jontrossbach

Very well said.

Maybe on a slow day. But I would wager 80 percent of those orders are not made in house when they are at high demand, which is what matters in this context. When they have a ton of orders they are most likely contracting most of them to TSMC.

Could you clarify what you think is happening?

To my knowledge Infineon, TI and STMicro are all technically "fabless". The term "fabless" is also misleading because "Fabless" actually means Foundry-less. This is part of the problem with reporters on this topic, they actually don't know that these companies don't have foundries and endorsed a "Fabless" business model long ago.

And the idea that a fire at a single fabrication site (not a foundry) caused this long of a chip shortage is not a likely scenario. Any explanation for the ongoing microchip shortage that has to do with fabrication sites and not foundries is likely a flawed understanding.

Yeah, my understanding from trying to follow this as closely as possible is that there are three different supply-chain problems all happening at once, getting conflated by journalist who don't have the expertise to report on the shortage correctly.

Sub-7nm chips appear to have had a genuine shortage from demand shocks during Covid.

EVs seem to be having a supply or demand shock for some reason.

And legacy node lithography chips (~250nm) were decommissioned and chip suppliers like autochip makers have been left scrambling to retrofit their designs to newer ~140nm-node tech.

Seems clear foundries have not been transparent with the public prior to the CHIPS Act.

Legacy lithography machines appear to have been decommissioned setting off a domino effect across the legacy node chip industry.

The 20’s Roaring Chip Shortage: Takeaways from the close of 2022

> I don't care if it is capable of rendering something invisible; the electrical continuity is obviously there.

Side channel leakage of these keys is the area of research you are talking about and that is why that type of research gets hella funded right now. They've gotten really good at hiding these signals. It's foundational to how fabless semiconductor companies operate, or else they would have their IP stolen faster than they could produce it. The chapter I referenced "a brief history of logic locking" goes over the missteps they've discovered over the years.

> got a patent number or other name for it?

I actually don't know where to find it in the literature. I learned it in class at my masters program at NYU and only because I thought to ask the professor. I don't have access to enough journals to trace back that far. It is probably in a material science journal somewhere with some crazy chemical compound name. But the EPIC paper I linked references a bunch of other logic locking schemes, which if you read some of those papers, which I think many are publicly available, you will be able to easily tell what they are saying is not possible without the flashing-glue.

> thanks for Patterson and Hennessey

And yeah, that Patterson and Hennessey talk is great. I remind myself python has the potential for 60,000 times speed-up when I think I'm bored with computers.

If you like my first article feel free to subscribe; It's title: Rage Against the Die-ing of the Locked Silicon. It's on logic locking, something I find many engineers overlook.

Going to start dropping thoughts into this substack. Maybe weekly.