Thread by @MichaelFiller

It’s important to unpack this contrarian view:

https://twitter.com/michaelfiller/status/1513322096520310790

The U.S. is about to spend > $50B 😳 on semiconductor manufacturing via the #CHIPSAct:
commerce.gov/news/press-rel…

Europe is considering something similar:
csis.org/blogs/perspect…

There is certainly value in fixing some of the most egregious supply chain issues and geopolitical risks in today’s semiconductor industry.

However, I fear we are not considering the most basic factors that underlie the current situation — factors that are at the core of how we manufacture ICs. As a result, I fear that we are unlikely to maximally leverage and benefit from #CHIPSAct funding. That would be a bummer.

According to @Wikipedia, a scriptorium is “a room in medieval European monasteries devoted to the writing, copying and illuminating of manuscripts commonly handled by monastic scribes.”
en.wikipedia.org/wiki/Scriptori…

This approach to text manufacturing was exceedingly time consuming, costly, and tightly controlled. The barrier to entry was really high — you had to be a monk or run a church! Luckily, the printing press changed everything.
en.wikipedia.org/wiki/Printing_…

But how did it change everything? This is a great opportunity to deploy the schema for fundamental manufacturing process innovation! medium.com/@processinnova…

Everything changed because we modularized letters in the form of type, which allowed the text manufacturing process to be “factored” into two distinct process steps: (1) type fabrication and (2) text fabrication (via the printing press). Type was the main process intermediate.

The result was a Cambrian explosion of uses for text because far more people could enter the text manufacturing market, which turned out to be much larger than most people thought.

Now, let’s talk ICs and fabs. Modern ICs are an awe-inspiring achievement. Interconnecting billions of nanoelectronic devices at high density yields an impressive combination of low voltage/high speed operation, functional diversity, and physical compactness.

At the same time, IC manufacturing is exceedingly time consuming, costly, and tightly controlled. The barrier to entry is really high. (Sound familiar?)

New fabs cost $10B+. cnbc.com/2021/03/23/int… The fact that a single tool costs $150M, no matter how technologically impressive, should be horrifying. wired.com/story/asml-ext…

The enormous costs mean few can do it, especially at the leading edge, which dramatically slows the rate of innovation. mckinsey.com/industries/adv…

This situation didn’t happen randomly. It is the natural result of the manufacturing approach we have used to fabricate ICs since the dawn of the industry — planar processing.
computerhistory.org/siliconengine/…
computerhistory.org/siliconengine/…

In planar processing, devices are monolithically integrated in or on a substrate (often a Si wafer). As a result, there is little/no modularity. Per step yields must be ~100%. This is a major, often overlooked and misunderstood, driver of manufacturing cost and cycle time.

Since factoring proved so successful for text manufacturing, let’s consider it in the context of ICs. Modularizing nanoelectronic devices (e.g., transistors) would allow devices and circuits to be fabricated in different sets of process steps.

“Nano-modular devices” could be manufactured in one location by one company and then interconnected to form “nano-modular ICs” in another location by another company (or individual?). To extend the analogy, nano-modular devices would be the type of nano-modular ICs.

Why might this be reasonable? It turns out that modularization is central to nearly all manufacturing (just not ICs!):

Maybe electronic components are somehow different? Nope.

Nano things must be different, right? Nope. The entire chemicals supply chain relies on the inherent modularity of molecules, which are smaller than today’s smallest nanoelectronic devices.

What might be the benefits of a nano-modular process paradigm? Some highlights include: (1) New device/circuit innovation trajectories, supply chains, and ecosystems; (2) Natural mixing of different materials and devices in 2D and 3D (the fancy term is heterogeneous integration);

(3) Streamlined incorporation of R&D into manufacturing, similar to that possible with software upgrades; (4) A dramatically expanded and distributed manufacturing base, ensuring resilience and accelerating innovation; (5) The emergence of entirely new contexts of use; …..

Could it be that easy? Yes and no. In some ways, it really is as simple as factoring device and circuit fabrication. In other ways, it will be hard. There are lots of technical details to work out. What are the key features of a nano-modular device?

How can nano-modular devices be interconnected to enable useful circuitry? Etc. We will need systems-level work of the type being pushed by the excellent @Ben_Reinhardt, @AdamMarblestone, and others.

I know factoring sounds too good to be true. I know it’s hard not to be blinded by the 114B transistors in an M1 Ultra. wired.com/story/apple-m1… But I also know that Gutenberg’s contemporaries thought handwritten scrolls were pretty awesome …