A Different Kind Of Material World

Drivers for the next wave of disruptions in chip manufacturing.


The semiconductor manufacturing world is poised for big change, and the driver will be materials.

Materials always have been a critical factor in semiconductors. Silicon is so important that an entire region of California is named after it. Rare earths have raised fears about nationalistic monopolies. And the shift from aluminum to copper interconnects at 130nm caused one of the most painful disruptions in semiconductor manufacturing.

Still, most of the changes in materials have been relatively slow-moving and methodical. Gallium nitride, silicon germanium and silicon-on-insulator have been waiting on the sidelines for years. It took almost a decade before FD-SOI gained a solid foothold in markets such as automotive and IoT. Likewise, the choice between cobalt and ruthenium had been discussed for the better part of a decade, with cobalt interconnects finally being implemented last year.

But the move to new materials is likely to start accelerating for several reasons. First, the reduction in performance scaling benefits at 10/7nm and beyond means chipmakers will need to look at alternative ways to regain that performance, and higher-mobility materials are a likely candidate. There has been serious talk about bulk CMOS running out of steam for the past two process nodes, and plenty of testing of III-V materials in commercial applications. There also has been extensive and continuing research into new transistors based on carbon nanotubes or graphene.

The problem facing the III-V compounds is they are expensive, they don’t conduct heat as well as silicon, and controlling leakage is a problem because of a narrower bandgap. But they do allow electrons to move much more freely than silicon, and as performance improvements diminish at each new node, new materials will become essential.

Second, while new materials are more expensive (at least initially), the push toward more heterogeneous integration in advanced packaging allows them to be phased in with less of a financial jolt. In fact, advanced packaging opens the door to a whole new wave of devices because new materials are just another die in a package, and it is far easier to manufacture them or grow them in smaller batches than trying to put everything onto a giant single die.

This also opens the door for in-package silicon photonics based on indium arsenide or indum gallium arsenide, which have suffered from economies of scale because they are difficult to work with using conventional silicon processes. But as chips developed separately and bundled into an advanced package, this approach begins to look much more attractive.

Third, while we’re used to thinking about processing inside of a device, there is a big push toward ubiquitous processing across a system, or across systems of systems. The classic von Neumann architecture generally is thought of as processing inside of a box of some sort, whether that’s an SoC or a smart phone or a computer. But there is growing support behind a model that leverages processing everywhere, and that model is being embraced by carmakers for autonomous driving and by the communications industry for millimeter-wave 5G.

The only way to make that work is to change the form factor of some processing elements. Sensors will be required on structures and equipment that have some electronic components, but that equipment also may have mechanical or biological components to them, as well. Inks that are conductive, devices that can be printed, and machines that require monitoring are all part of this equation, and different materials will be required to make this all work.

Taken as a whole, any of these developments could have a significant impact on chip manufacturing. Consolidation at the leading edge into one giant pure-play foundry (TSMC) and two IDMs (Samsung and Intel) could give way to a much more diverse collection of foundries, with specialization based on materials. And while none of the most advanced-node foundry operations are in jeopardy, materials are likely to make the future of chip manufacturing far more heterogeneous and diverse than it is today.


Riko R says:

Very insightful – and right on, IMHO

Leave a Reply

(Note: This name will be displayed publicly)