Do you think Intel can come back, or is Intel too dysfunctional/inefficient to recover?
Just one anecdotal story, one of the wafer guys had an idea to improve a process. He went to his boss who shot it down instantly. Wafer guy knew he was right so went to another manager who liked the idea and implemented it. It ended up working and being a great idea. Instead of being rewarded for his idea, wafer guy was fired by his manager for failing to follow the chain of command.
It's built into the culture at Intel to be territorial to advance. It's been that way for at least 30 years. I personally don't see it changing any time soon.
Grove certainly sends quite utilitarian vibes. He also says that a collaborative culture is important. Your comment suggests that neither he nor his successors managed to make Intel's culture collaborative enough.
Higher management is probably often an iterated prisoners dilemma. Collaborating is best for the company but defecting is good for the ambitious manager.
It did have a few ideas which were new to me and it presents a holistic consistent management philosophy which you don't get from reading a hodge-podge of blog posts.
Two (for me) new ideas: Matrix organisation is inevitable for large companies in search of the sweet spot between agility and efficiency. Knowledge workers are middle managers.
Overall: Good book but not "you have to read it" level.
But we are now living in a world where mobile is more important than desktop, and mobile chip designs have enough power to be a threat to the desktop market, and perhaps even the server market in the not-too-distant future. This is something that should worry Intel.
Why so? Few more material, and device generations are on the way. Litho has went through more generation changes in the last 1 decade, than in any previous one.
Whole new device classes are on the way: optics on silicon, new memory, logic in the backend, and logic on package.
Semiconductor engineering is still a career suicide though, that haven't changed in the last few decades.
Wow, so I need to thank myself for accidentally bumbling towards ops and dev? Meh, for me it’s just that in EU there’s a total dearth of opportunities in the industry
> optics on silicon,
Yep! That and quantum and post VNeumann architectures. There's some weird shit on the horizon that I probably won't live to see in household products, but I don't think it will be be optical deposition by layer. It might, who knows, I"m officially talking out my arse now. :)
And not even for I/O, but for on-die communications, and busses.
High speed SERDES may well be replaced by same-die optics, as a lower transistor count, dumber, and cooler solutions.
Could you go into more detail?
The long term is that the industry will get less, and less labour intensive as more, and more old, labour intensive fabs are retired, and new ones are 100% AMHS.
This is so true. I am currently a hardware engineer and planning to switch to software. If you are not in Nvidia, Apple or other mega-growth hardware companies, you are suffering atleast 40% paycut compared to software industry and also working about twice as hard, using obsolete tools and flows, and almost no opportunities for technical advancement or significant impact. Meanwhile the annual raises are cost-of-living adjustments. Its truly depressing.
That was only around 2000. There was a lot for me to get excited about since then:
* AMD64
* SMT
* Dual-core / multicore
* On-chip integrated graphics that weren’t awful (low-end discrete graphics aren’t a thing anymore)
* Gigantic L3 and L4 caches (a side-effect of on-chip graphics)
* Turbo mode making a comeback (yes, I know Turbo Mode switches actually slowed down computers)
Judging by Apple’s direction, it’s looking like we’ll start seeing actual (ultra-fast) RAM on-chip too, with Optane that could mean we can finally move-away from block-addressed storage and have a massive flat memory-space for everything .
I kinda laughed when AMD beat Intel to the punch not once, but TWICE. (1GHz, then 64b).
I also worked on the 740, 810 and 815 (both hardware and drivers, oddly, I did a lot of stuff). Game devs HATED us for the 810/815 because it was such huge volume they had to target it as the LCD.
RTL https://en.m.wikipedia.org/wiki/Register-transfer_level
RTL is a way to model a digital circuit and some common HW design languages are VHDL, Verilog amd Intel's proprietary iHDL.
Numbers are just Intel chipset families.
Intel has money and equipment and existing market share: there's no reason for them not to be profitable for the long term, as long as they stop acting like they don't have any competition (as they did until 2018) or that their only competition is AMD (as they currently are pretending).