The Next Chip Shortages?

The rollout of chiplets and heterogeneous designs could have unexpected implications on a global scale, creating a whole new round of chip shortages that will be much harder to fix.

It’s impossible to say for certain what will happen here in the wake of massive changes in chip design and a fluid and unpredictable geopolitical situation. The trade war between the United States and China began with the U.S. targeting specific companies such as Huawei and ZTE, but the focus has shifted over the past couple years to restricting the sale of technology for leading-edge chips. That includes equipment to manufacture, inspect, and test chips 7nm and below, as well as some confusion over EDA tools and some IP.

The result has been massive investment in infrastructure outside of China as the global supply chain bifurcates. That adds a degree of resiliency into the supply chain. The problem is that the vast majority of that investment has focused on chips developed at the most advanced process nodes, or specialty processes such as SiC and GaN. The disconnect is that most of the chips in use today are developed at mature nodes, and 200mm fabs have been running at or above capacity for some time. Moreover, that percentage will only grow as chipmakers begin packaging an assortment of chips in a package — including digital logic developed at the most advanced nodes and some developed at older nodes — due to the skyrocketing costs of designing and manufacturing chips at the most advanced nodes.

Some new 200mm fabs are planned, but SEMI predicts wafers per month will increase just 8.7% by 2026. Whilte that will provide some relief in the short-term, it may not be enough in the mid- to long-term.

Related to this, there are questions about just how much of a boost improvements in 193 immersion lithography will provide. Inverse lithography technology, which has been sitting on the back burner for years, is seeing a resurgence. So is the ability to print curvilinear shapes on masks. Both allow greater density in 193nm litho.

If that boost is significant enough, along with tools to take advantage of it, density in 14nm chips could increase to the point where it equals another node or two. That would eliminate the need for quadruple patterning and push density roughly to where EUV was first deployed. And this could happen at far lower costs, because much of the 200mm litho equipment in fabs is already fully depreciated. That makes 200mm chips very attractive, and it could significantly increase demand.

There are a lot of variables in play here, and plenty of uncertainty. No one knows what will happen on a geopolitical basis, or how different approaches to disaggregating SoCs will play out. Yet if all these different avenues for growth do pan out, spurring demand for chips developed with 193i lithography, the industry could see continued shortages of chips, wafers, and the equipment needed to make the chips.

Heterogeneous integration and the whole chiplet concept sounds like a great plan, but it doesn’t work without a robust supply chain behind it. The big unanswered questions are just how robust that supply chain really is, and when and whether a surge in demand will materialize to seriously stress it.