What could back an open market for chiplets?

By Chris Edwards, New Electronics | November 18, 2025

For as long as the idea of the system-on-chip (SoC) has existed, the graphs of design and non-recurrent engineering costs have been reliably exponential in the wrong direction. Each successive process node just pushes the number even further into the stratosphere. The payoff was, naturally, you get more for your money in terms of silicon capability with each notch of the ratchet.

Then, even silicon became more expensive on a per-transistor basis. Not by much, but it deviated from the long-term trend of Moore’s Law. At the International Electron Device Meeting in 2023, Google chip-packaging chief Milind Shah placed the switch at 20nm, just ahead of the foundries’ move to finFETs.

Chipmakers companies like AMD could see this problem coming and began, as the finFET generations appeared, the move to chiplets. Dividing a large, all-encompassing SoC into multiple chiplets makes it possible to dedicate the most advanced silicon to the circuits that can take best advantage. For AMD, those were the x86 cores in its high-end processors. The I/O and memory controllers? Those can use more mature, more analogue-friendly and cheaper process nodes.

To read the full article, click here