One of the most interesting things about hardware vs software is the barrier to entry.
Hardware intellectual property has an insane number of working parts and "factors" (electrical resistance, photo-lithography, metal layers, silicon is amazing) all have to be solved.
Software by contrast, follows the open source model. All those things you mentioned, are being done by people and academics in their space time on the internet. You're essentially assembling pre-configured tools
Hardware is more difficult, because the laws of physics are FAR harder to account for, than tying together APIs. Because APIs are after all, designed for humans to be able to understand and configure.
Software is also complex, but the good people in FOSS have already assembled their shit for free and are giving it to you.
Additionally, basically no hardware IP and blueprints are open sourced. Because of the free flow of goods (and not services), someone in an IP disrespectful country will clone you and sell your goods internationally. You have basically no resource to pursue that individual.
Launching an IP disrespecting internet business at scale however? That is almost impossible. As soon as your IP disrespecting competitor gets going, you can hamstring them.
Hence for hardware, create a develop, release, iterate cycle, and keeping the money flowing, you essentially need to employ a LOT of people to get your startup going.
Hence VS fund about 10 SaaS startups per hardware one.
Mostly I was talking about SaaS with the "make your first two engineering hires like this" guide. Clearly for photo-lithography with silicon on leading edge nodes, you just cannot do that.
> Hardware intellectual property has an insane number of working parts and "factors" (electrical resistance, photo-lithography, metal layers, silicon is amazing) all have to be solved.
Again, fabless companies don't have to do any of this. You get a PDK from the factory which provides all these parameters and even standard cell libraries so you don't even have to do circuit design. You just feed your logic into "compilers". You need "physical" designers to optimize the output of those stages, but these are not people looking at the physics, they just optimize P&R and floor plan according to the PDK specs and tool output basically to close simulated timing.
In practice once you get to a large enough and high performance product, you would need custom circuit designers. And step up even larger again and you'll likely be working directly with the manufacturer on defining and tweaking the process. At Apple scale, you probably even define your own custom process node. Once you get to this scale though it's pretty much like big software houses writing their own libraries or OSes or compilers or asm code because the standard available stuff is not quite enough. That all takes a lot of engineers too.
But that's not _required_ and almost certainly PA Semi would not have had a whole lot of engineers on that work, it would have been logic, analog, PD, V&V, power, primarily. And almost all would just work on the standard package they get from their fab. So mostly writing "software". And you don't need any money for a develop,release,iterate cycle on most of the hardware stack because it can be and is simulated.
Hardware intellectual property has an insane number of working parts and "factors" (electrical resistance, photo-lithography, metal layers, silicon is amazing) all have to be solved.
Software by contrast, follows the open source model. All those things you mentioned, are being done by people and academics in their space time on the internet. You're essentially assembling pre-configured tools
Hardware is more difficult, because the laws of physics are FAR harder to account for, than tying together APIs. Because APIs are after all, designed for humans to be able to understand and configure.
Software is also complex, but the good people in FOSS have already assembled their shit for free and are giving it to you.
Additionally, basically no hardware IP and blueprints are open sourced. Because of the free flow of goods (and not services), someone in an IP disrespectful country will clone you and sell your goods internationally. You have basically no resource to pursue that individual.
Launching an IP disrespecting internet business at scale however? That is almost impossible. As soon as your IP disrespecting competitor gets going, you can hamstring them.
Hence for hardware, create a develop, release, iterate cycle, and keeping the money flowing, you essentially need to employ a LOT of people to get your startup going.
Hence VS fund about 10 SaaS startups per hardware one.
Mostly I was talking about SaaS with the "make your first two engineering hires like this" guide. Clearly for photo-lithography with silicon on leading edge nodes, you just cannot do that.