[ChuckMcM]

Too late.

China dominates in the 'jelly bean[1]' chip business. This was the bread an butter for the dozen semiconductor shops in the Bay Area during the 70's through the early naughts. As fabs shut down they became 'fabless' and had their work shipped out to TSMC. Now they exist as a genre in China in much greater numbers than they do here. Companies that will make a couple of wafers of USB + ADC + random logic kinds of things.

And the innovation never comes from the chip companies, it comes from companies that are building the products. They go to the chip companies and say "Can you make a chip that does this?" and then a new jelly bean chip is born and if that product gets traction everyone decides to make one with their own bit of spin.

What is unique to the Chinese scene is that there are thousands of minute variations rather than any sort of ordered discipline. Transistors are free at the jelly bean chip level so there is no incentive to re-use or save on chips.

ARM has become the defacto microprocessor architecture because, in part, ARM will license it to anyone.

In the US all of these folks have coalesced into a few mega companies (TI, NXP, Microchip, Maxim) which don't have much in the way of competition locally. So innovation stops because conservative business practices rule.

This really became clear to me when I started perusing datasheet sites for the Chinese market and realizing there are hundreds if not thousands of chips for which there is no English data sheet. This is a complete reverse from the last century when nearly every data sheet was in English and getting a Chinese version was a challenge.

[1] Jelly Bean chips are those that are made in batches of 1 - 10 million with a set of functions that are fairly specific to their application.

[nerpderp83]

> And the innovation never comes from the chip companies, it comes from companies that are building the products.

I agree with you.

Eng has really got to stop thinking they are the source of ideas, most good ideas come from customers that have needs that can't be currently met. Eng balances design constraints, as companies grow, hierarchies get stronger and creativity drops. Maximum innovation comes from small companies solving customer (the ones building the products) problems.

The Jelly Bean chip business could get disrupted by the proper mix of analog and digital. The PSoC [1] was almost but not quite there

[1] https://en.wikipedia.org/wiki/PSoC

[ChuckMcM]

I think you are correct as well, its a matter of timing. Eventually there will be the equivalent of an FPGA that instead of having "LUT"s or "CLB"s they will have a bunch of useful bits of tech and an internal wiring bus that will be programmed by setting switches or blowing fuses. Then the variation will be how many pins on the package which will determine how many of these things you can bring out to pins.

I did a thought experiement of what it would take to make a 'universal' TTL chip, which was basically a 14 pin, 16 pin, and 18 pin device where all 74xx chips were actually inside of them and through a programming step you bonded one of them to pins on the lead frame. It required a 90nm or better feature size. The trick was you only have (number of pins) number of input stages and output stages (all the "TTL" part of TTL (these are fairly large by their nature). The logic was all internal 1.8v with a simple step down regulator to go from 5V down to 1.8V.

You could cover both the SSI and MSI catalogs. Of course there isn't anyone designing new stuff with TTL any more so it remains a thought experiment only.

[nerpderp83]

This would have been a godsend for managing inventory.

In a somewhat related idea, is that of unioning the footprint of alternate parts in a PCB design so that either part could be used. I have never used any high in EDA software, but if you had end to end design, schematic, pcb layout and parts databases with high level machine-readable symbolic pin descriptions, these could be inferred automatically (unioned footprints).

Using your idea, I wonder if you could automatically have "hot standby" components in the same package?

[skybrian]

Based on a link posted not long ago, FPGA's seem to already be changing this way:

http://semiengineering.com/fpgas-becoming-more-soc-like/

[fspeech]

As chip manufacturing goes DRAM used to be at the bottom of the ladder. Many logic fabs did DRAM first to learn the rope because the fab technology is simpler for DRAM (less layers than logic). DRAM fabs bought used equipment from logic fabs, so there were always swing capacity available. DRAM business was tough and the main challenge was yield (therefore cost and margin). I don't know if any of this has changed more recently as it certainly seems people are not bringing DRAM capacity online as expected -- maybe they are scared by the prospect that China may bring in a large amount of capacity.

The Chinese company mentioned in the article is only incidental to the story so the article title is kind of a click bait. Sure they decided to buy technology from the Taiwanese fab UMC, which is perfectly legal. UMC as a very large fab itself, could also be expected to provide the technology. For whatever reason, UMC seemed unable to deliver and had allegedly resorted to underhanded tactics instead.