21 comments

[ 3.6 ms ] story [ 44.8 ms ] thread
reading the first three paragraphs after a week like this makes me crazy.
It's incredible how in those few paragraphs he envisioned so many things that could be enabled by cramming more transistors in, that a half century later are still somewhat novel.
Honestly, I feel the opposite. What he imagined back then as absolutely crazy was standard only 20 years after. It is crazy how exponentials work. Wild to think where we are standing 60 years later and wild to think where we will stand in 60 years.
I guess the bit I was referring to as somewhat novel is when he referred to memory being combined with compute rather than separate - As 'compute in memory' or distributed memory and compute architectures, are recent enough. separate memory and compute still domiante, but with HBM, stacked cache, and GPUs using significantly more on die memory were now seeing some quite different architectures realized
I haven't been on the internet much this week - what happened?
(comment deleted)
I'm genuinely tempted to answer "The singularity happened".

A bit of an overstatement[0], but it's certainly the vibe.

The news is happening so fast now, I'm not even sure without looking it up if the second room temperature superconductor was last week or not, let alone whose latest AI is passing which exams or producing images that normal people share as they mistake them for photographs.

[0] especially as I don't like the term itself: https://kitsunesoftware.wordpress.com/2022/09/20/not-a-singu...

"With unit cost falling as the number of components per circuit rises, by 1975 economics may dictate squeezing as many as 65 000 components on a single silicon chip.

The future of integrated electronics is the future of electronics itself. The advantages of integration will bring about a proliferation of electronics, pushing this science into many new areas.

Integrated circuits will lead to such wonders as home computers—or at least terminals connected to a central computer—automatic controls for automobiles, and personal portable communications equipment. The electronic wristwatch needs only a display to be feasible today. But the biggest potential lies in the production of large systems. In telephone communications, integrated circuits in digital filters will separate channels on multiplex equipment. Integrated circuits will also switch telephone circuits and perform data processing.

Computers will be more powerful, and will be organized in completely different ways. For example, memories built of integrated electronics may be distributed throughout the machine instead of being concentrated in a central unit. In addition, the improved reliability made possible by integrated circuits will allow the construction of larger processing units. Machines similar to those in existence today will be built at lower costs and with faster turnaround."

Absolutely correct in every single prediction.

Imagine being so in tune with something you're creating you can forsee the future.
(comment deleted)
Is cramming a term of art? It’s always seemed low brow to me? In any case, RIP, you were a giant.
1. Yes.

2. Yes, but it's hardware. The brow is low. (Hardware person here).

When you reach a certain stature, you're informally granted more liberty of expression in your paper titles.

Wow, there's a surprising number of predictions he nailed here. One of the most impressive is the suggestion that "newly devised automation procedures could translate from logic diagram to technological realization without any special engineering" -- hinting at the possibility of logic synthesis, which wouldn't become viable until well into the 1980s (AFAIK?), but is now the standard methodology for complex digital ICs.
I always interpreted moores law as having 2 aspects, the scaling part (cramming more per mm2), and the economic aspect (for minimum component cost).

So when people say moores law is dead, it's often confusing as clearly things are still scaling (3nm crams significantly more transistors in than 7nm), but the cost per transistor has flattened, or possibly inverted a bit.

> but the cost per transistor has flattened, or possibly inverted a bit.

No, that part is still accurate. There's still a cost premium associated with really large, dense dies -- and it's one of the major reasons why you see some manufacturers using chiplets to circumvent that.

What a visionary article and a delightful read.

To give another context to look at this in retrospective, here's [0] a die shot from a 2n3906 transistor made in 1965 (still available today). It's roughly 60µm * 60µm in size. By comparison a high end chip made in 2022 has 80 billion transistors in an area of 814mm^2 [1].

That is three hundred and fifty thousand transistors (350,000!) in the same area 58 years later.

[0] https://en.wikipedia.org/wiki/2N3906#/media/File:2N3906_top_... [1] https://en.wikipedia.org/wiki/Hopper_(microarchitecture)

Sorry, but you are comparing apples with oranges. This transistor can whitstand 40 V and can drive 200 mA. The other one is a processor working maybe at 3.3V or 1.8 V.
Can you give an example of a state of the art small transistor circa 1965 to make apples to apples comparisons with? The integrated circuits in the Apollo instrumentation unit perhaps?