Thanks everyone for the responses. I think it seems, for the depth I'm interested in, it'll require deep study on my part, instead of my spare brain cycles. Here's my thinking that prompted this, sorry for the brain dump and lack of links, these are startupish ideas.
In business you have either the forces of disintermediation/vertical-integration, or specialization. We've moved from "real men have foundries" to fabless, with consolidation of foundries. I think of ARM as a software company. I am a software person. The goal of software is to turn hardware X into X-as-a-service, à la Amazon AWS. Part of the compromise of XaaS is compromise. What you loose in specialization you gain in flexibility; you shift capex into opex and promote venture capital.
Okay, so what?
What is post ARM? Three are two opposing forces of modern compute. On the one hand you have Amazon with Annapurna, and Google with its custom switches, consolidating the datacenter into a service, a really big black box with a internet-facing API. On the other hand you have the free-wheeling world of IoT, with all the big players trying desperately to create walled gardens. My thesis: this won't work for IoT.
What does this mean for future hardware? Well, think of what it would mean to turn something into a XaaS. Consider the foundry as a VHDL/Verilog to silicon. Simplify the frontend and the backend, i.e. limit/streamline the HDL and the output geometry(TSSOP/BGA/etc), in order to increase yield. The software would look like mix-and-match core (ARM/MIPS/lowRISC), pick on-chip bus, pick SRAM, pick memory controller (hey, cool you're dropping all DRAM controller for several NAND controllers and a small on-chip flash), pick accelerators, ethernet, etc/etc. Of course this is essentially already the case in hardware, except for the still large upfront capital expense.
But consider something else for IoT, something for the future electrician/carpenter/plumber. That is, change the customer from the end user to experts in the trades or proficient DIYers. To do this a startup would specify the physical/mechanical/electrical/thermal/acoustic/etc properties of modules, be it li-fi, smart sockets, servos, etc and make software to mix-and-match them into devices by these trades experts to solve problems. For example, someone who installs blinds can put together a modules to build something that automatically controls the blinds. To make this work requires some serious cross-disciplinary thinking. The backend would have this modules fully openly specified with factories wherever in the world competing to build them. The frontend would be educating and marketing to tradespeople as a way to make their practices more lucrative and increase their fees. The goal would be this software with lots of pre-designed module combinations, and tips and tricks for "blinging" up your home. To begin, the startup would have to design and manufacture its own modules to show viability, and then drop out of the picture and earn money on royalties collected from being interoperable with their design software.
Yes, if only I had a few megabucks lying around... :)
We have gone fabless but this hasn't altered the capital equation as much for hardware as one might hope. Chips tend to become 'pad limited' and 'dissipation limited' so there is a limit on what can be done in a certain volume. Some of the more creative stuff isn't about chips, its about interconnects. Which gets us to #2.
The "Internet of Things" isn't really an Internet, the disruption is that everything is a network and yes, Sun had it correct when they said the network is the computer. What you have is a collection of agents which cooperate to achieve a commanded objective. No one cares about 'smart dust' what they care about is the transformative aspects of real time contextual data. The IoT is about creating adapters which convert ambient information into data that can be collectively consumed and processed by computers. A billion barometers on smartphones taking samples of the pressure where they are, combining that with a set of GPS coordinates and transforming ambient data (air pressure at a known point) into a consumable dataset. Which when observed over time can inform on larger processes such as weather fronts. None of that needs "new chips" but it can benefit from easier assembly of existing capabilities.
In business you have either the forces of disintermediation/vertical-integration, or specialization. We've moved from "real men have foundries" to fabless, with consolidation of foundries. I think of ARM as a software company. I am a software person. The goal of software is to turn hardware X into X-as-a-service, à la Amazon AWS. Part of the compromise of XaaS is compromise. What you loose in specialization you gain in flexibility; you shift capex into opex and promote venture capital.
Okay, so what?
What is post ARM? Three are two opposing forces of modern compute. On the one hand you have Amazon with Annapurna, and Google with its custom switches, consolidating the datacenter into a service, a really big black box with a internet-facing API. On the other hand you have the free-wheeling world of IoT, with all the big players trying desperately to create walled gardens. My thesis: this won't work for IoT.
What does this mean for future hardware? Well, think of what it would mean to turn something into a XaaS. Consider the foundry as a VHDL/Verilog to silicon. Simplify the frontend and the backend, i.e. limit/streamline the HDL and the output geometry(TSSOP/BGA/etc), in order to increase yield. The software would look like mix-and-match core (ARM/MIPS/lowRISC), pick on-chip bus, pick SRAM, pick memory controller (hey, cool you're dropping all DRAM controller for several NAND controllers and a small on-chip flash), pick accelerators, ethernet, etc/etc. Of course this is essentially already the case in hardware, except for the still large upfront capital expense.
But consider something else for IoT, something for the future electrician/carpenter/plumber. That is, change the customer from the end user to experts in the trades or proficient DIYers. To do this a startup would specify the physical/mechanical/electrical/thermal/acoustic/etc properties of modules, be it li-fi, smart sockets, servos, etc and make software to mix-and-match them into devices by these trades experts to solve problems. For example, someone who installs blinds can put together a modules to build something that automatically controls the blinds. To make this work requires some serious cross-disciplinary thinking. The backend would have this modules fully openly specified with factories wherever in the world competing to build them. The frontend would be educating and marketing to tradespeople as a way to make their practices more lucrative and increase their fees. The goal would be this software with lots of pre-designed module combinations, and tips and tricks for "blinging" up your home. To begin, the startup would have to design and manufacture its own modules to show viability, and then drop out of the picture and earn money on royalties collected from being interoperable with their design software.
Yes, if only I had a few megabucks lying around... :)