I'd really love to see more folks trying to make neat chips again. There tends to be a player or two in each section of computing; a couple folks making NVMe controllers, two folks making USB4 chips, a half dozen folks making wifi chips. There's a lot of ARM cores out there but there's really only a handful of cell and tablet chip makers. Then the hyperscalers cranking out server designs, which while compute heavy don't need quite the sea of integrated peripheral connections, somewhat.
The foundry model is so interesting, but there's such an IP qualification barrier to being able to ship your off the shelf say PCIe IP on a new foundry.
Seems alas like Intel maybe perhaps is giving up, or just targeting the hyperscalers with money to throw at chip-making. Samsung seems to have a similar problem here, finding customers competent enough to use well the blank slate of capabilities offered.
Ideally the open chip design world would be buoying these efforts, would be enticing more people into making chips. It feels like if these companies want to have viable foundries, maybe they should be working to accelerate these efforts more, should be trying to get open source ram controllers and open source PCIe and USB designs thriving and usable, helping folks see and understand how to qualify such designs on their foundry.
The IP control of the major design blocks is tight right now. That keeps the foundry work from flooding in, imo. And it seems so fixable!
I'd really love to see more folks trying to make neat chips again. There tends to be a player or two in each section of computing; a couple folks making NVMe controllers, two folks making USB4 chips, a half dozen folks making wifi chips. There's a lot of ARM cores out there but there's really only a handful of cell and tablet chip makers. Then the hyperscalers cranking out server designs, which while compute heavy don't need quite the sea of integrated peripheral connections, somewhat.
The foundry model is so interesting, but there's such an IP qualification barrier to being able to ship your off the shelf say PCIe IP on a new foundry.
Seems alas like Intel maybe perhaps is giving up, or just targeting the hyperscalers with money to throw at chip-making. Samsung seems to have a similar problem here, finding customers competent enough to use well the blank slate of capabilities offered.
Ideally the open chip design world would be buoying these efforts, would be enticing more people into making chips. It feels like if these companies want to have viable foundries, maybe they should be working to accelerate these efforts more, should be trying to get open source ram controllers and open source PCIe and USB designs thriving and usable, helping folks see and understand how to qualify such designs on their foundry.
The IP control of the major design blocks is tight right now. That keeps the foundry work from flooding in, imo. And it seems so fixable!