Archive 27/02/2024.

Gameplay - basic constraints


A while ago I wrote an article about The silicon industry invisible revolution. The main thesis was that general-purpose processors no longer meet their user needs. I mean, they still meet some of them, and can meet all of them, but at the price of increased energy consumption (you know, they are all Turing machines and can emulate each other, but that adds overhead).

This revolution is hidden from consumers. Sometimes, you can pick up in the news that this-and-this manufacturer builds its own chips. Or bought a chip factory, but people in general do not pay attention to the value chain being in the invisible part for them.

Let me copy a map that was a part of that analysis:

Special use cases put pressure on manufacturers to build specialised chips. Factories and designers do not scale easily, so this situation is being played in two different ways:


  1. Big players acquiring factories. Their purpose is to develop know-how, patents, build solutions that fit specialised user needs and prevent others from doing so. Since they already own those factories, they will do everything to hold those constraints for as along as possible, to prevent competition.
  2. Small players, notably in the IoT world, where there is still need to build and launch very niche designs, and where building a chip is not a competitive advantage, do everything to keep costs down, and avoid creation of constraints around factories. That includes sharing designs, building blocks, and also sharing machines that can physically produce chips.

The game of constraints is simple - identify what all players want, and what you can control. Than start controlling it, and you will make everyone else life harder.


Good stuff here. Been thinking since last September about how Apple has been ramping its “custom silicon” efforts from still relative “sleeper” (OK, the iPhone was always pretty fast, & had somewhat customized ARM since at least the A4/iPhone4 days) to full-blown marketing bonanza:

Interesting is how the “neural cores” have seen the most improvement, and are underwriting Apple’s edge computing approach to Machine Learning, which is in turn underwriting the now front-and-center “privacy first” moat strategy (Google, FB, and others can’t follow there).

Recently had a chat about the question where on a WMap to place “Privacy” as a (somewhat counter-factual/“negative” of sorts?) feature. Some initial statements out in the wild point to a strong version of privacy being next to impossible for the data-exploited masses = commoditized, such that only the wealthy could afford it = “custom” territory. So is Apple playing an aspirations angle here?


That’s indeed an interesting aspect, and I will look at it, probably in the near future. My not-so-initial thoughts would be that ‘privacy’ is a social construct and it is not a single component, in the same way as ‘security’ means many things. ‘Privacy’ is either about hiding things that may not get you sympathy but have no control about (employers and f.e. having kids). ‘Privacy’ is about not being judged, about separation of action and maybe a couple of different things…

It is also a powerful marketing weapon. When Apple says ‘privacy’, f.e. in Siri, it means that most probably computing will happen on a device that does belong to you. But… is it really different from a cloud based voice dictation?

Tricky. If possible, I’d like to move this aspect to a new thread.


Yes, it gets complicated quickly… sure, move it wherever, I was merely noting the connection to the custom silicon