Personally I do prefer to buy custom IC's in standard package as it leave the option open for later multi sourcing if a company sublicenses the chip. Think back to when ua741 was sole source, now half the single opamps on the planet have footprint compatible devices as well as numerous spec compatibles.
It will always be a challenge to have complex devices on a industry standard pinout, but I see room for legacy "copies" to take this route.
Either way, at the end of the day PCB changes aren't that expensive below 100MHz
First of all, your statement of at least two assembly houses isn't always true. Second of all, assembly houses are not substrate suppliers. Flip-chip BGA substrates are unique per device. The volume of that substrate is not shared by any other semiconductor company or device. Flip-chip BGA substrates is where it all goes away from anything "standard".
The article was trying to bring focus to the key products at the heart of systems. You can say those were always sole-source from a silicon perspective, but not a packaging perspective. Many of the key processors and FPGA products started out in standard packaging more than a decade ago.
When you say "We design in almost exclusively multisourced parts, the exception is the CPU, any programmable logic, things like video encoders/decoders, some specialist opamps, and RF IC's"
You are essentially agreeing with the premise of the article that the heart of your system is not standard packaging and is not multi-sourced. The biggest cost of your BOM is now in 100% custom packaging.
Interesting that the article starts off with standard packaging is over and then lists a few very special parts that were never multisourced anyway. You can't get an NVidea GPU from anyone other than NVidea and you can't get a Xilinx FPGA from anyone but Xilinx. I feel the whole premise of this article is a load of balony.
If I look wider into the field there are a number of specialised parts and even those are available in a variety of standard packages.
The larger Vendors (ie. TI etc.) offer parts in bare die with bump but the bulk of the market is still standard.
We design in almost exclusively multisourced parts, the exception is the CPU, any programmable logic, things like video encoders/decoders, some specialist opamps, and RF IC's, all of which only come from one manufacturer anyway.
As an engineer designing circuitry, one of my tasks is to consider obsolescence and multi-sourcing. I totally agree that there are a huge proliferation packages, especially in BGA, QFN, and DFN footprint styles. This also applies to power transistor footprints. To say that everything used to be standard is pushing the envelope as well. Consider the now fairly standard off-the-shelf inductors, where a few decades ago, almost every inductor was a custom or semi-custom part. A big reason for the proliferation of the packages has been the inability of prior packaging to meet the needs of people who design more (Moore?) stuff onto smaller boards. There has simultaneously been a vast improvement in circuit board fine-pitch manufacture and pick-and-place equipment capability. As such there is not the huge penalty that used be in place for non-standard packages. I have begun to really like the interchangeable packages for MOSFETs that drop onto a 5x6mm SO8-style pattern (with lots of copper across pads 5-8). These packages look lots different depending on the manufacturer, but I can even specify substitutes that are not look-alike because they are form-fit-function replacements.
If someone comes out with a rediculous package that is unique, the market will take care of them. We designers, if doing our jobs, will not use a part in a package that is available from only one source, probably a small or unstable source in many cases, in a design that is expected to have a 5-30 year longevity. Any engineer that does so needs some serious correction.
As for the availability of "standard" parts, I think it is better than it has been in a long time. Chip resistors, capacitors, and inductors abound from multiple sources. One need only look at any of the national distributors to see that small active devices (transistors, OpAmps, diodes, linear regulators/references, etc.) are almost always available in very standard packaging if you consider the SOT (3-4 terminal devices), SOD (2 terminal devices), and SO (multipin devices) series packages. It is not difficult to restrict one's self to non-BGA, non-DFN, and non-QFN packages for fairly standard circuitry as my company recently did for CPLD-controlled motor drive circuit for automotive application.
I guess we're just in between two ages ... Sooner or later the entire (high speed) system will be made within a single package - be it by (finally) finding a way to have RAM, ROM, CPU, imaging sensor, whatnot all done by the same technology, be it by 3D stacking. If that's done and no real demand remains for high-pin-count high-speed extensibility remains standard packaged systems might evolve which could be pin compatible like classic TTL chips in DIL packages.
On the other hand I see a trend to total (vertical) concentration of the market where, in the end, the most powerfull chips are not availlable for anybody except the absolute top players. 10 years ago I attempted to get an offer for an imaging sensor (each chip had a 4 figure $ price) in 20000 chip quantities - which is really a lot for the industrial imaging market - but obviously that was way below the minimum quantity ... guess I should start reading about genetics, maybe there's some fun left in that field ;-) ...
Traditionally Chip Packaging was for protecting the fragile semiconductor from the environment and expanding on-chip I/Os at wafer type geometries to PCB type geometries with a standard foot / toe print. But packages are circuits too and proper electrical design has driven the development of new "Packaging" technologies for the last 2 decades starting with the now pervasive solder bumped flip chip technology which does not even have a package in the conventional sense. In the interest of maximizing electrical performance there are a lot of packages with custom designs just like on chips themselves.
Perhaps its time to bury the "Package" term altogether at least for the performance driven segment. Recently some have tried to introduce the term "Interconnectology" but it sounds too uncomfortably close to "Proctology". The best that I can come up with is Off Chip Interconnects & Integration ( OCII ).
Replay available now: A handful of emerging network technologies are competing to be the preferred wide-area connection for the Internet of Things. All claim lower costs and power use than cellular but none have wide deployment yet. Listen in as proponents of leading contenders make their case to be the metro or national IoT network of the future. Rick Merritt, EE Times Silicon Valley Bureau Chief, moderators this discussion. Join in and ask his guests questions.