting, and we look forward to witnessing the advancements that chiplets and additive manufacturing will bring.
