Electronics Weekly Electronics Design & Components Tech News
We have all heard that Arm is looking at charging OEMs for licences as well as charging semiconductor companies and now we hear it is building a chip and may adopt the fabless semiconductor industry model.
All this is presumably designed to beef up Arm’s potential earnings clout in the eyes of investors at the upcoming IPO.
But investors are cute people – they can see that charging OEMs and going fabless could kill Arm. This tinkering around with Arm’s business model could backfire on Softbank, with investors shunning the IPO.
One of the oldest axioms in the semiconductor industry is ‘don’t compete with your customers’.
Abandoning a fair-to-all licensing business model for a fabless model looks good on paper – after all Intel makes $60 billion a year selling microprocessor chips while Arm makes $3 billion a year selling microprocessor IP – but no one wants to go back to the x86/PC model where two companies controlled the supply of processors of a dominant market architecture. It was the fear of a repeat of that situation which persuaded regulators to ban Nvidia’s takeover of Arm.
The industry won’t stand for a repeat of the x86 monopoly. It will dump Arm just as soon as it can if Arm tries it on.
Not sure why anyone is making a big thing of this. ARM have built several test chips in the past, and I’m sure this is just one more. Of course doing it near their IPO doesn’t seem the best timing, but sometimes you need a real test chip to prove your simulations for new processes are on the right track. They did the same for Hi-K, FinFET and other industry changes. Remember it’s isn’t just processor IP that ARM sell, even though that gets all the publicity.
Hopefully stories about them wanting to change their licencing model are just silly stories.
As of course are comments about people changing to RISC-V. That architecture is at least a decade behind ARM, and can’t beat ARM or Intel without violating numerous patents.
What ARM need to do is get their heads down and fill in the weakspots in their product lineup. They know where they are.
If Arm wanted to dispel the rumours Mike they only have to say so.
Like most companies, the sensible policy is never to comment on rumours as otherwise you get people trying to sue you in the US courts for all sorts of nonsense claims.
Trouble is the last rumour (charging OEMs) turned out to be true
Apple outclassed ARM in mobile. Risc-V is open source. So far x86 has the performance crown. AMD has a license to modify ARM and make it competitive with Apple. What does ARM have to offer?
Compatibility with every smartphone in the world, FE?
Ah Yes, Multi-Chip Modules as they used to be called 30 years ago, my dear Armchair, it’s what you do when scaling isn’t delivering.
Chiplets my dear David, chiplets…
The reality is the Arm model is already dead. There are only 10 companies in the world with enough cash to drive a 2nm chip and market big enough to pay for it. Hence the need to move to chiplets – create a compute chiplet with enough CPU and GPU, and then let the rest of the industry build Io, memory and other differentiated gubbins around that. Arm needs to move, and while they will compete with Samsung, et al, they need to kickstart the market before they are throttled by the Angstrom age….
MCMs would pop up every few years and dissappeaagaim as fabs found ways to suck the chiplet functions on to a single die. They are mainline products only in very high value markets like military and space.
Fractions of a cubic mm count in a phone design and you lrapidly lose real estate advantage using MCMs. There are additional processing steps for an MCM which significantly erode value for commodity products.
MCMs are wonderful in the right application but I doubt they are much of a solution for ARM.