My understanding is, they design the specification of the ARM assembly language, and the design of the hardware (not individual chip fab but like... abstractly, somehow?)
But if they specify the design of the hardware, why would one chipmaker have faster/better CPUs if they are using the same spec as another chipmaker with slower/worse CPUs?
What's the business model? Do they get paid per ARM chip sold, or yearly for the rights to sell ARM chips, etc?
> Do they get paid per ARM chip sold, or yearly for the rights to sell ARM chips, etc?
Both. Basically at the top end You have licensing that big companies like Apple pay a large one-off sum ( comparatively large from ARM's perspective , but tiny in Apple's view ) for the use of ARMv8. ( i.e You will need to pay again for ARMv9 or ARMv7 ), and do custom design with the Instruction Set as along as it remain compatible with ARMv8. At the bottom end you are simply buying a blueprint from ARM, add some other IP you want and pay per ARM chip.
Anandtech has an article [1] about it and most of it are still relevant.
Since we are taking about business It is also worth mentioning the revenue of ARM before Softbank acquisition was less than the Net profits of Intel.
> why would one chipmaker have faster/better CPUs if they are using the same spec as another chipmaker with slower/worse CPUs?
That's not the interesting thing, really. ARM licensees don't compete on how fast the ARM itself is, they compete on what else is on the same chip that they've designed themself. Radios, high-speed serial interfaces, analog-digital, etc.
(the big exception is Apple, who have their own silicon team improving the ARM implementation)
ARM have a very wide range of price/performance/area/power options. You don't always want the biggest and best.
I don't know exactly how compensation works, but they have basically two products:
- The ARM ISA (the "specification")
- A series of chip designs implementing such ISA (the Cortex series)
Customers can either license the Cortex design to implement in their own silicon or license the ISA and design their own CPU (like Apple did). There isn't one single Cortex core, but many of them at different performance/price points
AMD and Intel license IA-32/AMD64 and extensions from eachother, and they have agreements which make this mutual licensing apply to basically any ISA extension. The original x86 patents are long since expired, but every year they add extensions and cross-license them to eachother.
Base x86 is partially open partially owned by Intel. X86-64 has components owned by AMD. AMD and Intel cross-license. You can’t start a new x86 chip company without paying amd/intel (if you want to use modern isa)
AFAICT, ISA's are heavily patent protected. The patents are usually on implementation details, but it's difficult to implement without stepping on the patents, so...
Intel & AMD have a cross license, x86-64 was actually created by AMD. And before then, (ie, in the 70s & 80s) Intel was forced to license to third parties to qualify for defense & government contracts.
I assume the "forced to license to third parties" part applies in the case of the Zilog Z80?
But re: patents, in the case of Cyrix:
> Focused on removing potential competitors, Intel spent many years in legal battles with Cyrix, consuming Cyrix financial resources, claiming that the Cyrix 486 violated Intel's patents, when in reality the design was proven independent.
[And this is despite Cyrix having white-box reverse-engineered Intel's chips to figure out how to be software- and socket-compatible with them! That seemingly didn't matter, as long as in the end the design they actually put in their own chip wasn't encumbered by Intel's patents.]
> Intel lost the Cyrix case, which included multiple lawsuits in both federal and state courts in Texas. Some of the matters were settled out-of-court and some of the matters were settled by the court. In the end after all appeals, the courts ruled that Cyrix had the right to produce their own x86 designs in any foundry that held an Intel license. Cyrix was found to never have infringed any patent held by Intel.
With that case-law in place, it sounds like anyone demanding that third-parties license their ISA these days is effectively bluffing.
Relatively few companies have an ISA license; I doubt it's a significant income source for ARM versus licensing their core designs: only Apple and Qualcomm are left producing parts based on their own implementations in any real volume.
From what I understand, ARM will give you various different implementations of different ARM cores, depending on the licensing terms. My guess is that this includes:
- RTL optimized for simulation
- RTL optimized for synthesis, maybe with variants for FPGA vs ASIC
- Various lower-level implementations, complete with layout. Just a “block” that you drop onto your chip design. I imagine that these are something like “here’s a block that you can give to TSMC for their 16nm process”
They sell basically three things: rights (and some specifications) to design your own chips with their interface, rights to use their core designs in your SoCs, and services related to those two things.
> What's the business model? Do they get paid per ARM chip sold, or yearly for the rights to sell ARM chips, etc?
From what I understand ARM designs and licenses some chips but also licenses the right to design chips implementing the same ISA. So not all ARM processors are designed by ARM.
Typically fabless CPU/MCU companies design and implement netlists to these CPU designs. It includes all of the logic to implement the CPU. This netlist/VHDL/Verilog design is the protected IP they sell, as well as the rights to re-use, etc.
The IP itself can be placed into semiconductor fabric and combined with many other parts such as peripherals, flash memory, sram, etc.
Some companies (eg Apple) have an ARM architectural license, which gives them the ability to design custom microarchitectures, but using the same ARM ISA.
My understanding is, they design the specification of the ARM assembly language, and the design of the hardware (not individual chip fab but like... abstractly, somehow?)
But if they specify the design of the hardware, why would one chipmaker have faster/better CPUs if they are using the same spec as another chipmaker with slower/worse CPUs?
What's the business model? Do they get paid per ARM chip sold, or yearly for the rights to sell ARM chips, etc?