It’s Only Common Sense: How AI Consumes the Electronics Supply Chain
Dan’s Biz Bookshelf: ‘Building a Story Brand 2.0’
When TTM opened its new UHDI facility in New York earlier this year, it wasn't simply adding another PCB factory. As the largest circuit board builder in the United States, TTM was investing in technologies that bring PCB fabrication closer to the complex world of semiconductor packaging.
The advent of AI helps explain why that matters. The enormous compute power required by AI is pushing chip performance to new levels, but those chips don't operate in isolation. They depend on advanced packages, IC substrates, and ultimately PCBs capable of handling greater densities, faster signals, more power, and more heat. If the rest of that hardware can't keep up, even the most advanced processor can't deliver its full potential.
I-Connect007 has been following how AI's demands are rippling through the electronics ecosystem, from silicon to systems. Who is building the hardware behind AI? What new demands does it place on electronics manufacturers, and how far will PCB technology have to advance to keep pace?
The companies making investments provide another piece of the picture. Amkor recently expanded its planned Arizona advanced packaging and test investment to approximately $12 billion, while TSMC and Amkor are developing a U.S. advanced packaging partnership. ASE is expanding panel-level packaging for AI applications, and companies such as MKS, Koh Young, Kulicke & Soffa, Indium, and Teradyne are developing the process, inspection, assembly, materials, and test technologies needed to manufacture complex packages at scale. The announcements vary, but they point to the fact that building the hardware behind AI requires an increasingly connected manufacturing ecosystem.
The following articles and interviews offer some insight into I-Connect007's coverage, from advanced packaging and heterogeneous integration to the board underneath it all.
What Heterogeneous Integration Means for EMS Providers
Published May 14
Intel Fellow Ravi Mahajan explains why heterogeneous integration has become important: Different functions can't necessarily be optimized on the same silicon platform, so advanced packaging brings those optimized technologies together into a functioning system. He specifically addresses what this transition means for EMS providers and the broader manufacturing ecosystem.
More Than Moore Enabled by Advanced Packaging and Heterogeneous Integration
Published August 13
Chetan Patil of Marvell explains the underlying technological problem. Making larger monolithic chips is becoming more difficult and expensive, while AI and HPC require greater compute performance and memory bandwidth. Chiplets and advanced packaging allow different functions to be optimized separately and then integrated into one system.
TTM Building the Infrastructure Around the Chip
Published September 17
Nolan Johnson interviews TTM President and CEO Edwin Roks about the company's new UHDI facility and the move toward finer-feature PCBs, semiconductor-level dimensions, substrates, and heterogeneous packaging. Roks also makes the AI connection directly: Chips are only part of the infrastructure required to support AI, and PCBs remain critical to the system.
IC Substrates vs. UHDI: The Future of Interconnect
Published March 15
Marcy LaRont talks with Jan Pedersen of NCAB about the increasingly overlapping territory between IC substrates and UHDI PCBs. Pedersen says the rapid development of chips has driven miniaturization at the interposer and substrate levels while PCBs haven't evolved at the same pace. He identifies AI as one of the primary drivers pushing UHDI into more applications.
Advanced Packaging for AI: Reliability Starts at the Cu/Cu/Cu Microvia Junction
Published April 20
Kuldip Johal of MKS' Atotech shows what all this means at the manufacturing level. Greater AI compute density and data movement increase I/O density from chip to interposer to IC substrate to PCB, putting greater demands on PCB architecture and microvia reliability. He concludes that in AI hardware, the bottleneck isn't necessarily the chip; the board underneath it can become the limiting factor.
