Top Three Things to Watch in 2023

It is the start of a new year, 2023, so time for predictions. I'm not going to predict a lot of specific things, but I will predict what I think will be the biggest stories in the semiconductor industry in 2023:

• The fracturing of the global semiconductor supply chain
• 3D heterogeneous integration
• Intel

The End of Globalisation

Until the shortage of semiconductors at the start of the pandemic, largely caused by the automobile companies inadvisedly canceling all their orders, I don't think the average politician could have told you what a semiconductor was. And I don't just mean in the US, around the world. Probably, the same goes for the CEOs of those foolish automobile companies (Tesla aside — they knew exactly what a semiconductor was). Suddenly, it became the number one thing on everyone's to-do list. Every region in the world decided it needed its own semiconductor industry not controlled by other nations. See my post China, US, Europe: Everybody's Got a CHIPS Act.

One thing for sure is that this will result in a lot of inefficiencies and so, ultimately, higher prices for semiconductors. Long ago politicians decided every country needed one or two of its own automobile companies, and a flag-carrier airline. This has resulted in a huge overcapacity for car manufacturing (I've read 30%) and lots of extremely unprofitable airlines (hello ITA Airlines, fka Alitalia, Italy's loss-making national champion, and still unprofitable).

Obviously, the Biden administration's October 7th's rules will accelerate this. It is worth emphasizing that these rules only affect the most advanced processes and the reality is that most chips are not designed in these nodes. For example, 28nm is still a huge node, the last really effective planar node. This means that its manufacturing costs are low. When I said "really effective" was no ignoring 20nm which was also a planar node but the leakage currents were too high and the node has pretty much faded away. A good summary from late December last year is Modern Diplomacy's Semiconductor Arms Race” in Microelectronics.

3D Heterogeneous Integration

Chiplets, advanced packaging, 3DIC, 3DHI, or others. No matter term you use, this has already become one of the hottest areas in semiconductor design and I expect this to remain so. I wrote about this most recently in my post 3D Heterogeneous Integration (3DHI).

A couple of interesting datapoints. At the recent RISC-V Summit (see my post RISC-V Summit 2022) Ventana Micro Systems announced that its datacenter level RISC-V processor would be available as a chiplet as well as packaged. Another datapoint is that static memory at 3nm appears to be no denser than at 5nm, at least as reported at IEDM. This means means putting a large cache on an SoC makes no sense. Putting the memory on top of the processor, in a less advanced process, as AMD has done with what it calls V-cache, seems a better approach. I think we will see more announcements like this ijn 2023, things available as chiplets, and things that it makes no sense to integrate on the most advanced nodes, which are hard-to-design and expensive to manufacture.

This will also be the year that we start to see real use of UCIe interfaces, the Universal Chiplet Interconnect Express.

Sometimes the chiplet approach is called More than Moore. But that's not actually fair to Gordon Moore, in his famous piece in Electronics, he also went on to predict chiplets when he said:

It may prove to be more economical to build large systems out of smaller functions, which are separately packaged and interconnected.

Intel

There are several different reasons to keep an eye on Intel during 2023. One is its process roadmaps which Intel believes will give it back process leadership. Above is the current state of the process roadmap as reported at IEDM by Ann Kelleher, who has to deliver it. I reported on that in my post IEDM Keynote: Ann Kelleher on Future Technology. This year will be a make or break with Intel 3 coming into risk production and a lot of progress to be expected on 20A and 18A set for delivery in 2024.

The next thing to watch at Intel is whether it can successfully create a large, profitable foundry business. I'm not sure what to make of the fact that Randhir Thakur, who was leader of the foundry unit IFS, is leaving, but not until the end of Q1 2023. Intel is clearly creating more separation between its manufacturing and other parts of the company. It remains to be seen if it goes all the way and splits it into a separate manufacturing/foundry company, and everything else (especially all the processors). Tying into the chiplet section above, Intel is a leader in advanced packaging with its EMIB and Foveros technologies, which it will also provide as part of its foundry offering.

Intel is also investing a lot in new fabs. Two in Arizona, two planned in Ohio, at least on planned in Germany. Again, with weakening of the semiconductor industry overall, it remains to be seen if this buildout continues on schedule. A lot of capex has been cut across the industry already in 2022. To some extent, this fab building program goes along with the semiconductor supply chain starting to break down into regional blocs. If the US is going to return to being a semiconductor manufacturing powerhouse again, Intel is the one to watch (along with Micron in memories).

 

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.

.