What goes around comes around...
Reminds me of what mainframe manufacturers apparently used to do in the 1960/70s.
If you wanted a speed improvement, for a wad of cash an engineer would 'upgrade' your system - by removing line(s) of wait code!
Want to enable custom features on a chip? That will cost you … and keep sending the message to chipmakers that this can be the new standard for future software-driven tweaks to pricing models. The idea of businesses paying money to enable certain features on a chip via software is one step closer to reality this week. Base …
Software defined silicon might help when customers want only a low number of chips for their spec and the manufacturer can use the same chip with feature addons switched off/on to develop and produce the chip in higher numbers.
Using that to try and get the customer to pay for addons already in the chip is risky. What are the odds the built in feature is already outdated by newer technology at the time the customer is interested in using it?
It's not actually software-defined silicon at all -- that's just another industry buzzword. It's actually compartmentalised functionality with software controlled multiplexers that either connect to or bypass some compartments. All the hardware is there all the time. You're just denied access to some of it unless you pay up (and probably on a subscription basis so they've got your goolies in their grasp once you sign up).
So, instead of making four, eight and ten core processors with differing combinations of graphic and ML cores, you just get one reasonably maxed out option and pay for what you need? Preferably via a subscription, because everyone loves a good subscription? Well, everyone who sells subscriptions loves a good subscription.