The Garmin Watches don't have DRAM for power reasons. The processor is a memory constrained microcontroller without a hardware memory management unit. Without the MMU you don't get as powerful memory virtualization or process isolation.
People with direct experience of the manufacturing sector in the US and abroad (including me) will certainly tell you that America's capabilities have deteriorated, at least relative to other countries. The machine tool industry is considered a good bellwether for industrial capacity and there's been some good research on this topic if you're interested: http://www.bismarckanalysis.com/Machine_Tools_Case_Study.pdf
I've looked at the financials of hearing aid companies before. And the short answer is that the money is not going to the hearing aid company. Of the $5k you spent on your hearing aid, the hearing aid company was probably paid $500, and the other $4.5k went to the audiologist.
The problem is that most consumers' revealed preference is for better performance rather than replaceability or device longevity (I know some people here will express a different preference, but you are a minority of consumers). Consumers don't factor device longevity into their purchase decisions, so companies don't prioritize it during the design process. And it needs to be prioritized because design decisions for longevity involve tradeoffs with other device performance metrics like size, weight, battery life (you can improve battery longevity by decreasing the depth of discharge of the battery before a recharge is required), processor performance, price.
Unfortunately this article clearly demonstrates the problems companies face when they try to improve longevity at the cost of performance. Apple implemented processor throttling based on measured battery condition to improve the longevity of their devices, but even in this article, seemingly focused on device longevity, they still receive criticism for this decision:
"But even consumers who hang onto their old iPhones for as long as possible learned in 2017 that Apple released a software update that slows down old phones to counteract aging lithium-ion battery problems."
I can't speak to the CC26xx series specifically, but normally deeply embedded cores running a ROM stack like that have some additional SRAM which can be loaded at boot for firmware patches.
It's so amazing how electronics have progressed. a startup called Octavio Systemsn is now selling a single package that can run a Linux system. I wouldn't count it as running Linux on a single chip, (power system, microprocessor, RAM) because I believe there are multiple dies within the package, but still, super impressive.
This video was very cool. Are there any IC's that can perform analog computing for neural networks on the market now? I'm picturing something like an FPGA but with a bunch of op amps that you can connect into summers or amplifiers.
If not, how would one practically implement an analog computer for neural network programming (without several tables full of op-amps?)