|
|
|
|
|
|
by imtringued
22 days ago
|
|
|
Industry wants to have the best memory, SATA, PCIe controllers, floating point units, SRAM units, etc. Those components need to be optimized to the particular process and the companies that are best at developing them are also the same companies making the tools used to design semiconductors in the first place. Cadence has to build a feature to design analog electronics in their software package. Then they have to make sure that their designs are compatible/manufacturable at all the foundries. This means they will inevitably develop all the basic components anyway and there is not much point to avoid paying them for the end result. An integer adder or a multiplier is a commodity, the same adder optimized for a specific fab is a specialized good worth tens of thousands, if it means you can avoid the hassle of a tape out just to test your novel design. Nobody is realistically competing on adder designs and for the complex analog stuff, the development effort may add up to several tapeouts. |
|
|
I feel like there's a ton of designs where absolute fastest is not that crucial. Plenty of products would be fine shipping DDR4-3200, or lower.
And it feels like there's significant opportunity for innovation lost when we have to accept blackbox designs, when these are pre-made systems. We could be doing incredibly neat NVMe Computational Storage things... but theres very few people able to experiment with NVMe in any meaningful way, that aren't gatekept by these expensive blocks. Near memory compute also has similar "what if" scenarios. We can treat these as "just buy off the shelf" commodity parts, until we can't: we can hardly guess how much innovation is being left behind by this current treatment.
Sure, your arguments that Cadence pre-qualifies designs at foundries is very compelling! It's incredibly nice to not have to do the hard work, to leave it all to someone else. I still think it's a bad bargain for society & that if we could collectivize the effort to make chip-making something available to people, if we could get back to a more pure Silicon Foundry style model that wasn't so gatekept, we could see something other than the incredibly sad fate that the past two decades have held: endless buy outs, endless consolidation of chip making, with far fewer new companies getting started.